C C60 Overview ▶ Fullerenes for Drug Delivery Capacitive and direct current (dc)-contact MEMS switches are among the most important micromachined devices for high-frequency applications due to their near-ideal RF performance. Dc-contact switches function similarly to conventional relays: micromachined beams or plates move under the influence of an appropriately applied force (e.g., electrostatic force) to open or close a metal-to-metal contact. While micromachined beams or plates are also utilized in capacitive switches, these switches rely on metal-to-dielectric contacts to implement their on and off states. Capacitive switches are particularly attractive for demanding high-frequency communications, electronic warfare, and radar systems due to their ultralow loss (< 0.1–0.2 dB up to 40 GHz), high isolation (>20–50 dB for frequencies beyond 10 GHz), very high linearity (>66 dBm third-order intercept point), and near-zero power consumption (
tens of nJ per switching cycle and zero quiescent power for electrostatically actuated switches). When compared to solid-state switches, capacitive switches are relatively slow devices with speeds ranging in the tens to hundreds of microseconds range. This speed is primarily limited by switch inertia and squeeze film damping. Their relatively large lateral dimensions of tens or hundreds of mm allow capacitive switches to handle several hundred mW of RF power. Long-term operation, however, can only be achieved if they are hermetically sealed in order to avoid contaminationand humidity-induced failure. Hermetically sealed capacitive switches have successfully switched over Cancer Modeling ▶ Models for Tumor Growth Capacitive MEMS Switches Dimitrios Peroulis School of Electrical and Computer Engineering, Birck Nanotechnology Center, Purdue University, West Lafayette, IN, USA Synonyms Electrostatic RF MEMS switches; Micromechanical switches; RF MEMS switches Definition Capacitive micro-electro-mechanical systems (MEMS) switches are a special type of micromachined switches that control radio frequency (RF) signal paths in microwave and millimeter-wave circuits through mechanical motion and contact. B. Bhushan (ed.), Encyclopedia of Nanotechnology, DOI 10.1007/978-90-481-9751-4, # Springer Science+Business Media B.V. 2012 C 364 Capacitive MEMS Switches, Fig. 1 (a) Side-view and (b) top-view schematics of a typical shunt capacitive MEMS switch. Both the up and down states are shown Capacitive MEMS Switches a b 100 billion cycles at room temperature and under low RF power conditions (20 dBm). Despite the aforementioned RF advantages, capacitive switches are currently not available commercially and are not widely utilized in defense or communication systems. This is primarily due to the facts that (1) high-yield manufacturing processes are not widely available yet and (2) their main failure modes such as dielectric charging, dc/RF gas discharge and metal creep and the physics behind them have not been adequately understood and addressed today. Switch Structure and Actuation Mechanisms Figure 1 shows a typical capacitive MEMS switch [1]. This is a shunt switch configuration, and is the dominant capacitive switch configuration in the literature today. The signal travels down the center conductor, and if the switch closes, will return along the outside conductors. It is, however, possible to design geometries for series configurations. Their characteristics, nevertheless, are very similar to the ones found in shunt switches. Consequently, this entry focuses primarily on shunt capacitive switches. The switch in Fig. 1 consists of a movable beam or plate that is anchored to the switch substrate. While typically the term “plate” may better characterize the geometry of Fig. 1 for typical lateral switch dimensions, it is common in the RF MEMS literature to refer to this geometry as a fixed-fixed beam. Hence the term “beam” is adopted here to describe such capacitive switch geometries. Cantilever beams are also possible particularly for series switch configurations. The fixedfixed beam anchors are typically metallic and are connected to the RF line. For example, they can be connected to the ground planes of a coplanar waveguide line as shown in Fig. 1b. The movable beam is typically composed of a thin-film (thickness h of 0.5–2 mm) metal such as gold, aluminum, nickel, or molybdenum. It is also possible that the beam is comprised of multiple layers including thin-film dielectrics such as silicon nitride or silicon dioxide and metals. One or more metallic pads are placed underneath the beam. In the simplest case, a single metallic pad is placed under the beam as shown in Fig. 1b. In this case, this pad is the center conductor of the coplanar waveguide line. This pad is usually covered by a thin (0.1–0.3 mm) dielectric layer such as silicon nitride, silicon dioxide, or more recently amorphous [2] or ultrananocrystalline [3] diamond. The beam’s length and width are determined by the required down-state switch capacitance as explained in the following section. This results in length and width in the tens to hundreds of mm for 5–40 GHz capacitive switches. The thickness of the dielectric layer also impacts the down-state capacitance. While high capacitance is in general required, thicknesses lower than 0.1 mm are hard to achieve in practice due to the need to handle high electric fields (dc and/or RF) across this Capacitive MEMS Switches 365 Capacitive MEMS Switches, Table 1 Main advantages and drawbacks of common actuation schemes for capacitive RF MEMS switches Actuation mechanism Electrostatic Advantages Zero quiescent power consumption, easy biasing circuit, fast transient response (tens of microseconds) Magnetostatic Low voltage, high contact pressure, potentially low power with latching mechanism Electrothermal Low voltage, high contact pressure, size comparable to electrostatic schemes Piezoelectric Same as electrostatic by with low voltage Drawbacks Need to generate high voltage, high voltage may lead to charging and breakdown issues Low quiescent power consumption requires latching, slower than electrostatic due to increased switch size High quiescent power consumption (mW), slow response time (tens to hundreds of milliseconds) Difficult to achieve high-quality piezoelectric layer and integrate it with RF circuit at low temperatures dielectric. The gap ðg0 Þ between the bottom surface of the movable beam and the top surface of the dielectric layer is determined by the need to minimize the upstate capacitance. A low up-state capacitance is necessary for low insertion loss. Capacitive switches in the 5–40 GHz range typically have gaps in the 2–5 mm range. The switch has two states of operation. In the up state, the RF signal goes through the signal line almost unaffected by the movable beam. The up state is also called “zero biased” or “on state”. In the down state the RF signal does not go through the RF line because it is reflected (see following section). The down state is also called the “biased” or “off state.” An actuation mechanism is required to move the switch beam between these two states. Several possible actuation schemes exist, including electrostatic, electrothermal, magnetostatic, and piezoelectric. Table 1 summarizes the main advantages and drawbacks of each actuation scheme. The vast majority of reported capacitive switches are electrostatically actuated. They use the same actuation principle as the one originally proposed when the first capacitive switch was invented and reduced to practice in 1994 [4, 5]. In this original scheme a dc actuation voltage is applied between the C movable beam and the actuation pad underneath it (Fig. 1). The beam is attracted due to the generated electrostatic field and collapses on the switch dielectric layer. Despite the need to generate a high dc voltage (30–100 V), which can be readily accomplished using a dc-dc converter, electrostatic switches exhibit the most desirable electromechanical characteristics, including the fastest possible response, zero quiescent power consumption, and the easiest possible biasing circuits. The resulting high fields though may lead to (gas and solid) dielectric charging and their associated reliability issues. More detailed discussion can be found in the last section. RF Performance Figure 2 shows a simple but physically meaningful and accurate lumped-element equivalent circuit of the switch shown in Fig. 1. It also includes typical values for all the equivalent circuit components [1]. The parameters a and b in this figure represent the attenuation constant and propagation constant of the transmission line respectively. The up- and down-state capacitance values are the most critical ones in this equivalent circuit. The up-state capacitance can often be accurately calculated by a typical quasi-static expression CUP ¼ Cpp þ Cff ¼ e0 A þ Cff g0 þ etdr where Cpp and Cff are the parallel-plate and fringingfield capacitances, respectively, A is the RF area of the switch (A ¼ Ww in Fig. 1), g0 is the initial switch height, td is the dielectric layer thickness, and er is this layer’s dielectric constant. For typical switch geometries, the fringing-field capacitance could reach 25–50% of the parallel plate capacitance. If improved accuracy is needed, a full-wave simulation is performed to estimate the switch up-state capacitance. The up-state capacitance must be sufficiently small to minimize the up-state insertion loss. Assuming a welldesigned switch where the contributions from L and Rs can be ignored in the up state, the switch up-state reflection coefficient can be calculated from S11 ¼ joCUP Z0 2 þ joCUP Z0 C C 366 Capacitive MEMS Switches Capacitive MEMS Switches, Fig. 2 Lumped-element equivalent circuit of the capacitive switch shown in Fig. 1. Typical values are provided for the lumped components of this circuit Z0,a,bl Z0,a,bl Typical Values C L Rs Switch Impedance Zs = Rs + jw L + f0 = CUP 1 , C= jw C CDOWN 1 2p X-band K-band CUP CDOWN 0.1/6 pF 0.04/3 pF L 4 – 80 pH 6 – 50 pH Rs 0.1 – 0.3 Ω 0.1 – 0.3 Ω Zs = 1/jwC for f << f0 Rs for f = f0 jwL for f >> f0 LC where Z0 is the characteristic impedance of the transmission line (typically 50 O). For example, an up-state capacitance of 70 fF results in S11 < 10 dB up to approximately 30 GHz. The up-state switch ohmic loss is the other critical up-state characteristic. The total ohmic loss of the switch can be calculated as Loss ¼ 1  jS11 j2  jS21 j2 This depends on (a) the attenuation a (dB/cm) of the transmission line underneath the movable beam and (b) on the switch series resistance Rs . Well-designed capacitive switches can exhibit a total loss of less than 0.1 dB up to 40 GHz. Figure 3 shows numerical values for this loss for typical switch characteristics. The switch down-state capacitance is more complicated to calculate because the switch beam may not be perfectly flat against the dielectric layer. Even in good designs this may not be possible due to the roughness of the layers involved. A model that is often used to capture the nonideal down-state switch capacitance is [1] CDN e0 A 1 er ¼ td þ 2 r þ er td ! where r is the roughness amplitude. The down-state fringing-field capacitance is not included in this equation because it is typically not significant (<5% of the parallel-plate capacitance) due to the small dielectric layer thickness. As shown by the equation above, the experimentally achieved down-state capacitance can Capacitive MEMS Switches, Fig. 3 Simulated ohmic loss for a typical shunt capacitive switch. The attenuation a (dB/cm) depends on the transmission line characteristics vary greatly depending on the true contact area and the dielectric layer characteristics including its roughness. In practice, it is difficult to avoid a 30–50% degradation of the down-state capacitance compared to the theoretical parallel-plate value. A high down-state capacitance (2–5 pF) is typically required in order to achieve an acceptable isolation (>20–50 dB) level at the desired frequency. One way to achieve this is to decrease the dielectric layer thickness. However, this dielectric layer needs to sustain very high electric fields (50–150 V/mm) across its thickness. Given typical fabrication process limitations that prohibit hightemperature growth processes for the dielectric layer Capacitive MEMS Switches 367 C 0 Capacitive MEMS Switches, Fig. 4 Simulated and measured down-state scattering parameters of a capacitive MEMS switch (After [6] with permission) S11 (sim) S11 (meas) −5 −10 C S [dB] S21 (sim), Lp = 2 pH −15 S21 (sim) S21 (meas) −20 −25 0 5 10 15 20 25 30 35 40 Frequency [GHz] (please see fabrication section), dielectric layers thinner than 0.1–0.2 mm become impractical. A second way to increase the down-state capacitance is to increase the dielectric constant. For instance, bariumstrontium-titanate (BST) or strontium-titanate-oxide (STO) films with dielectric constants up to 400 can be employed. Besides additional fabrication complexities, these films have not been thoroughly studied in MEMS switches and may exhibit unacceptably high dielectric charging. Most switches typically employ some form of silicon dioxide or silicon nitride dielectric with dielectric constant in the 3.9–7.5 range. For switches with very low inductances (L < 1–2 pH), the switch isolation can be approximately calculated as S21 ¼ 2 2 þ joCDN Z0 A more accurate calculation reveals that the switch inductance and series resistance also determine the total down-state isolation. In particular, its inductance cancels the down-state capacitance at the switch resonant frequency f0 ¼ 1 pffiffiffiffiffiffiffiffiffiffiffi 2p LCDN This frequency can be adjusted by controlling the switch physical geometries. Switch inductances in the range of 1–100 pH can be readily achieved [1]. However, higher switch inductance typically results in a higher switch series resistance. Typical switch resistance values range in the 0.1–2 O range. The series resistance is the primary limiting factor of the switch isolation at its resonant frequency. At that frequency the switch isolation can be calculated as S21 ¼ 2Rs 2Rs  at f ¼ f0 2Rs þ Z0 Z0 Figure 4 shows measured and simulated results of a typical capacitive switch [6] with CDN ¼ 1:1 pF; L ¼ 87 pH; Rs ¼ 1:95 O. This figure also shows the expected performance when the inductance is reduced to L ¼ 2 pH. Electromechanical Considerations: Static Behavior Figure 5 shows a simple but physically meaningful one-dimensional electromechanical model of the switch geometry of Fig. 1. The beam is modeled as a spring-mass system with a spring constant k. This spring constant depends on (a) the beam geometry, (b) the electrostatic force distribution on the beam, and (c) the residual stress of its structural film. The residual stress s (MPa) is due to the fabrication process and depends on the exact deposition conditions. Typically a tensile stress (s > 0) is needed in order to avoid buckling. The spring constant can be expressed as k ¼ k1 þ k2 C 368 Capacitive MEMS Switches R I m, L k /2 VC VS k/2 h er , td Fe g0 Capacitive MEMS Switches, Fig. 5 One-dimensional electromechanical model of a capacitive RF MEMS switch where k1 depends on the first two factors and k2 depends on the residual stress. The exact values can be calculated based on the specific switch design. For example, for the fixed-fixed beam of Fig. 1 and assuming that W ¼ L=3 and that the electrostatic attractive force is uniformly distributed along the beam section directly above the coplanar waveguide center conductor, the spring constant can be calculated as [1] k ¼ k1 þ k2  3      h 27 h 3 ¼ 32Ew þ 8sð1  nÞw L 49 L 5 where v is the beam’s Poisson’s ratio. For usual beam geometries and fabrication processes with residual stress in the order of 10–50 MPa, the second term dominates the spring constant. Typical spring constant values range from 10 to 50 N/m in order to provide sufficiently high restoring force and avoid stiction issues. While low spring-constant designs have been successfully demonstrated [7], special care needs to be taken in avoiding stiction and self-actuation due to high RF power [8]. It is also important to mention that the above spring constant calculations are based on small-deflection theory. A nonlinear spring constant may need to be derived if this condition is not satisfied. The electrostatic force Fe on the switch beam can be calculated as Fe ¼ @We 1 2 @CðgÞ 1 e0 AV 2  ¼ V 2 @g 2 g2 @g where We is the stored electrostatic energy, g is the switch gap between the beam and the actuation pad, and V is the applied electrostatic voltage. The last Capacitive MEMS Switches, Fig. 6 Simulated gap-voltage relationship for a capacitive MEMS switch with the following characteristics: L ¼ 300 mm; w ¼ 100 mm; W ¼ 120 mm; h ¼ 1 mm; g0 ¼ 3 mm; E (Young’s modulus) = 79 GPa; s ¼ 10 MPa. These results have been obtained with the PRISM center online simulation tool in MEMShub [13] approximation is based on assuming a parallel-plate capacitance approximation and by ignoring the dielectric contribution. The static switch gap can be calculated by taking into account the static equilibrium of the forces applied on the beam 1 e0 AV 2 ¼ k ð g0  gÞ 2 g2 The above equation can be solved for the applied voltage as rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2k 2 g ð g0  gÞ V¼ e0 A This equation is plotted in Fig. 6 for typical switch parameters. As Fig. 6 shows, there are two possible gaps for any given actuation voltage. This is not observed in practice and is a result of the unstable behavior of the beam. In particular, for small actuation voltages, the electrostatic force is increased proportionally to g12 . However, the restoring force is only increased proportionally to g. Hence, there is a critical gap beyond which the restoring force cannot hold the beam and the beam collapses on the dielectric surface. This critical gap can be found from the Capacitive MEMS Switches 369 C than g0 . It is hard to analytically calculate the hold-down voltage because it depends on many fabricationdependent conditions such as the adhesion force between the beam and the dielectric layer. Typical hold-down voltages are in the range of 5–15 V [1]. Consequently, the gap-voltage relationship is strongly hysteretic. It is also worth mentioning that nonideal conditions such as an initial beam curvature and nonlinear bending are not included in this model. These effects can be captured by more complicated nonlinear beam models such as the ones presented in [9, 10]. Electromechanical Considerations: Dynamic Behavior Capacitive MEMS Switches, Fig. 7 Simulated shape of the movable beam of a capacitive MEMS switch for different bias voltages. The characteristics of the switch geometry are the same as in Fig. 6. These results have been obtained with the PRISM center online simulation tool in MEMShub [13] The dynamic behavior of the capacitive MEMS switch of Fig. 1 can be approximately captured by a onedimensional model as described by the following equation mg00 ðtÞ þ bg0 ðtÞ þ kgðtÞ ¼ F previous equation by taking its derivative and setting it up to zero @V 2 yields ¼ 0 ! gc ¼ g0 @g 3 The voltage, therefore, required for actuating the beam, called the pull-in or pull-down voltage Vp , is given by sffiffiffiffiffiffiffiffiffiffiffiffi 8kg30 Vp ¼ V ð g c Þ ¼ 27e0 A Figure 6 also shows the voltage-gap relationship as obtained by a two-dimensional beam model [13]. The gap plotted is between the center of the beam and the actuation pad. This curve is slightly different because the beam is not deformed as a perfectly flat object as assumed by the one-dimensional model. Figure 7 shows the actual deformed shape of the movable beam for voltages up to the pull-down voltage. Actuation voltages in the range of 30–100 V are typical in RF MEMS switches. Notice, however, that once the beam is actuated, the voltage required to hold the beam down (holddown voltage Vh ) is much lower because the gap between the beam and the actuation pad is much lower where m is the switch mass, b is the damping coefficient, and f is the externally applied force. RF MEMS switches are typically packaged in an environment of 1 atm in order to avoid excessive ringing due to an underdamped response. As a result, the switch damping is dominated by squeeze-film damping as the gas under the beam is displaced during the switch motion. Due to the small gap g0 , it is in general difficult to accurately calculate the damping coefficient particularly in the near-contact region. This is further complicated by the possible existence of holes in the beam that aid its fabrication and substantially improve its switching speed. While there are several published approximations that can yield a reasonable approximation to the damping coefficient [1], accurate macro-models based primarily on rarefied gas dynamics only recently started becoming available [11, 12]. An equivalent way to characterize the switch damping is by its mechanical quality factor defined as Q¼ k o0 b qffiffiffi k where o0 ¼ m is the switch mechanical frequency. Typical switches show mechanical frequencies in the 20–100 kHz range and quality factors in the 0.5–2 C C 370 Capacitive MEMS Switches 4 Gap Height (μm) 3 2 1 p=1 atm p=0.5 atm 0 20 40 60 80 100 Time (μs) Capacitive MEMS Switches, Fig. 8 Simulated switching time (pull-down) of a capacitive MEMS switch for different bias voltages. The characteristics of the switch geometry are the same as in Fig. 6. These results have been obtained with the PRISM center online simulation tool in MEMShub [13] range. Figures 8 (switch closure) and 9 (switch opening) illustrate dynamic responses as calculated by two-dimensional beam models that accurately capture squeeze-film damping [13]. Notice that the displacement at the center of the beam is plotted in these graphs. The switching speed can also be estimated based on the simple one-dimensional model. While it is difficult, in general, to derive an exact analytical solution, this model can provide reasonable approximations for common cases. For example, for Q > 2, the closing time can be estimated by [1] ts  3:67 Vp Vs o0 where Vs is the applied voltage. Other limiting cases can be found in [1]. In general, closing times in the 5–50 ms range can be achieved. A similar range is typically possible for the release times. The one-dimensional model can also be used to estimate the velocity and acceleration of the switch. Switching velocity in the 1–10 m/s range can be observed in the near-contact region. Due to its small mass, the switch acceleration can exceed 106 m/s2 in the same region. Capacitive MEMS Switches, Fig. 9 Simulated switching time (release) of a capacitive MEMS switch for different pressure levels. The characteristics of the switch geometry are the same as in Fig. 6. These results have been obtained with the PRISM center online simulation tool in MEMShub [13] Fabrication Methods Capacitive switches can be fabricated with conventional micromachining processes and require a small number of masks. Figure 10 illustrates the masks of a typical fabrication process. • Step (a): The first mask defines the circuit metal of Fig. 1 after this metal layer is deposited on the substrate through evaporation, sputtering, or electroplating. Gold, aluminum, and copper are common metal choices. A smooth metal surface is particularly important directly underneath the switch beam in order to minimize local electric field enhancement. • Step (b): The second mask defines the dielectric layer to cover the portion of the metal that will be under the beam. Unless a high melting temperature metal has been deposited (e.g., tungsten), a relatively low-temperature process is required for the deposition and patterning of the dielectric layer in order to avoid damaging the circuit metal. Plasma-enhanced chemical vapor deposition (PECVD) is the most common process for depositing silicon nitride/oxide films. This is usually followed by a reactive ion etching (RIE) step that helps etching the unwanted dielectric layer parts. Capacitive MEMS Switches Capacitive MEMS Switches, Fig. 10 Simplified typical fabrication process for a capacitive MEMS switch 371 a C d b e c • Step (c): The third mask is used to define the sacrificial layer of the switch. The sacrificial layer is the layer upon which the beam will be deposited, and this material needs to be removed at the end of the process to release the beam. The beam anchor points are defined in this step by selectively etching the sacrificial layer. This can be a dry- or a wet-etch step. More complicated processes may involve an additional planarization step before the next mask. The choice of the sacrificial layer material is critical as it controls many important parameters of the switch design, including the residual stress of the beam. Common material choices include photoresists, other photoconductive polymers, and polyimides. There is no sufficient understanding in the open literature of the exact processes that are involved in controlling the beam layer residual stress in the presence of a sacrificial layer. Important parameters though include, among others, the atomic structures of each film and the deposition temperatures of each film. • Step (d): The fourth mask defines the actual beam layer. A variety of processes can be utilized including evaporation, sputtering, and electroplating of the beam layer(s). This step is also critical in determining the final residual stress of the beam. • Step (e): The beam is finally released by etching (wet or dry) the sacrificial layer and by drying (if needed) the switch wafer. Etching of the sacrificial layer can also influence the beam residual stress particularly if a high-temperature process is necessary. If wafer drying is needed, this needs to be done carefully in order to avoid damage to the beam or causing stiction to the substrate. Stiction may occur if drying involves removing a liquid with high surface tension (e.g., water) underneath the beam. Such a liquid will pull the beam down as it evaporates. Special drying processes and equipment based on supercritical carbon dioxide have been successfully developed [14] and followed by many MEMS researchers. The last step in the fabrication process is packaging, which is discussed in the following section. Packaging Hermetic packaging is required for capacitive RF MEMS switches to avoid any contamination- or humidity-induced early failure. While conventional hermetic packages exist, they are not well suited for capacitive RF MEMS switches or circuits. First, if switches need to be inserted in conventional hermetic packages, these switches will have to be diced first since several thousands of them can be simultaneously fabricated on a wafer. Dicing released switches is particularly dangerous for the switches and may considerably reduce the process yield. Second, conventional hermetic packages are expensive (
tens of dollars/package) and not well-suited for cost-driven consumer applications. Third, they typically exhibit a relatively high insertion loss, which is often much higher than the switch itself (e.g., a DC-40 GHz package could exhibit 0.6 dB at 20 GHz [1]). As a result, it is important to follow a cost-effective on-wafer hermetic packaging scheme. In this case, a wafer-scale package is first completed and then dicing follows. A wide variety of approaches have been developed so far to accomplish this. These approaches can be divided into three main categories: C C 372 Capacitive MEMS Switches • Two-wafer hermetic packages completed by fusion, glass-frit, thermocompression, eutectic, or anodic bonding. These packages are created by bonding two wafers together using one of the aforementioned approaches. The main advantage of these techniques is that they result in excellent hermetic bonds. Their main drawback is that they may require high temperatures (300–1,000 C depending on the technique) with the exception of low-temperature eutectic bonds (e.g., indium–gold bonds). In addition, these techniques tend to be relatively expensive since packaging cost usually accounts for 60–80% of the total cost. • Two-wafer quasi-hermetic packages completed by low-temperature polymer or solder-bump bonding. This technique is similar to the previous one, except that sealing is achieved by low-temperature bonding (room temperature  150 C). A wide variety of polymers can be used for this. While low temperatures can be achieved, packages fabricated with such bonding typically exhibit very low but nonzero leak rates [1]. • Hermetic packages fabricated by on-wafer microencapsulation using micromachining techniques [15]. These techniques do not require a second wafer cap. Instead, every switch or switching circuit is encapsulated in a tiny package on its own wafer by a microfabricated technique. Typical temperatures in this process are in the 200–250 C range. These techniques are ideally suited for the small size and high RF bandwidth of MEMS devices and typically result in low-cost fabrication. Particular attention needs to be paid though to ensure compatibility between the switch and package fabrication processes. Figure 11 shows a switch packaged with this technique. pursuing this particularly due to the large cell phone market size. The most important of the high-frequency circuit applications (> 10 GHz) are high-isolation switching packets, true-time delay networks and phase shifters, reconfigurable impedance tuners for amplifiers and antennas, high-quality-factor reconfigurable filters, and tunable oscillators. Many of these circuits exploit the near-ideal RF performance of capacitive switches. Consequently, optimal performance can be usually achieved by employing a circuit- or sub-system-level package instead of a device-level package. Examples of several of these circuits can be found in [1]. Circuits and Applications Failure Mechanisms and Reliability Capacitive MEMS switches may be employed in a number of circuits mostly for communication, radar, and electronic warfare systems [1]. Variable capacitors and impedance tuners for cell phones and other radios in mobile form factors constitute the most important applications in the commercial sector. MEMS variable capacitors (varactors) can be formed by connecting in parallel several capacitive MEMS switches and selectively activating them. Several companies including WiSpry and Cavendish Kinetics are Capacitive MEMS switches suffer from high electric dc and/or RF fields through narrow gaps and dielectric layers. These fields can readily reach 5–50 V/mm in the up state and may increase further during actuation. Such fields may cause field emission and ionize the gas in the switch gap [16]. The long-term effects of field emission and gas discharge are not known at this point. In addition, when such fields are applied across a thin-film solid dielectric, charges may get trapped in the dielectric layer leading to dielectric charging. Capacitive MEMS Switches, Fig. 11 The memtronics switch (After [15] with permission) packaged Capacitive MEMS Switches A number of studies have been completed (see for example [17, 18]) focusing mostly on charges trapped in the bulk of the dielectric. However, surface charging of the solid dielectric that is also influenced by gas ionization is potentially more detrimental to the switch performance and is not well understood today. Charging phenomena are the leading cause of failure in capacitive switches today. Long-term drift of actuation voltage, stiction, and breakdown can be observed as a result of these charging issues. Besides solid and gas dielectric charging, metal creep is another potential failure mechanism. Creep may be developed in the movable beam material if a switch is subjected to a constant stress. For example, if a switch is left in its down state for a long time (typically tens to thousands of hours), the beam material may creep resulting in a temporary or permanent change of the switch spring constant. Several recent papers show it is potentially an area of concern for capacitive MEMS switches [19, 20]. Creep at high temperature may be even a more significant area of concern. Other possible failure modes include • Beam buckling due to high temperatures. This may be caused during release process, normal hightemperature operation, or due to high RF currents through the movable beam under high RF power conditions. • Self-actuation of the switch movable beam due to high RF power [8]. A high RF voltage may result in self-actuation because the attractive electrostatic force is proportional to the square of the switch voltage. This limits the switch power handling. • Hot switching failure. When a capacitive switch needs to interrupt high RF currents or sustain high transient RF voltages, abrupt chemical changes may occur at its surfaces leading to premature wear and failure. This is related to the dielectric charging phenomena. • Shock-induced failure. High shocks (>30,000– 100,000 g) may result in beam fracture particularly if contact is achieved. Such events are rare in most applications. Failures related to cycling-induced fatigue, crack generation, and fracture are not typically observed under normal operating conditions. However, they may become important at extreme temperatures particularly for movable beams based on thin-film metals. Despite the aforementioned failure modes, the best switches today have achieved over 100 billion cycles under typical laboratory conditions when driven by 373 C 30 kHz bipolar bias waveforms with approximately 35 V peak amplitude. These devices were hot-switched at a power level of 20 dBm at 35 GHz [21]. However, these cannot be considered typical results. Early failures are found in several wafer samples. Additional research is required to increase the observed reliability and limit early failures that are commonly due to poor fabrication process control. Cross-References ▶ Basic MEMS Actuators ▶ NEMS Piezoelectric Switches ▶ Piezoelectric MEMS Switch References 1. Rebeiz, G.M.: RF MEMS Theory, Design, and Technology. Wiley, Hoboken (2003) 2. Webster, J.R., Dyck, C.W., Sullivan, J.P., Friedmann, T.A., Carton, A.J.: Performance of amorphous diamond RF MEMS capacitive switch. Electron. Lett. 40(1), 43 (2004) 3. Goldsmith, C., Sumant, A., Auciello, O., Carlisle, J., Zeng, H., Hwang, J.C.M., Palego, C., Wang, W., Carpick, R., Adiga, V.P., Datta, A., Gudeman, C., O’Brien, S., Sampath, S.: Charging characteristics of ultra-nano-crystalline diamond in RF MEMS capacitive switches. In: Proceedings of the IEEE MTT-S International Microwave Digest, Anaheim, CA, pp. 1246–1249, May 2010 4. Goldsmith, C.L., Kanack, B.M., Lin, T., Norvell, B.R., Pang, L.Y., Powers, B., Rhoads, C., Seymour, D.: Micromechanical microwave switching. US Patent 5,619,061, 31 Oct 1994 5. Goldsmith, C.L., Yao, Z., Eshelman, S., Denniston, D.: Performance of low-loss RF MEMS capacitive switches. IEEE Microwave Wireless Compon. Lett. 8(8), 269–271 (1998) 6. Peroulis, D.: RF MEMS devices for multifunctional integrated circuits and antennas. PhD Dissertation, The University of Michigan, Ann Arbor, (2003) 7. Peroulis, D., Pacheco, S.P., Sarabandi, K., Katehi, L.P.B.: Electromechanical considerations in developing lowvoltage RF MEMS switches. IEEE T. Microw. Theory 51(1), 259–270 (2003) 8. Peroulis, D., Pacheco, S.P., Katehi, L.P.B.: RF MEMS switches with enhanced power-handling capabilities. IEEE T. Microw. Theory 52(1), 59–68 (2004) 9. Snow, M.: Comprehensive modeling of electrostatically actuated MEMS beams including uncertainty quantification. MSc Thesis, Purdue University, West Lafayette, (2010) 10. Younis, M.I., Abdel-Rahman, E.M., Nayfeh, A.: A reducedorder model for electrically actuated microbeam-based MEMS. J. Microelectromech. S. 12(5), 672–680 (2003) 11. Guo, X., Alexeenko, A.: Compact model of squeeze-film damping based on rarefied flow simulations. J. Micromech. Microeng. 19(4), 045026 (2009) C C 374 12. Parkos, D., Raghunathan, N., Venkattraman, A., Alexeenko, A., Peroulis, D.: Near-contact damping model and dynamic response of micro-beams under high-g loads. In: Proceedings of the IEEE International Conference on Micro Electro Mechanical Systems (MEMS), Cancun, Mexico, pp. 465– 468, Jan 2011 13. Ayyaswamy, V., Alexeenko, A.: Coarse-grained model for RF MEMS device. MEMShub.org, http://memshub.org/ resources/prismcg. Accessed on Mar 2011 14. Tousimis, http://www.tousimis.com/. Accessed on March 2011 15. Memtronics, http://www.memtronics.com/. Accessed on March 2011 16. Garg, A., Ayyaswamy, V., Kovacs, A., Alexeenko, A., Peroulis, D.: Direct measurement of field emission current in E-static MEMS structures. In: Proceedings of the 24th IEEE International Conference on Micro Electro Mechanical Systems (MEMS 2011), pp. 412–415, Jan 2011 17. Peng, Z., Yuan, X., Hwang, J.C.M., Forehand, D.I., Goldsmith, C.L.: Superposition model for dielectric charging of RF MEMS capacitive switches under bipolar control-voltage waveforms. IEEE T. Microw. Theory 55(12), 2911–2918 (2007) 18. Papaioannou, G., Exarchos, M.-N., Theonas, V., Wang, G., Papapolymerou, J.: Temperature study of the dielectric polarization effects of capacitive RF MEMS switches. IEEE T. Microw. Theory 53(11), 3467–3473 (2005) 19. Hsu, H.-H., Peroulis, D.: A viscoelastic-aware experimentally-derived model for analog RF MEMS varactors. In: Proceedings of the 23rd IEEE International Conference on Micro Electro Mechanical Systems (MEMS 2010), Wanchai, Hong Kong, pp. 783–786, Jan 2010 20. McLean, M., Brown, W.L., Vinci, R.P.: Temperaturedependent viscoelasticity in thin Au films and consequences for MEMS devices. IEEE/ASME J. Microelectromech. S. 19(6), 1299–1308 (2010) 21. Goldsmith, C., Maciel, J., McKillop, J.: Demonstrating reliability. IEEE Microwave Mag. 8(6), 56–60 (2007) Capillarity Induced Folding ▶ Capillary Origami Capillary Flow Prashant R. Waghmare and Sushanta K. Mitra Micro and Nano-scale Transport Laboratory, Department of Mechanical Engineering, University of Alberta, Edmonton, AB, Canada Synonyms Passive pumping; Surface tension–driven flow Capillarity Induced Folding Definition The fluid flow in an enclosed conduit due to simultaneous changes in the inherent surface energies of fluid and solid surface of the conduit. Introduction Everything in the universe has its own state of energy, which is represented by possible combinations of 132 elements of periodic table. The simplest form of each element is an atom and each atom has three different components: proton, electron, and neutron. Further, each element has a fixed number of electrons, which arrange themselves in different shells, and the number of electrons in each shell can be determined by Bohr’s theory. Several individual elements or atoms do not have sufficient number of electrons in their outer shell and this makes it unstable and further the element tries to become stable by searching for required electrons. The finding of sufficient electrons at the outer shell results in the formation of a molecule or in bulk cluster of molecules. The surface or interface formation with this cluster of molecules creates an imbalance in the arrangement and orientation of the molecules in the cluster, mainly across the interface. This imbalance represents the energy of system. Every system in the universe tries to attain the minimum state of energy and, therefore, in the case of liquid in the air, it has been observed that the raindrops always attain the spherical shape. Liquid molecules have capability to orient themselves. Hence, they can form the shape of minimum energy but in the case of solid molecules, they try to minimize the state of energy by covering up the liquid if it comes in contact. In this case, the solid surfaces are not in equilibrium with the saturated vapor, i.e., they are not in a state of minimum energy. The moment at which conduit or channel with high energy level surfaces comes in contact with lower energy level liquid interface, the solid surface tries to envelope itself with the liquid to attain the minimum energy. In the case of the higher energy liquid in comparison with the solid surfaces, the liquid interface tries to minimize its surface. The prior case can be illustrated with the water (72 dyn/cm) in glass channel (
200–300 dyn/cm) and the mercury (
700–735 dyn/ cm) with the same glass channel is an example of the latter case. Capillary Flow The recent developments in the microfabrication technologies allowed to fabricate features of sizes from micro- to nanoscales. The surface to volume ratio of the feature increases as the scale of the feature decreases which in turn makes surface forces dominant over other forces. Because of high surface forces, very high pressure is required to pump the fluid in microchannels. Hence, researchers have developed different nonmechanical pumping mechanism with the help of electrokinetic and/or megnetohydrodynamic approaches. Generation and actuation of electric and/ or magnetic field are an additional burden on the system which increases both the fabrication process and cost of the device. Therefore, attempts are being made to develop a flow without any external means. The fluid flow can be achieved by controlling surface chemistry, fluid properties like surface tension, or by changing the geometries. Such transport of fluid is called autonomous flow or autonomous pumping which is an ideal transport mechanism for microfluidic applications. The dominance of surface forces at microscale plays a significant role in deciding the pumping approach in microfluidic devices. Hence, nowadays surface tension–driven flow in microfluidic devices has widely attracted the attention of researchers. As explained earlier, the capillary action is the interplay between the surface energies or surface tension between the fluid and solid surface in contact. Moreover, at microscale, due to very high surface to volume ratio, the possibility of available surface area is very high. One can easily pump the fluid with capillarity provided the fluid has lower surface energy than the solid surfaces. For optimum design and function of any microfluidic device which works on capillary flow principle, it is essential to predict its behavior in advance and it is therefore necessary to perform the theoretical analysis of capillary flow within the microchannels. The prediction of the temporal variations in the flow front position along capillary length is the ultimate goal of the theoretical analysis. Therefore, generally the analysis is performed to predict the flow front position, i.e., penetration depth for given working and operating conditions. Over the last century, the capillary phenomenon has become a topic of interest due to its importance in several areas. First time in literature, Washburn proposed a closed form solution for the penetration depth in a channel of millimeter dimension. The closed form solution is derived by balancing the surface tension force to viscous force and it is observed that the 375 C penetration depth is proportional to the square root of the time. As explained earlier, capillary phenomenon is the change in the surface energy process and to encompass the concept of change in the surface energy, one can use the thermodynamic approach for analysis. The surface energy of solid is the topic of ongoing debate. Moreover, there are several effects like dynamic contact angle, inlet effects, reservoir effects, suspension flow, etc., which require tedious and cumbersome analysis. Attempts are also being made to present analysis with microscopic energy balance approach where different forces are accounted in terms of different forms of energy. On the other hand, hydrodynamic models, based on the conventional fluid mechanics principle, are easy to implement. Such models are mainly developed by two distinct approaches in the literature, namely, differential and integral approach. The moving fluid-air interface with the differential approach becomes computationally costly, whereas the integral approach with moving control volume provides a simple form of ordinary differential equation. In such modeling or analysis, the governing equation for flow front transport is obtained by balancing different forces like viscous, inertial, gravity, pressure forces, etc. The velocity-dependent terms like inertial and viscous terms of the governing equation are determined with the velocity profile across the channel. In the literature, the steady state assumption is applied from the very entrance of the channel neglecting the entrance length effect. This assumption has been widely adopted till date as done by Washburn where the time and length scale used for the validation of the theoretical model was big. Hence, the assumption of a steady state velocity profile holds true. Whereas, in the case of microfluidic channels the length and timescale is very small; hence, the Washburn prediction does not follow the observation as demonstrated by Saha and Mitra [1, 2]. The later part of this entry is dedicated to emphasize the importance of such microscale effects in the analysis. Main emphasis is given to the integral approach based on modeling due to the ease of its adaptability. Therefore, in the next section, the overview of the modeling of the capillary transport is discussed in brief. Mathematical Modeling Figure 1 shows the microchannel of width 2B and depth of 2W is considered for the theoretical modeling. C C 376 Capillary Flow Capillary Flow, Fig. 1 Schematic of the microchannel of width 2B, depth 2W considered for theoretical modeling [6] 2W x o z 2B y h The momentum equation in integral form for homogeneous, incompressible, and Newtonian fluid can be written as [3]: X @ Fz ¼ @t  Z Z h 0 B B Z W W Z B B rvz dxdydz þ Z W W vz ðrvz Þdxdy (1) P where Fz refers to all forces present during the development of the fluid-air interface, r is the fluid density. Generally, forces present during the capillary   transport are viscous ðFv Þ, gravity Fg¼ 4rghBW , pressure  forces at the flow front Fpf , and at the inlet Fpi . X Fz ¼ Fv |{z} þ Fg þ Fpf þ Fpi |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl} velocity dependent (2) used widely to determine the pressure force at the entrance of the microchannel. Levin et al. [7], for the first time in the literature, claimed that atmospheric pressure cannot be used as entrance pressure at the inlet of the microchannel. The pressure field expression for circular capillary is derived by assuming a separate hemispherical control volume as fluid source other than the control volume considered within the microchannel. Further several researchers have used same expression for rectangular microchannels with an assumption of equivalent radius. In such analysis, with an equivalent radius assumption, the hemispherical control volume of equivalent radius of projected area of rectangular microchannel entrance is presented in Eq. 3. ( )  2 pffiffiffiffiffiffiffiffi d2 h dh 1:772m dh pðo;tÞ ¼ patm  1:11r BW 2 þ 1:58r þ pffiffiffiffiffiffiffiffi dt dt BW dt (3) velocity independent The Eq. 1 contains three velocity-dependent terms, namely, transient, convective, and viscous force which is generally determined as velocity profile, vz , across the channel. The fully developed flow assumption, i.e., Poiseuille flow assumption, is widely used in the literature neglecting the transience in the velocity profile [4, 5]. The consequence of such assumptions particularly at microscales will be discussed in detail in the later part. The steady state velocity profile can be used to determine the velocity-dependent terms of the momentum equation. Moreover, remaining pressure force terms are calculated with available expressions (from the literature) for pressure fields at respective locations. The pressure force at the fluid-air interface can be determined by well-known Young-Laplace equation with fluid surface tension (s) and equilibrium contact angle (ye). The approximated pressure field expression at the entrance of the microchannel is The importance of an appropriate entrance pressure field expression for rectangular microchannel is also discussed in detail in later part of the study. Finally, determining all terms of Eq. 1 and rearranging as per order of differential operator, one can obtain the dimension form of the ordinary differential equation which governs the capillary transport in the microchannels. Further, nondimensional governing equation as shown in Eq. 4 can be obtained by performing nondimensional analysis with characteris2BÞ2 tic time, t0 ¼ rð12m and characteristic length h0 ¼ 2B.   2 d 2 h dh dh þ ðC3 þ C4 h Þ  þ C 2 2   dt dt dt  þ C5 h þ C6 ¼ 0 ðh þ C1 Þ (4) The coefficients of Eq. 4 are tabulated in Table 1. Two nondimensional numbers are obtained, i.e., Bond Capillary Flow 377 Capillary Flow, Table 1 Constants of the generalized nondimensional governing equation for a capillary flow in a microchannel with fully developed velocity profile [3] Constants C1 Expressions 0:55 pffi g 0:958 1 pffiffiffi 0:295 g C2 C3 C4 C5 Bo 144Oh2 gcosye 72Oh2 C6 number (Bo) and Ohnesorge number (Oh). The constants of this equation are functions of different nondimensional groups like Ohnesorge number (Oh), Bond number (Bo), and aspect ratio (g). The Ohnesorge number represents the ratio of viscous to m surface tension force, i.e., Oh ¼ pffiffiffiffiffiffiffiffi , the Bond num- or developing velocity profile instead of the steady state velocity profile as explained in the following section. The transient momentum equation in the direction of the flow for pressure-driven flow is: r The solution of Eq. 4 predicts the penetration depth with capillary flow. The numerical [8] and analytical [3] solutions of Eq. 4 are available in the literature. Revisiting the Assumptions for Microscale Applications As mentioned earlier, the velocity-dependent terms of momentum equation are determined with an assumption of a steady state. It is assumed that at the very entrance of the microchannel the flow is fully developed. However, in reality, three different flow regimes can be observed in the capillary flow: entry regime, Poiseuille regime, where the flow is fully developed, and the regime behind the fluid-air interface, i.e., surface tension regime. The steady state assumption, i.e., parabolic velocity profile assumption, is valid for steady state flow, whereas capillary flow is inherently a transient phenomenon. The parabolic velocity profile assumption is a reasonably good assumption for macroscale capillaries as shown by several researches [5, 7, 9]. Moreover, such assumptions are only valid in the case of a very high viscous fluid or very low Reynolds number flow which may not be true in every case [10]. Hence, it is important to consider and analyze such transience in the analysis at microscale. This can be tackled by considering the transient @vz @ 2 vz dp ¼m 2 þ dz @t @x (5) The velocity in Eq. 5 is a combination of steady and transient part of velocity as depicted in Eq. 6 [11]: vz ðx; tÞ ¼ vz1 ðxÞ þ vzt ðx; tÞ (6) where vz1 ðxÞ is the fully developed or steady state velocity, i.e., vz1 ðxÞ ¼ 2Brs ber dictates the ratio of gravity to surface tension force,  2 Þ . i.e., Bo ¼ rgð2B s C x B2 dp 1 B 2m dz 2 (7) The transient part of the velocity can be obtained by separation of the variable method which can be given as [6]: # 1 dp vzt ðx; tÞ ¼ 2 ð1Þ cosðln xÞexp Bml3n dz n¼1    nl2n t 1 X n " (8) where n is the kinematic viscosity of the fluid and ln ¼ ð2n1Þp 2B . By combining Eqs. 5 and 7 the transient velocity profile vz ðx; tÞ can be obtained by: ( ! 1 2 ð1Þ vz ðx; tÞ ¼ cosðln xÞexp Bm l3n n¼1 ) 1 2 dp 2 2  ðnln tÞ þ ðB  x Þ 2m dz 1 X n (9) Further, the average velocity across the channel can be represented as: vz ðtÞavg " 1 X B2 96 ¼ exp 1 3m ð2n  1Þ4 p4 n¼1 !# ð2n  1Þ2 p2 nt dp   4B2 dz (10) C 378 Capillary Flow Capillary Flow, Fig. 2 Transient response in the difference in the penetration depths with the fully developed (steady state) and developing (unsteady) velocity profile under different conditions [6] 18 Bo = 0.0076; Oh = 0.0075 Bo = 0.0076; Oh = 0.05 16 Bo = 0.01; Oh = 0.0075 % difference in penetration depths C 14 12 18 16 10 14 8 10 12 8 6 6 4 4 2 0 2 0 10 20 30 40 50 0 0 200 400 600 800 1000 1200 Time(t*) Finally, the velocity profile in terms of the penetration with transient velocity is: " # ( 1 X   B2 4 n cosðln xÞexp nl2n t ð1Þ vz ðx; tÞ ¼ a1 3 2m ðBln Þ n¼1 9 8 > > > >   = dh < x2 1 þ 1 2  1   > dt P > B > ; :a1 1  b1 exp l2n nt > n¼1 (11) where h i 2 ðfÞ4  4exp  f3 t i a1 ¼ h 2 ðfÞ4  6exp  f3 t and f ¼ ln B. A similar approach can be followed as explained in the previous section and further the governing equation for capillary transport can be derived with the transient velocity profile provided in Eq. 11. Moreover, the difference in the penetration depth with both approaches, i.e., with the steady state and transient velocity profile, under different operating conditions can be compared. Figure 2 shows the difference in the penetration depth in such cases where the difference in the penetration depth is more at the beginning of the filling process, as shown in the inset of Fig. 2. This difference in the penetration depth decreases as the flow progresses along the microchannel where the flow becomes a fully developed flow. This can also be explained with the help of boundary layer theory which is the effect of fluid viscosity. The boundary layer thickness increases as the viscosity of fluid increases because of the retardation of flow due to increase in the viscosity, whereas in the case of the fluid density, the effect is opposite to viscosity. Therefore, the difference in penetration depth with the high density fluid (Bo ¼ 0.01) is higher than the difference with the high viscous fluid. It is evident from the analysis that the transience effect in the analysis has a significant impact on the filling process prediction, particularly at the beginning of the filling process. At microscale, such difference needs to be accounted prior to the design. As discussed earlier, the pressure force at the entrance of the microchannel is determined with the help of the pressure field at the microchannel entrance. Several researchers [3–8] have adopted the pressure field expression with an equivalent radius assumption. Levin et al. [6] developed an entrance pressure field expression for circular capillary, assuming Capillary Flow 379 Capillary Flow, Fig. 3 The fluid volume from infinite reservoir considered as control volume for pressure field expression analysis in the case of rectangular microchannel. The arrow shows the direction of the fluid flow from the reservoir into the microchannel [6] 2B lc rs rc C ( 4g þ 3ð1  gÞ 24 ) ( ) 1 2 6 2 R1 d2 h þ 2þ þ 1  ln p 2p p 10 p B dt2 4ð1  gÞ 6 4g þ 3ð1  gÞ   p2 5 6 )  ð2  gÞ ð1  gÞ dh 2 4m    2p p2 dt B þr  ð2  gÞ þ z=0 Oc Os a hemispherical control volume as a separate control volume at the entrance which is responsible for a sink flow at the entrance of capillary and the pressure field. Moreover, a similar expression for rectangular capillaries is extended with an equivalent radius assumption. In such cases, the radius of circular capillary is replaced by the equivalent radius of projected area at the entrance of the channel. This is not a realistic representation for noncircular capillaries particularly for high aspect ratio microchannels where it is not appropriate to consider the hemispherical control volume for the sink flow or pressure field at the microchannel entrance. In the case of such geometries, the control volume needs to be considered as a combination of semicylinder and hemisphere as shown in Fig. 3. The detailed derivation of the pressure field expression with this control volume can be seen in [6] which is: pð0; tÞ ¼ patm  rB l inlet plane Microchanne C ð1  gÞ dh p dt (12) where R1 represents the radial distance far away from the control volume in the reservoir, where the sink action, i.e., entrance pressure force, disappears. One can re-derive the governing Eq. 4, using pressure field expression presented in the Eq. 12, and determine the effect of such a pressure field on the analysis. Figure 4 shows the comparison of variations in the penetration depth with recently proposed pressure and with equivalent radius field expressions. The approximated pressure field overpredicts the penetration depth. The difference in the penetration depth with the proposed pressure field is significant, which shows that it is important to consider the proposed pressure field for a rectangular microchannel rather than an approximated pressure field. The transport with a capillary action is the balance among surface, viscous, and other body forces which retard the flow as it progresses. Hence, the capillary flow always attains a steady state which is generally termed as an equilibrium penetration depth in the literature. If the length of the channel is longer than that of the equilibrium penetration depth, then flow front cannot reach the outlet and, therefore, the assistance to the capillary flow is attempted in such cases. Passive or nonmechanical pumping approaches combined with the capillary flow serve this enhancement. The scaling analysis suggests that the gravity force is less dominant at microscale [12], but several researchers have demonstrated that gravity can be used as an assistance to the capillary flow [13–15]. Generally, the capillary flow analysis is performed with an assumption of infinite reservoir. Hence, the reservoir effect and the gravitational force from the reservoir are generally neglected in the analysis. To accommodate the entrance effect of finite size reservoir at the inlet of the microchannel in the theoretical modeling, the entrance pressure field is developed for the arrangements shown in Fig. 5. The rectangular microchannel with rectangular reservoir on the top of the microchannel is considered and the pressure field with the gravity and reservoir effect is developed in the flow [16]. Moreover, this pressure field C 380 12 Capillary Flow Bo = 0.0076; Oh = 0.0075 2B 2W 2 2 Reservoir 8 z 6 y H x 4 2 h Capillary flow front Pressure field with equivalent radius Proposed pressure filed 0 0 10 20 30 40 50 Time(t*) Capillary Flow, Fig. 4 The comparison of variations in the penetration depth with equivalent radius and recently proposed pressure field expressions. Figure 4 shows the comparison of penetration depth for g ¼ 0:9 with the corresponding difference in the penetration depth [6] is used to obtain the governing equation for capillary flow under the influence of gravity head from the reservoir. The reservoir with three different levels of fluid in the reservoirs (H*), namely, 10, 50, and 100, is considered for the analysis. Figure 6 shows the variations in the penetration depth (h*) with different operating conditions. This analysis represents the interplay between the surface tension force and gravity head from the reservoir. The capillary flow takes place in the channel which remains same for all three cases, whereas the level of the fluid from the reservoir increases from case I to case III. Thus, for three cases, the capillary effect is the same but the gravity head is different. At the beginning of the transport, the fluid from the reservoir offers less inertia to the fluid transport and the capillary force dominates over the gravity from the reservoir. Therefore, at the beginning of the transport, the penetration depth with a lower reservoir fluid level (H* ¼ 10.0) is higher than the other two penetration depths as shown in the inset I. Similarly, the penetration depth with the highest reservoir fluid level (H* ¼ 100.0) has the lowest penetration depth as compared to others. Moreover, as the fluid progresses in the microchannel, the momentum from the reservoir fluid assists the capillary flow and the gravitational force, 2B Capillary Penetration depth(h*) 10 2W1 1 Capillary Flow, Fig. 5 Schematic of a gravity-assisted capillary flow in a vertically oriented capillary of width 2B1 and depth 2W1. The additional gravitational head from the fluid in a finite reservoir of size (2B2  2W2) is assisting the capillary flow [16] due to which the fluid from the reservoir becomes dominant over the capillary force within the microchannel. This results in transcendence among the penetration depths with a different gravity head. The penetration depth with the highest gravity head (H* ¼ 100) surpasses the penetration depth with gravity head H* ¼ 50 and H ¼ 10* in inset I and II two, respectively. This can be attributed to as an interplay between the surface tension force, i.e., the capillarity and gravitational force from reservoir. In the case of microfluidic applications, the sizes of reservoir and microchannel are comparable to each other. Hence one cannot neglect the effect of the reservoir in such cases, particularly if it is surface tension–driven pumping. Further, one can assist the capillary flow with an appropriate arrangement of reservoir. There are always certain limitations to the autonomous pumping which make them inadequate in long microchannels. Hence, it is important to enhance the pumping ability by other means. Further enhancement in the capillary flow can be achieved by coupling the capillary flow with the electroosmotic flow which is one of the electrokinetic pumping mechanisms. In most of the cases, the inner wall of a microchannel always has surface charges due to different mechanisms like ionization, dissociation of ions, isomorphic substitution, etc., [17]. These surface charges Capillary Flow 120 90 Flow front penetration (h*) Capillary Flow, Fig. 6 Transient response of a flow front transport for different gravitational heads in the reservoir with Bo ¼ 0.0055, Oh ¼ 0.0084, B1/W1 ¼ 0.05 B2/W2 ¼ 0.2. I. Flow front penetration rate for H* ¼ 100 surpasses the penetration rate for H* ¼ 50. II: Flow front penetration rate for H* ¼ 100 surpasses the penetration rate for H* ¼ 10. III: Flow front penetration rate for H ¼ 50 surpasses the penetration rate for H* ¼ 10 [16] C 381 H* = 10.0 H* = 50.0 74 H* = 100.0 72 Bo = 0.0055 Oh = 0.0084; B1/W1 = 0.05; B2/W2 = 0.2; 70 C 68 60 III 66 12 44 13 14 15 16 60 40 56 36 30 32 52 I 28 2.5 2.0 0 0 14 3.0 3.5 II 48 4.0 4 28 5 42 6 7 8 56 9 10 70 Time(t*) distribute ions of the electrolytes in a specific pattern when brought into contact with an electrolyte which is generally termed as the formation of electrical double layer (EDL). After applying the electric field across the channel, the movement of the ions takes place, which results in the movement of the fluid due to an electric field [18]. The electrolyte solution is transported with the capillary action; one can further assist the capillary flow. An additional body force due to electroosmotism is added to Eq. 2, which accommodates the additional effect of electroosmosis. Further one can analyze the interplay between the capillarity and electroosmotisms as presented in the recent studies [19]. Figure 7 shows the variation in the penetration depth of the capillary flow under the influence of electroosmotism. Through a nondimensional analysis, a new nondimensional number is proposed, i.e., Eo which represents the ratio between the surface tension force and electroosmotic force. The direction of the electroosmotic flow can be reversed by changing the electric field direction. Hence, negative and positive Eo numbers are observed in the analysis. The negative Eo numbers represent the change in the direction of the electric field as compared to positive Eo numbers. The pure capillary flow can be seen as Eo ¼ 0. The variation in the penetration depth under three different operating conditions is shown in Fig. 7. As observed in the pure capillary case (Eo ¼ 0), the penetration depth attains the equilibrium penetration depth, whereas in the case of Eo numbers, the equilibrium penetration depth increases with increment in the magnitude of Eo numbers. This represents that in the case of electroosmotic flow with Eo number, the capillary flow is assisted by electroosmotism. In this analysis, the nondimensional length of the microchannel (L*) is considered as 300 and with –Eo ¼ 0.01 the entire filling of the microchannel is observed. In the case of positive Eo numbers, it is observed that the electroosmotism acts in a opposite direction of the capillary flow. Hence, it retards the flow and this can be observed by the decrement in the equilibrium penetration depth with the increment in the positive Eo numbers. For the enhancement in the capillary flow, a transport with additional gravity head and electroosmotic forces are considered. A generalized theoretical modeling for a gravity-assisted capillary flow with reservoir effects and electroosmotically assisted capillary flow is reported in brief. It is observed that even though the scaling among forces suggests that the 382 300 Oh = 0.0070 E0 = −0.005 Bo = 0.0075 E0 = −0.001 E0 = −0.01 γ = 0.006 250 ε* = 0.07 L* = 300 200 300 Bκ = 4.0 E0 = 0.01 150 E0 = 0.001 E0 = 0.0 E0 = 0.005 100 50 250 Flow front penetration (h*) Capillary Flow, Fig. 7 Variation in the penetration depth for vertically oriented channel with water as electrolyte, where B ¼ 100 mm, W ¼ 400 mm, L ¼ 75 mm, z ¼ 75 mV and constant contact angle is 27 . The inset shows the variation in the electric field within the electrolyte as the flow front progresses under different applied voltages [19] Capillary Flow Penetration depth (h*) C E0 = −0.5 E0 = −0.1 200 E0 = −0.05 150 100 50 0 0 50 100 150 Time(t*) 0 0 200 400 600 800 200 250 1000 Time(t*) gravitation force is negligible at microscale, the reported analysis infers that with a finite reservoir, an added advantage due to gravity can be a useful tool to transport the fluid at microscale. This added force for the capillary transport can be utilized without any additional burden in the design of the LOC device. The electroosmotically assisted capillary flow model suggests that in a combined flow the electrokinetic parameters have an important influence on the capillary flow. Such electrokinetic flow approaches can be coupled to enhance the capillary flow transport in the microchannel. The wetting properties of the fluid decide the capability of pumping with a capillary flow. Therefore, it is important to know the precise magnitude of wetting properties like contact angle and surface tension of the working fluid. The microfluidics has become a promising option for biomedical application and inclusion of biomolecules is an unavoidable part in such applications. In most cases, the biomolecules are attached with the microbeads and transported to the desired locations. It is evident from the experimental analysis that the inclusion of microbeads changes the wetting behavior drastically [20]. Therefore, it is necessary to consider the effect of microbeads in the fluid for the analysis. This can be done by considering the following expressions for surface tension and contact angle: density and viscosity which are functions of volume fraction of microbeads. Such correlations of the surface tension and contact angle are provided in [20]. Such expressions for the variation in the contact angle and surface tension with the volume fraction can be readily used in modeling transport processes of microbead suspensions in micro-capillaries, used in the microfluidic devices. In passive pumping, particularly with the capillary flow, different aspects due to microscale effects like aspect ratio–dependent velocity profile, contact angle at four walls, fluid-air interface dynamics in the case of suspension flow, etc., need to be investigated in detail. Theoretically the concept of electroosmotically assisted capillary flow has been presented but the experimental demonstration of such phenomena is also an interesting area of research. Moreover, wetting of biomolecule suspensions under transient effects instead of the steady state is also needed to be studied. The experimental study of the flow behind the front and at the entrance of the microchannels is also an interesting study to perform. Capillary Origami Cross-References ▶ AC Electroosmosis: Basics and Lab-on-a-Chip Applications ▶ Electrowetting ▶ Micro/Nano Flow Characterization Techniques ▶ Micropumps ▶ Surface Tension Effects of Nanostructures ▶ Wetting Transitions References 1. Saha, A., Mitra, S.K.: Numerical study of capillary flow in microchannels with alternate hydrophilic-hydrophobic bottom wall. J. Fluid Eng. Trans. ASME 131, 061202 (2009) 2. Saha, A., Mitra, S.: Effect of dynamic contact angle in a volume of fluid (VOF) model for a microfluidic capillary flow. J. Colloid Interface Sci. 339, 461–480 (2009) 3. Xiao, Y., Yang, F., Pitchumani, R.: A generalized flow analysis of capillary flows in channels. J. Colloid Interface Sci. 298, 880–888 (2006) 4. Washburn, E.: The dynamics of capillary flow. Phys. Rev. 17, 273 (1921) 5. Chakraborty, S.: Electroosmotically driven capillary transport of typical non-Newtonian biofluid in rectangular microchannels. Anal. Chim. Acta 605, 175–184 (2007) 6. Waghmare, P.R., Mitra, S.K.: A comprehensive theoretical model of capillary transport in rectangular microchannels. Microfluid. Nanofluid. (2011). doi:10.1007/s10404-0110848-8 7. Levin, S., Reed, P., Watson, J.: A theory of the rate of rise a liquid in a capillary. In: Kerker, M. (ed.) Colloid and Interface Science, p. 403. Academic, New York (1976) 8. Marwadi, A., Xiao, Y., Pitchumani, R.: Theoretical analysis of capillary-driven nanoparticulate slurry flow during a micromold filling process. Int. J. Multiph. Flow 34, 227 (2008) 9. Dreyer, M., Delgado, A., Rath, H.: Fluid motion in capillary vanes under reduced gravity. Microgravity Sci. Technol. 4, 203 (1993) 10. Bhattacharya, S., Gurung, D.: Derivation of governing equation describing time-dependent penetration length in channel flows driven by non-mechanical forces. Anal. Chim. Acta 666, 51–54 (2010) 11. Keh, H., Tseng, H.: Transient electrokinetic flow in fine capillaries. J. Colloid Interface Sci. 242, 450 (2001) 12. Nguyen, N., Werely, S.: Fundamentals and Applications of Microfluidics. Artech House, New York (2003) 13. Yamada, H., Yoshida, Y., Terada, N., Hagihara, T., Teasawa, A.: Fabrication of gravity-driven microfluidic device. Rev. Sci. Instrum. 79, 124301 (2008) 14. Jong, W.R., Kuo, T.H., Ho, S.W., Chiu, H.H., Peng, S.H.: Flows in rectangular microchannels driven by capillary force and gravity. Int. Commun. Heat Mass Transf. 34, 186–196 (2007) 15. Kung, C., Chui, C., Chen, C., Chang, C., Chu, C.: Blood flow driven by surface tension in a microchannel. Microfluid Nanofluid 6, 693 (2009) 383 C 16. Waghmare, P.R., Mitra, S.K.: Finite reservoir effect on capillary flow of microbead suspension in rectangular microchannels. J. Colloid Interface Sci. 351(2), 561–569 (2010) 17. Hunter, R.: Zeat Potential in Colloid Science, Principle and Applications, Principle and Applications. Academic, London (1981) 18. Israelachvili, J.N.: Intermolecular and Surface Forces. Academic, London (1998) 19. Waghmare, P.R., Mitra, S.K.: Modeling of combined electroosmotic and capillary flow in microchannels. Anal. Chim. Acta 663, 117–126 (2010) 20. Waghmare, P.R., Mitra, S.K.: Contact angle hysteresis of microbead suspensions. Langmuir 26, 17082–17089 (2010) Capillary Origami Supone Manakasettharn, J. Ashley Taylor and Tom N. Krupenkin Department of Mechanical Engineering, The University of Wisconsin-Madison, Madison, WI, USA Synonyms Capillarity induced folding; Elasto-capillary folding; Surface tension–powered self-assembly Definition Capillary origami is folding of an elastic planar structure into a three-dimensional (3D) structure by capillary action between a liquid droplet/bubble and a structure surface. Why Capillary Origami? The fabrication of 3D structures is one of the major challenges for micro- and nano-fabrication. Folding of an elastic planar structure after patterning and release is one technique to fabricate a 3D structure using self-assembly. The term origami is taken from the Japanese art of paper folding; while the actuation of the folding is accomplished by using capillary forces of a fluid droplet, hence the technique has been termed capillary origami. The combination of the folding process with capillary forces has resulted in a new technique for micro- and nano-fabrication. C 384 History The term capillary origami was first introduced in 2007 by Charlotte Py et al. to describe the folding of a polydimethylsiloxane (PDMS) sheet into a 3D structure by using capillary forces created by a water droplet [1]. As early as 1993, Syms and Yeatman demonstrated that 3D structures could be fabricated by folding surfaces using capillary forces produced by molten solder [2]. Later Richard R. A. Syms introduced the term surface tension–powered self-assembly to describe the technique [3, 4]. Both of these techniques are quite similar in that 3D structures can be produced by folding elastic thin films. Both use capillary forces for self-assembly. In the first example, various liquids such as water are used, while for the second study molten metals such as solder were used, which then solidified fixing the 3D microstructures. Principles At the macroscale level, the influence of capillary forces is negligible compared to other forces such as gravity, electrostatic, or magnetic. Because capillary forces scale linearly with the characteristic size of the system, at sub-millimeter dimensions capillary forces begin to dominate since the majority of other forces decrease much more rapidly than the first power of the length. For example, a human cannot walk on water because capillary forces produced at the water surface are much smaller than the gravitational force acting on a human, which scales as the cube of the length. On the other hand, the much smaller water strider can easily walk on water because capillary forces are large enough to balance the gravitational force produced by the water strider. For capillary origami, capillary forces need to be large enough to counteract the weight of the liquid droplet and the structural forces of the planar layer. In terms of energy, for capillary origami one needs to consider the interplay of three different energies: capillary energy, bending energy, and gravitational potential energy. For a two-dimensional (2D) model, the capillary energy per unit length of the interface (2D analog of the surface energy) is defined as Ec ¼ L0 g, where L0 is the length of the interfacial surface of the fluid and g is the surface tension [5]. The bending LB energy per unit length is approximately Eb ¼ 2R 2 , Capillary Origami 1,000 No folding 100 Lcrit / Lec C Folding 10 No folding 1 0.01 0.1 1 10 Lec / Lc Capillary Origami, Fig. 1 Folding criteria plotted from (1) assuming complete circular folding and neglecting the effect of gravity where L is the length of the structure and R is the radius Eh3 of curvature [6]. B ¼ 12ð1v 2 Þ is the bending rigidity of the structure, where E is Young’s modulus, h is thickness of the layer, and v is Poisson’s ratio. If one only considers the mass of the fluid, assuming that it is much larger than the mass of the structure, then the gravitational potential energy per unit length is Eg ¼ rSgz, where r is the density, S is the surface area, g is the constant of gravity, and z is the height of the center of mass. By neglecting the effect of gravity and assuming complete circular folding, de Langre et al. [7] derived simplified criteria for folding considering the interplay between capillary and bending energies, which are expressed as pffiffiffi pffiffiffi Lc Lcrit 2p < < 2 2p Lec Lec (1) where Lcrit is the critical qffiffiffiffi length of a structure for g folding to occur, L ¼ c qffiffiffi rg is the capillary length [5], B and Lec ¼ g is the elasto-capillary length [8]. The simplified criteria for folding derived from (1) can be plotted as shown in Fig. 1. To fold a structure pffiffiffiffiffiffi Brg requires LLecc < 2 or Lc > L2ec or g > 2 indicating that the capillary length must be larger than half of the elasto-capillary length so that the capillary effect can overcome bending rigidity of the structure. pffiffiffi The other requirement for folding is LLcrit > 2p ffi 4:44 or ec Lcrit > 4:44Lec confirming that the length of the structure should also be long enough for a liquid droplet to Capillary Origami C 385 1 mm 1 mm 500 μm Capillary Origami, Fig. 2 Capillary origami 3D structures of a pyramid, a cube, and a quasi-sphere obtained by folding triangle-, cross-, and flower-shaped PDMS sheets, respectively, actuated with a water droplet (Reprinted with permission from [10]. Copyright 2007, American Institute of Physics) Capillary Origami, Fig. 3 Capillary origami 3D structures formed from triangle- and flower-shaped templates using soap bubbles (scale bar: 2 cm) (Images reprinted from [11] with permission) wet the surface to produce sufficient capillary forces to fold the structure. For the 3D structures the folding criteria become more complex. In particular in 3D the critical length also depends on the shape of the initial template such that Lcrit ffi 7Lec for squares and Lcrit ffi 12Lec for triangles [9]. Figure 2 shows examples of capillary origami structures of a pyramid, a cube, and a quasi-sphere obtained from folding triangle-, cross-, and flower-shaped PDMS sheets, respectively [10]. Besides using liquid droplets, capillary origami structures can be constructed by using soap bubbles as shown in Fig. 3. The weight of a soap bubble is much less than that of a liquid droplet especially for large droplets capable of covering centimeter size structures when gravitational forces become significant. A soap bubble was shown to fold a centimetersize elastic structure, which cannot be accomplished using a liquid droplet [11]. Petals of a flower also can be folded into a structure similar to capillary origami when submerged in water as shown in Fig. 4. The folding of the flower in water is accomplished by the interplay of elastic, capillary, and hydrostatic forces. During submersion, hydrostatic pressure pushes against the back of petals, and surface tension prevents water from penetrating through the spacing between petals resulting in trapping an air C C 386 Capillary Origami Capillary Origami, Fig. 4 The folding of an artificial flower when submerged in water (Reprinted with permission from [12]. Copyright 2009, American Institute of Physics) a b 50 μm c d Capillary Origami, Fig. 5 (a) Schematics of initial templates. (b)–(d) SEM images of 3D microstructures after folding (scale bar: 50 mm) (Reprinted with permission from [13]. Copyright 2010, American Institute of Physics) bubble inside a flower. The inside of the folded flower remains dry protected by the air bubble [12]. Applications Capillary origami has been used to fabricate a number of 3D microstructures. Figure 5 illustrates the self-assembly of structures with various geometries. The initial planar templates are shown in Fig. 5(a), and the folded final 3D microstructures are shown in Fig. 5(b)–(d). The initial planar templates with lengths ranging from 50 to100 mm and a thickness of 1 mm were fabricated from silicon nitride thin films deposited and patterned by using standard micromachining processing typically used for integrated circuit and MEMS fabrication. Water droplets then were deposited on the templates to fold 3D microstructures [13]. Figure 6 shows another example of microfabrication of a quasi-spherical silicon solar cell based on capillary origami. After fabrication by conventional micromachining processing, the initial flower-shaped silicon template was folded into a sphere using a water droplet. Unlike, conventional flat solar cells, this spherical solar cell enhanced light trapping and served as a passive tracking optical device, absorbing light from a wide range of incident angles [14]. Structures formed by capillary origami also can be actuated by using electrostatic fields to reversibly fold and unfold then. For this application we need to take into account the interplay of capillary, elastic, and electrostatic forces. As shown in Fig. 7, an electric field was applied between the droplet and the substrate. When the voltage was increased, the electrostatic force increased eventually overcoming capillary forces resulting in unfolding of the PDMS sheet. When the voltage was decreased below a certain threshold, the electrostatic force was no longer strong enough to prevent capillary forces from again folding the elastic sheet [15]. Capillary Origami Cr/Au 387 + n+ p Etch undercuts SO C Fold into a sphere I wa ter C Ag wire Capillary Origami, Fig. 6 Three schematics from left to right showing steps to fabricate a spherical-shaped silicon solar cell. The image at the far right shows the final spherical-shaped silicon solar cell (Images reprinted from [14] with permission) Electrode Droplet Ui 0V + + ++ + + + + + 700 V 200 V Elastic sheet E Isolating layer Counter electrode Capillary Origami, Fig. 7 Capillary origami controlled by an electric field. The schematic of the experimental setup is shown at the far left and three images to the right show the results of increasing voltage from 0 V to 700 V and decreasing voltage from 700 V to 200 V. Images from [15] – reproduced by permission of The Royal Society of Chemistry Capillary origami is a simple and inexpensive method to fabricate 3D structures at the sub-millimeter scale. By using capillary forces, intricate and delicate 3D thin-film structures can easily be fabricated, which would be difficult to obtain by other means. More applications exploiting the advantages of capillary origami itself or in combination with electric fields can readily be envisioned. Ultimately one expects to see more commercial products based on this versatile technique. 3. Syms, R.R.A.: Surface tension powered self-assembly of 3-D micro-optomechanical structures. J. Microelectromech. Syst. 8, 448–455 (1999) 4. Syms, R.R.A., Yeatman, E.M., Bright, V.M., Whitesides, G.M.: Surface tension-powered self-assembly of microstructures – the state-of-the-art. J. Microelectromech. Syst. 12, 387–417 (2003) 5. Berthier, J.: Microdrops and digital microfluids. William Andrew Pub, New York (2008) 6. Timoshenko, S., Woinowsky-Krieger, S.: Theory of plates and shells, 2nd edn. McGraw-Hill, New York (1959) 7. de Langre, E., Baroud, C.N., Reverdy, P.: Energy criteria for elasto-capillary wrapping. J. Fluids Struct. 26, 205–217 (2010) 8. Bico, J., Roman, B., Moulin, L., Boudaoud, A.: Adhesion: elastocapillary coalescence in wet hair. Nature 432, 690 (2004) 9. Py, C., Reverdy, P., Doppler, L., Bico, J., Roman, B., Baroud, C.N.: Capillarity induced folding of elastic sheets. Eur. Phys. J. Spec. Top. 166, 67–71 (2009) 10. Py, C., Reverdy, P., Doppler, L., Bico, J., Roman, B., Baroud, C.: Capillary origami. Phys. Fluids 19, 091104 (2007) 11. Roman, J., Bico, J.: Elasto-capillarity: deforming an elastic structure with a liquid droplet. J. Phys. Condens. Matter 22, 493101 (2010) 12. Jung, S., Reis, P.M., James, J., Clanet, C., Bush, J.W.M.: Capillary origami in nature. Phys. Fluids 21, 091110 (2009) 13. van Honschoten, J.W., Berenschot, J.W., Ondarcuhu, T., Sanders, R.G.P., Sundaram, J., Elwenspoek, M., Tas, N.R.: Elastocapillary fabrication of three-dimensional microstructures. Appl. Phys. Lett. 97, 014103 (2010) Cross-References ▶ Self-assembly ▶ Surface Tension Effects of Nanostructures References 1. Py, C., Reverdy, P., Doppler, L., Bico, J., Roman, B., Baroud, C.N.: Capillary origami: spontaneous wrapping of a droplet with an elastic sheet. Phys. Rev. Lett. 98, 156103 (2007) 2. Syms, R.R.A., Yeatman, E.M.: Self-assembly of threedimensional microstructures using rotation by surface tension forces. Electron. Lett. 29, 662–664 (1993) C 388 14. Guo, X., Li, H., Yeop Ahn, B., Duoss, E.B., Jimmy Hsia, K., Lewis, J.A., Nuzzo, R.G.: Two- and three-dimensional folding of thin film single-crystalline silicon for photovoltaic power applications. Proc. Natl Acad. Sci. U.S.A. 106, 20149–20154 (2009) 15. Pineirua, M., Bico, J., Roman, B.: Capillary origami controlled by an electric field. Soft Matter. 6, 4491–4496 (2010) Carbon Nanotube Materials ▶ Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling Carbon Nanotube-Metal Contact Wenguang Zhu Department of Physics and Astronomy, The University of Tennessee, Knoxville, TN, USA Synonyms Carbon nanotube-metal interface Definition Carbon nanotube-metal contacts are widely present in many carbon nanotube-based nanodevices, and their electronic structures may significantly influence the operation and performance of carbon nanotube-based nanodevices. Overview Carbon nanotubes (CNTs) are quasi-one-dimensional materials with remarkable mechanical and electronic properties promising a wide range of applications from field-effect transistors (FETs) and chemical sensors to photodetectors and electroluminescent light emitters. In most of these CNT-based nanodevices, metals are present as electrodes in contact with the CNTs. Many factors including the CNT-metal contact geometry, Carbon Nanotube Materials microscopic atomic details at the interface, and the resulting electronic structure can play a significant role in determining the functionality and performance of the devices. For instance, it has been demonstrated that an individual semiconducting CNT can operate either as a conventional FET or an unconventional Schottky barrier transistor, depending on the properties of the metal-CNT contact. In general, the electrical transport characteristics of the CNT-metal systems are sensitive to the choice of metal element as the electrode. CNT-Metal Contact Geometry There are two types of interface geometries of CNTmetal contacts, i.e., end contact and side contact [1, 2]. The end-contact geometry refers to the cases where metals are merely in contact with the open ends of one-dimensional CNTs, as illustrated in Fig. 1a. This contact geometry can be naturally achieved in the catalytic CVD growth of CNTs, where CNTs sprout from catalytic metal particles with the CNT axis normal to the metal surface. Figure 1b shows a sample experimental image of an end contact between a single-wall CNT and two Co tips in an in situ electron microscopy setup. The side-contact geometry refers to the cases where metals are in contact with the sidewall of CNTs, as illustrated in Fig. 1c. This contact geometry occurs when a CNT lays on the surface of a flat metal substrate. In most of CNT-based nanodevices, such as CNT FETs, metal strips are deposited from above to cover the CNTs laying on the surface as to build electrodes, fully covering sections of the CNTs, as shown in Fig. 1d. Among these two contact geometries, the side-contact geometry is more technologically relevant to CNT-based nanodevices. Bonding and Wetting Properties of Metals on CNTs In the end-contact geometry, metals form strong covalent bonds with carbon atoms at the open ends of CNTs [3]. The bonding energy can be as high as, e.g., 7.6 eV for a single bond at a CNT–Co contact, according to density functional calculations. Due to the strong covalent nature of the bonding, large mismatch-induced strains or high tensile strength can be built up at the interface. Carbon Nanotube-Metal Contact a b c d Carbon Nanotube-Metal Contact, Fig. 1 A schematic illustration of (a) an end contact and (c) a side contact between a CNT and a metal (From Palacios, J.J., Pérez-Jiménez, A.J., Louis, E., SanFabián, E., Vergés, J.A.: Phys. Rev. Lett. 90, 106801 (2003), Fig. 1). (b) A CNT forming end contacts with Co tips (From Rodŕiguez-Manzo, J.A. et al.: Small, 5, 2710–2715 (2009), Fig. 1). (d) A CNT forming side contacts on gold electrodes (From Anantram, M.P., Léonard F.: Rep. Prog. Phys. 69, 507–561 (2006), Fig. 24) 389 C In the side-contact geometry, metals and CNTs form much weaker bonds due to the nearly chemically inert side walls of CNTs [3]. Single-wall CNTs are built up of a cylindrically closed sheet of graphene, in which carbon atoms arranged in a honeycomb structure form very stable sp2-hybridized covalent bonds with the pz-orbitals of carbon extending normal to the sidewalls. The interaction between metals and CNTs in the side-contact geometry is determined by the hybridization between the carbon pz-orbitals and the unbonded orbitals of the metals. Alkali and simple metals have binding energy around 1.5 eV per atom. Some transition metal atoms with unpaired d electrons, such as Sc, Ti, Co, Ni, Pd, Pt, form strong bonds with a binding energy around 2.0 eV per atom, whereas the transition metals with fully occupied d orbitals such as Cu, Au, Ag, and Zn have relatively weak binding with a binding energy less than 1.0 eV per atom. On the other hand, the binding energy of metal on CNTs also depends on the radius of CNTs. In general, the larger is the radius, the weaker the binding energy. The wettability of metals on CNTs is critical to the electrical transport properties at CNT-metal contacts. In addition, CNTs can be used as templates to produce metallic nanowires with controllable radius by continuously coating the sidewalls of CNTs with metals. Experiments using different techniques such as electron beam evaporation, sputtering, and electrochemical approaches have achieved continuous coating of Ti and quasi-continuous coating of Ni and Pd on CNTs [3, 4]. Such metallic nanowires are ideal to be used as conducting interconnects in nanodevices. Metals such as Au, Al, Fe, Pb form isolated discrete clusters rather than a uniform coating layer on the surface of CNTs. Figure 2 shows sample TEM images of Ti, Ni, Pd, Au, and Fe coatings on CNTs [4]. The correlation between the wettability of these metals and their binding energies on CNTs is clear, i.e., metals with relatively strong binding energies with CNTs tend to form uniform coatings. Electronic Structures of CNT-Metal Contacts The electronic structure of CNT-metal contacts has a significant impact on the operation and performance of CNT-based nanodevices [2]. Due to the one-dimensional nature of CNTs and their special C C 390 Carbon Nanotube-Metal Contact, Fig. 2 TEM images of (a) Ti, (b) Ni, (c) Pd, (d) Au, (e) Al, and (f) Fe coatings on carbon nanotubes contact geometries, CNT-metal contacts exhibit some unusual features when compared to traditional planar contacts. For metallic CNTs, as in contact with metals, ohmic contacts are normally formed at the interface, where no interface potential barrier exists, and the contact resistance is primarily determined by the wettability of the Carbon Nanotube-Metal Contact metal and the local atomic bonding and orbital hybridization at the interface [2]. Palladium is found to be optimal as electrodes to make ohmic contacts with metallic CNTs. Semiconducting CNTs form either ohmic contacts or Schottky barriers at the interface with metals [2]. Figure 3 schematically illustrates the energy levels for an ohmic and a Schottky contact between a metal and a semiconducting CNT. A distinctive feature of CNTmetal contacts from traditional planar metal/semiconductor interfaces is that the height of the Schottky barrier formed at CNT-metal contacts strongly depends on the work function of the metal for a given semiconducting CNT [5, 6]. In general, at traditional planar metal/semiconductor interfaces, the Schottky barrier height shows very weak dependence on the metal work function due to the so-called Fermi-level pinning effect [7]. The strong dependence of the Schottky barrier on the metal work function in CNTmetal contacts is attributed to the reduced dimensionality of CNTs, which entirely changes the scaling of charge screening at the interface, making the depletion region decay rapidly in a direction normal to the interface and thus significantly weakening the Fermi-level pinning effect [5]. Experimental and theoretical work has shown that the interface Schottky barrier regions are much thinner in one dimension than those in three dimensions. In this case, charge carrier tunneling through the Schottky barriers becomes important. Because of the involvement of tunneling and thermionic emission in the carrier transport at the interfaces, the dependence of the on-current of CNT transistors on the Schottky barrier becomes very strong. Figure 4 shows experimental CNT-FET on-current and Schottky barrier height as a function of the CNT diameter for three different metal electrodes, Pd, Ti, and Al [6]. When the metal work functions are in the valence or conduction band of the semiconducting CNTs, ohmic contacts will likely be formed at the interface. Experimental measurements have shown that certain metals with high work functions, such as Pd, can produce nearly ohmic contacts with semiconducting CNTs. Ohmic contacts are more desirable in devices where contact resistance needs to be minimized. In addition to the metal work function, other factors, such as the contact geometry and the chemical bonding at the interface also play important roles in the transport properties of CNT-metal contacts. Carbon Nanotube-Metal Interface C 391 Carbon Nanotube-Metal Contact, Fig. 3 A schematic illustration of the energy levels for an ohmic and a Schottky contact between a metal and a semiconducting CNT Ec EF Ec Ev Ev metal metal Ohmic Schottky 10−5 Pd contacts −6 10 0.0 Ti contacts AI contacts 0.2 −7 0.4 10−5 10−9 Ion (A) Carbon Nanotube-Metal Contact, Fig. 4 Experimental CNTFET on-current (left axis) and computed Schottky barrier height (right axis) as a function of the CNT diameter for three different metal electrodes, Pd, Ti, and Al 10−8 10−10 10−11 10 −6 10 −7 10 −8 1 2 + 0.6 3 Schottky Barrier (eV) Ion (A) 10 0.8 0.5 1.0 1.5 d (nm) 2.0 2.5 1.0 −12 10 0.6 Cross-References ▶ Carbon Nanotubes for Chip Interconnections ▶ Carbon-Nanotubes ▶ CMOS-CNT Integration References 1. Banhart, F.: Interactions between metals and carbon nanotubes: at the interface between old and new materials. Nanoscale 1, 201–213 (2009) 2. Anantram, M.P., Léonard, F.: Physics of carbon nanotube electronic devices. Rep. Prog. Phys. 69, 507–561 (2006) 3. Ciraci, S., Dag, S., Yildirim, T., G€ ulseren, O., Senger, R.T.: Functionalized carbon nanotubes and device applications. J. Phys. Condens. Matter 16, R901–R960 (2004) 0.8 1.0 1.2 1.4 Diameter (nm) 4. Zhang, Y., Franklin, N.W., Chen, R.J., Dai, H.J.: Metal coating on suspended carbon nanotubes and its implication to metal-tube interaction. Chem. Phys. Lett. 331, 35–41 (2000) 5. Léonard, F., Tersoff, J.: Role of fermi-level pinning in nanotube schottky diodes. Phys. Rev. Lett. 84, 4693–4696 (2000) 6. Chen, Z.H., Appenzeller, J., Knoch, J., Lin, Y.-M., Avouris, P.: The role of metal  nanotube contact in the performance of carbon nanotube field-effect transistors. Nano Lett. 5, 1497–1502 (2005) 7. Tung, R.T.: Recent advances in Schottky barrier concepts. Mater. Sci. Eng. R 35, 1–138 (2001) Carbon Nanotube-Metal Interface ▶ Carbon Nanotube-Metal Contact C C 392 Carbon Nanotubes 10 Carbon Nanotubes CPU Transistor Count 109 2x every 2 years 1 107 Microns ▶ Ecotoxicology of Carbon Nanotubes Toward Amphibian Larvae ▶ Computational Study of Nanomaterials: From LargeScale Atomistic Simulations to Mesoscopic Modeling 0.1 Feature Size 0.7x every 2 years Carbon Nanotubes (CNTs) 0.01 1970 1980 1990 2000 65 nm 45 nm 32 nm 2010 105 103 2020 ▶ Chemical Vapor Deposition (CVD) ▶ Physical Vapor Deposition Carbon Nanotubes for Chip Interconnections, Fig. 1 Moore’s Law: Transistor count has doubled while feature size has decreased by 0.7X every 2 years (Figure reprinted with permission from Kuhn [1]) Carbon Nanotubes for Chip Interconnections Motivation Gilbert Daniel Nessim Chemistry department, Bar-Ilan Institute of Nanotechnology and Advanced Materials (BINA), Bar-Ilan University, Ramat Gan, Israel Synonyms Carbon nanotubes for interconnects in integrated circuits; Carbon nanotubes for interconnects in microprocessors Definition Chip interconnections electrically connect various devices in a microprocessor. Today’s established technology for interconnects is based on copper. However, it may be technically challenging to extend copper use to future interconnects in microprocessors with smaller lithographic dimensions due to materials properties limitations. Carbon nanotubes are currently investigated as a potential replacement for future integrated circuits (microprocessors). Although carbon nanotubes are a clear winner against copper in terms of materials properties, multiple fabrication challenges need to be overcome for carbon nanotubes to enter the semiconductor fab and replace copper for chip interconnections. Following over 40 years of successful fulfillment of Moore’s law, stating that the number of transistors in a chip doubles every 2 years, we have already moved from microelectronics to nanoelectronics [1]. Although the “end of scaling” has been predicted many times in the past, enormous technical challenges, especially quantum mechanical issues and billiondollar lithography investments, are a serious threat to further miniaturization (Fig. 1). Today’s latest processors are manufactured using the 32-nm technology. To move toward the 22-nm node and beyond, issues such as lithographic limitations, leaking currents in ultra-thin dielectrics (only a few monolayers thick), insufficient power and thermal dissipation, and interconnect reliability must be resolved [1]. At the transistor level, the performance is negatively affected by increased off-state currents due to short channel effects, increased gate leakage due to tunneling through nanometer-thin dielectric layers, and increased overall gate capacitance due to decreasing gate pitch. Although quantum mechanical tunneling and leakage currents may eventually stop further scaling, efficient heat removal from a chip is currently the biggest obstacle. In this respect, the many kilometers of copper interconnects present in today’s chips are the main culprit for heat generation. For instance, in 2004, Magen et al. [2] showed that for a microprocessor fabricated with the 0.13-mm-node technology consisting of 77 million transistors, interconnects consumed more than 50% of the total dynamic power. Carbon Nanotubes for Chip Interconnections Given the increased length of interconnects, their reduced cross section, and the increased current densities circulating into the interconnects of our latest chips, the problem has been further exacerbated. Additionally, copper interconnects are a major contributor to the total resistance-capacitance (RC) delay of the chip, can fail by electromigration, and need a liner to avoid diffusion into the silicon. Bottom line, the interconnect issue is so serious that the International Semiconductor Roadmap [3] (ITRS, an expert team assessing the semiconductor industry’s future technology requirements for the next 15 years) indicates copper interconnects as a possible dealbreaker to further miniaturization for IC nodes beyond 22 nm. Many technology options are currently under investigation to replace copper for interconnects. Among them, we can mention other metals (mainly silicides), wireless, plasmonics, and optical interconnects. Most notably, there has been an intense research effort on new nanotechnology materials such as carbon nanotubes, which, at the theoretical level, could solve all the above technical issues suffered by copper. The plan of this section is to first introduce the reader to copper interconnects’ fabrication and limitations. Next, we will compare copper to carbon nanotubes (CNTs) and detail possible models for implementation. An important paragraph will focus on the state-of-the-art of CNT fabrication, prior to concluding on the outstanding issues and outlook for future CNT-based chip interconnections. Background on Copper Interconnects and Dual-Damascene Process In 1997, IBM introduced the revolutionary “dualdamascene” process to fabricate copper interconnects and to replace aluminum interconnects, the industry standard at the time. Compared to aluminum, copper presents two major advantages: (1) 50% lower resistivity (Cu  1.75 mm cm versus Al  3.3 mm cm) and (2) higher current densities before failure by electromigration (up to 5  106 A/cm2) [4, 5]. Although, as a material, copper was a clear winner against aluminum, fabrication challenges delayed its introduction. Historically, we may be at a similar juncture with carbon nanotubes compared to copper as we were with copper compared to aluminum in 1997: in spite of their superior materials properties, 393 C mainly fabrication issues are now preventing the introduction of nanotubes in the semiconductor industry to replace copper interconnects. Copper diffuses into silicon, generating mid-gap states that significantly lower the minority carrier lifetime and which lead to leakage in diodes and bipolar transistors. Copper also diffuses through SiO2 and low-k dielectrics, and therefore requires complete encapsulation in diffusion barriers. Since no dry etches were known for copper, IBM’s bold innovation of polishing using chemical mechanical polishing (CMP), after electroplating the copper, was significantly at odds with the technological processes at that time in semiconductor fabrication. The copper dual-damascene process consists of the following steps: • Develop a pattern for wires or vias by patterned etching of the dielectric. • Deposit a barrier layer (usually Ta) to prevent copper diffusion into silicon. • Deposit a copper seed layer. • Fill the vias with copper using electrodeposition. • Remove excess copper using CMP. • Repeat the process to lay the alternating layers of wires and vias which will form the complete wiring system of the chip (Fig. 2). Typical microprocessor design follows a “reverse scaling” metallization scheme with multiple layers of interconnects labeled as local, intermediate, and global interconnects, with increasing width. Very thin local interconnects locally connect gates and transistors within a functional block and are usually found in the lower two metal layers. The wider and taller intermediate interconnects have lower resistance and provide clock and signal within a functional block up to 4 mm. Global interconnects are found at the top metal layers and provide power to all functions in addition to connecting functional blocks through clock and signal. They are usually longer than 4 mm (up to half of the chip perimeter) and exhibit very low resistance to minimize RC delay and voltage drop. Below are a typical cross section of an I.C. chip and a possible implementation using CNTs (Figs. 3 and 4). Limitations of Copper Interconnections Copper interconnects have efficiently scaled down to the current 32-nm-node microprocessors, although this C C 394 Carbon Nanotubes for Chip Interconnections, Fig. 2 Dual-damascene process of copper filling an interconnect via (Figure reprinted with permission from Jackson et al. [21]) Carbon Nanotubes for Chip Interconnections SiN 6. Treach and via etch 1. SiN deposition Metal 1 Contacts Devices 2. Oxide dielectric deposition 7. Barrier/ seed deposition 8. Copper till 3. Via patterning 4. Partial via etch 9. Copper CMP and SiN cap layer 5. Treach Patterning Passivation Dielectric Etch Stop Layer Wire Via Dielectric Capping Layer Copper Conductor with Barrier/Nucleation Layer Global Carbon Nanotubes for Chip Interconnections, Fig. 3 Typical cross sections of hierarchical scaling in current microprocessor (Figure reprinted with permission from the Semiconductor Industry Association [22]) Intermediate Metal 1 has required many technological advances to allow ever-shrinking copper cross sections to carry increasing currents without failure. However, we may be very close to smashing against a technical wall because of materials failure and related fabrication issues. Alternative materials or technologies would require many changes in semiconductor fabrication and Pre-Metal Dielectric Tungsten Contact Plug Metal 1 Pitch massive investments; thus the large semiconductor companies are doing the impossible to extend copper application to future nodes. It is clear that only when up against an insurmountable technical wall will the semiconductor industry switch to a new technology. Electrical resistance is a major issue now that copper interconnect cross sections are comparable to Carbon Nanotubes for Chip Interconnections 395 C Carbon Nanotubes for Chip Interconnections, Table 1 Selected critical parameters for copper use as interconnects in future IC nodes (Data from the Semiconductor Industry Association [22]) Year of production (Estimated) MPU/ASIC metal 1 ½ pitch (nm) (contacted) Total interconnect length (m/cm2) – Metal 1 and 5 intermediate levels, active wiring only Barrier/cladding thickness (for Cu intermediate wiring) (nm) Interconnect RC delay [ps] for a 1-mm Cu intermediate wire, assumes no scattering and an effective r of 2.2 mO-cm Carbon Nanotubes for Chip Interconnections, Fig. 4 Schematic view of possible implementation of carbon nanotube via interconnects in lieu of copper (Figure reprinted with permission from Awano et al. [7]) the mean free path of electrons in copper (
40 nm in Cu at room temperature). Grain boundary and surface scattering are significant contributors to the increased resistance, especially now that we have reached nanoscale dimensions. At the microstructural level, the grain boundaries play an important role, hence, among other fabrication concerns, controlling the copper grain size during electrodeposition has allowed to limit the grain boundary scattering impact thus far. The steep rise in interconnect resistance for smaller IC nodes is a major source of RC delays and directly affects the chip reliability by increasing the risk of electromigration failure, a major issue for further downscaling. Electromigration is the transport of material caused by the gradual movement of the copper ions due to the momentum transfer between conducting electrons and diffusing metal atoms, which occurs for high current densities, which can create voids leading to open circuits. Given that downscaling leads to a reduction of the interconnects’ cross section, the problem is amplified at subsequently smaller nodes. To compound the issue, the need for a resistive diffusion barrier layer, also called a liner (usually Ta), to avoid copper diffusion into silicon, further reduces the available conductive copper cross section, thus increasing the risk of electromigration failure, especially as the operating temperature rises. 2010 2015 2020 45 25 14 2,222 4,000 7,143 3.3 1.9 1.1 1,132 3,128 9,206 In addition to the increased resistance and the electromigration failure risk, many other aspects of the dual-damascene process are becoming potential sources of failure as the node shrinks. Among the many integration concerns, we can mention materials issues such as interface adhesion between the different materials (copper, low-k dielectrics, etc.), liner effectiveness, metal voids, CMP interface defects, etc. Concurrently, there is a long list of process-related issues such as the need for etch/strip/ clean processes (to avoid damage to low-k dielectric materials), atomic layer deposition (ALD) processes to deposit liners, copper plating and CMP techniques, etc. A few interesting numerical estimates taken from the 2009 projections from ITRS, [3] provide the reader with the magnitude of the technical challenge to extend copper interconnect technology (Table 1). As already mentioned, many alternative technologies are currently being investigated for replacement of copper as interconnect material that would require significant chip redesigns and new fabrication technologies. Some examples include optical interconnects, radio frequency (RF) interconnects, plasmonics, and 3-D interconnects (probably still copper). The interested reader can find more details on these alternative technologies in the review paper from Havemann et al. [6] (now a little dated) or in the latest ITRS report on interconnects [3]. The Case for Carbon Nanotube Interconnects An interesting solution, which has been the subject of intense research in recent years, is to replace copper C C 396 Carbon Nanotubes for Chip Interconnections, Fig. 5 Graphical representations of ideal graphene sheet, SWCNT, MWCNT (Figure reprinted with permission from Graham et al. [22]) with carbon nanotubes. If a reliable and repeatable fabrication process consistent with Complementary Metal Oxide Semiconductor (CMOS) technology requirements could be developed, integration into existing chip architectures may not require significant process redesign (Fig. 5). Carbon nanotubes, which can be visualized as rolled sheets of graphene, have been widely investigated as a promising new material for many electrical device applications [5, 7] (e.g., transistor (CNT-FET), interconnects) as they exhibit exceptional electrical, thermal, and mechanical properties [8]. When comparing materials properties, CNTs are a clear winner against copper. Studies show that CNTs are stable for current densities up to 109 A/cm2, two orders of magnitude higher than copper. CNTs can exhibit multichannel ballistic conduction over distances of microns. Because of their higher chemical stability relative to copper, diffusion barriers (liners) are not needed for CNTs, thus allowing a larger conductive cross section compared to copper for the same technology node. Additionally, their mechanical tensile strength (100 times that of steel) and their high thermal conductivity (comparable to diamond) give CNTs an edge compared to copper. Finally, growing CNTs in high aspect ratio vias could allow the design of chips with higher interlayer spacing to reduce overall RC losses and to decrease chip-layer energy dissipation [9]. Before examining possible models of CNT-based interconnect architectures, it is important to clearly understand CNTs’ electrical properties, which represent the most critical material limitation to resolve with respect to copper. The electronic band structures of single-wall CNTs (SWCNTs) and of graphene are very similar. For graphene and metallic SWCNTs, the valence band and the conduction band Carbon Nanotubes for Chip Interconnections touch at specific points in the reciprocal space. For semiconducting SWCNTs, the conduction band and the valence band do not touch. Semiconducting SWCNTs have been extensively studied as channels in transistor devices while metallic SWCNTs have been considered for applications such as IC interconnects and field emission. The resistance of a CNT contacted at both ends is the sum of three resistances [5, 10]: RCNT ¼ RQ þ RL þ RCONTACT where RQ is the quantum resistance, RL is the scattering resistance, and RCONTACT is the contact resistance. We will now discuss these three resistances. An ideal (defect-free) metallic SWCNT electrically contacted at both ends, in the absence of scattering or contact resistance, exhibits a resistance R ¼ 2 RQ  13kO as a SWCNT has two conduction channels. The quantum resistance RQ ¼ 6:5kO is due to the mismatch between the number of conduction channels in the nanotube and the macroscopic metallic contacts. The one-dimensional confinement of electrons, combined with the requirement for energy and momentum conservation, leads to ballistic conduction over distances in the order of a micron. The scattering resistance is due to impurities or nanotube defects that reduce the electron mean free path, and depends on the length l of the SWCNT:   1 l RL ¼ 2 RQ l0 For defect-free SWCNT lengths below a micron, we can neglect the scattering resistance The contact resistance, which results from connecting the SWCNT to a contact (usually metallic), depends strongly on the material in contact with the nanotube, and on the difference between their work functions. The work functions of multiwall CNTs (MWCNTs) and SWCNTs have been estimated to be 4.95 and 5.10 eV, respectively [10]. Palladium has been found to be one of the materials minimizing the contact resistance, better than titanium or platinum contacts (which exhibited nonohmic behavior when in contact with CNTs) [10]. For interconnect applications, most often bundles of SWCNTs are considered. It is important to note that the coupling between adjacent SWCNTs is negligible Carbon Nanotubes for Chip Interconnections Carbon Nanotubes for Chip Interconnections, Fig. 6 Equivalent circuit model of metallic SWCNT used in HSPICE simulations (Figure reprinted with permission from Naeemi and Meindl [14]) RC1 397 dx RQ /2 RQ /2 RC2 (RQ / l0 ) = 216KΩ /μm Rshunt = (RQ / l0 )dx Re = (RQ / le )dx RV lM .dx lk .dx Re since, for defect-free SWCNTs, the electrons would rather travel along the SWCNT axis (ballistic path) than across SWCNTs because of the large inter-CNT tunneling resistance (2–140 MO) [10]. Thus, the resistance of a bundle of SWCNTs can be viewed as a parallel circuit of the resistances of the individual SWCNTs. If we have n SWCNTs, the resistance of the bundle will be: bundle ¼ RSCWNT n The above overview related to SWCNTs. The electrical properties of MWCNTs have not been as extensively studied because of the additional complexities arising from their structure, as every shell has different electronic characteristics and chirality, in addition to interactions between the shells [11]. Geometrically, the interwall distance in a MWCNT is 0.34 nm, the same as the spacing between graphene sheets in graphite. What still has to be clarified is how the conductivity of a MWCNT varies with the number of walls. Initially, it was thought that the conductance of a MWCNT occurred only through the most external wall, which seems to be the case at low bias and temperatures, where electronic transport is dominated by outer-shell conduction. However, theoretical models and experimental results indicate that shell-to-shell interactions can significantly lower the resistance of MWCNTs with many walls [5, 10]. One view is that the conductivity of a MWCNT with n walls is simply n times the conductivity of a SWCNT. Li et al. [12] experimentally measured an electrical resistance of only 34.4O for a large MWCNT with outer diameter of 100 nm and inner diameter of 50 nm (¼ > 74 walls). This value was much lower than RV = (dV (x)/ I0) CQ .dx Rshunt RSCWNT C CE .dx I0 = 25 μA the one that could be calculated assuming all walls participated separately in the electrical conduction (i.e., calculated as the parallel of the resistances for each wall) showing that interwall coupling contributes to additional channels of conductance. Naeemi et al. [13] also assumed intercoupling between CNT walls in their models to increase the channels of conduction with increasing number of walls. However, their conductance was lower compared to that measured experimentally by Li’s team. Bottom line: The conductivity of MWCNTs increases with the number of walls but the exact relationship has not yet been exactly clarified. Models of CNTs as Interconnects Various studies investigated replacing copper interconnects with bundles of CNTs (SWCNTs or MWCNTs) or with one large MWCNT. A first approach consists of using densely packed SWCNTs. Modeling SWCNTs as equivalent electrical circuits and using SPICE simulations, Naeemi et al. [14] showed that a target density of SWCNTs of at least 3.3  1013 CNTs/cm2 was required. Currently achieved maximum densities for CNTs in vias barely reach 1012 CNTs/cm2, which is still an order of magnitude smaller than required. Furthermore, since statistically only one-third of the SWCNTs grown are metallic (the other two-third are semiconducting), the conduction of the bundle will only occur in the metallic SWCNTs (Fig. 6). Naeemi et al. [14] also compared SWCNT bundles to copper as local, intermediate, and global interconnects. They showed that, in SWCNT bundles, resistance and kinetic inductance decreased linearly with the number of nanotubes in the bundle, while magnetic inductance changed very slowly. The resistance of C C 398 le = 1.6 μm scopper-Bulk scopper@22 nm-Node 0.10 VAB = 0 VAB = 0.05 V VAB = 0.2 V VAB = 0.5 V 0.01 0.10 1.00 10.00 100.00 SWNT–Bundle Length, l (μm) Conductivity, σ(μΩ–cm)−1 Conductivity, σ (μΩ–cm)−1 1.00 Carbon Nanotubes for Chip Interconnections 1.00 Cu Wire, W = 100 nm Cu Wire, W = 50 nm Cu Wire, W = 20 nm SWCN, D=1 nm 0.10 l = l0b/a MWCN, D = 10 nm MWCN, D = 20 nm MWCN, D = 50 nm MWCN, D = 100 nm 0.01 0.1 1.0 100.0 10.0 Length, l (μm) 1000.0 Carbon Nanotubes for Chip Interconnections, Fig. 7 Conductivity of densely packed SWCNT bundles versus length for various bias voltages (Figure reproduced with permission from Naeemi and Meindl [14]) Carbon Nanotubes for Chip Interconnections, Fig. 8 Conductivity of MWCNTs with various diameters compared to Cu wires and dense bundles of SWCNTs (Figure reproduced with permission from Naeemi and Meindl [13]) a bundle of SWCNTs with sufficient metallic nanotubes was smaller than the resistance of copper wires, while capacitance was comparable. SWCNT bundles also fared better compared to copper in reducing power dissipation, delay, and crosstalk. For local interconnects, they quantified the improvements as 50% reduction in capacitance, 48% reduction of capacitance coupling between adjacent lines, and 20% reduction in delay. For intermediate interconnects, the improvements were more marked, especially in terms of improved conductivities. For global interconnects, dense SWCNT bundles proved critical to improve bandwidth density (Fig. 7). Using MWCNTs, which are all electrically conductive as they exhibit multiple channels of conduction (compared to only one-third metallic SWCNTs), could lower the resistivity of the bundle, although fewer of them can be packed in the same space because they usually have larger diameters (but also require a lower packing density compared to SWCNTs). In a different modeling study, Naeemi et al. [13] explored the suitability of MWCNTs as replacement for copper interconnects. They concluded that for long lengths (over 100 mm), MWCNTs have conductivities many times that of copper and even of SWCNT bundles. However, for short lengths (less than 10 mm), dense SWCNT bundles can exhibit a conductivity that is twice that of MWCNT bundles. Thus, for via applications, they recommended using dense bundles of SWCNTs or, alternatively, bundles of MWCNTs with small diameter (i.e., with few walls) (Fig. 8). Using an individual MWCNT with large diameter could offer high conductivity due to the participation of multiple walls to significantly increase the channels for conduction. However, as previously mentioned, the exact relationship between the number of walls and the conductance has yet to be clarified. In conclusion, the choice of nanotubes may differ depending on the type of interconnect. For instance, dense bundles of SWCNTs or MWCNTs with few walls may be more suitable for small-section vertical vias, while dense bundles of larger MWCNTs may be more appropriate for long-range interconnects. The option of using a large MWCNT which fills all the space available needs to be further investigated. It is also plausible that hybrid systems of copper/SWCNTs/ MWCNTs may be the best solution; for instance, small-section vertical vias may be replaced by dense SWCNT bundles, while larger long-range horizontal interconnects may still use copper, dense bundles of MWCNTs, or even metal-CNT composites. Practical Implementation: Fabrication State of the Art and Outstanding Issues Reliable and repeatable high-yield CNT fabrication compatible with CMOS standards is the main Carbon Nanotubes for Chip Interconnections Carbon Nanotubes for Chip Interconnections, Fig. 9 Pictorials comparing the “grow-in-place” and “grow-then-place” techniques (Reproduced with permission from Professor Carl V. Thompson [18]) 399 Grow-In -Place C2H2 C Grow-Then-Place C2H2 C C2H2 C2H2 Chemically-directed Field-directed Catalyst: Ni, Co, Fe ... bottleneck in replacing copper in chip interconnections. Although hundreds of research teams have focused their efforts on nanotube growth and thousands of papers detailing growth recipes have been published, surprisingly, very few have focused on the growth on conductive substrates at CMOS-compatible processing temperature [7, 10, 15, 16]. It is still a challenge to reliably and consistently synthesize CNTs on conductive layers at temperatures below 400–450 C, the maximum temperature allowed in CMOS fabrication to avoid disrupting previous diffusion patterns. Furthermore, it is still difficult to precisely control CNT diameter and height, although chemical vapor deposition (CVD) from thin films of controlled thickness [17] or from nanoparticles [16] of controlled size has shown encouraging results. To utilize carbon nanotubes in industrial applications, two main approaches have been considered: “grow-in-place” and “grow-then-place” [18] (Fig. 9). Grow-then-place: This technique consists of first preparing nanotubes and subsequently transferring them to a substrate. Arc discharge and laser ablation are the main techniques used to synthesize free-standing nanotubes. The nanotubes may be subsequently selected (e.g., separating SWCNTs or metallic SWCNTs) and purified prior to use. To transfer them to another substrate, CNTs are usually functionalized in a way that they will attach to pre-patterned areas of the substrate which will attract functionalized CNTs. An interesting technique for interconnect vias is based on using electrophoresis to push CNTs dispersed in a liquid solution into a matrix with pits (e.g., porous alumina matrix). The advantages of this method are that it places no restrictions on the process or temperature used for CNT synthesis and allows to pretreat the CNTs (e.g., select, purify, functionalize). The major drawback is that no successful and repeatable technique to transfer the CNTs to the substrate has been developed to date. The challenge of resolving this issue appears too high to make this technique a candidate for the CMOS industry. However, free-standing, purified, CNTs are manufactured by many companies and sold for other applications (e.g., CNT-polymer composites). Grow-in-place: this technique usually consists of preparing the sample with a catalyst present in the locations where the nanotubes will be synthesized. For instance, a thin catalyst film can be deposited using e-beam evaporation or sputtering; alternatively, nanoparticles can be deposited on a substrate. Synthesis is usually performed using thermal or assisted (e.g., plasma) CVD. This method has several advantages: (1) good control of nanotube position (CNTs will grow where there are catalysts), (2) proven recipes to obtain crystalline CNTs (at least on insulating substrates), (3) proven capabilities to obtain carpets of vertically aligned CNTs, (4) physical contact with the substrate, (5) electrical contact with the substrate, and (6) CVD techniques are commonplace in the CMOS industry. The major drawbacks are that (1) the processing temperature should be below 400–450 C (CMOScompatibility), thus putting serious limits on the synthesis method and (2) the CNTs should be directly synthesized on the substrate of choice, usually C 400 Carbon Nanotubes for Chip Interconnections h c CNTs b a Via hole Ta barrier 1 μm Co catalyst particles SiO2 g TiN cap Cu Si substrate d Spin on Glass (SOG) e f Planarization by CMP Ta barrier Cu Ti contact layer 5 nm Carbon Nanotubes for Chip Interconnections, Fig. 10 Process to synthesize CNTs into pits, SEM cross section, and TEM showing crystalline MWCNT (Reprinted with permission from Yokoyama et al. [23]) a metallic layer to provide electrical contact. Although growing dense carpets of crystalline CNTs on insulating substrates such as alumina or silicon oxide has been achieved by many, CNT growth on metallic layers still remains a serious challenge. Interactions between the catalyst and the metallic substrate (e.g., alloying) are the major impediments for the successful growth of dense carpets of CNTs on metallic layers. In addition to interesting results obtained from university research, good progress on the growth and characterization of carbon nanotubes for interconnects has been achieved by industrial laboratories, initially by Infineon and now by the Fujitsu laboratories. In 2002, Kreupl et al. [19] of Infineon showed that bundles of CNTs could be grown in pits of defined geometry. More recently, Awano et al. [7] of Fujitsu grew bundles of MWCNTs in a 160 nm via at 450 C and measured an electrical resistance of 34O (the CNT density observed was 3  1011 CNTs/cm2). This follows a previous result obtained earlier by the same team where they grew MWCNTs into a 2 mm via at temperatures close to 400 C with the lowest resistance measured of 0.6O after CMP and annealing in a hydrogen atmosphere [5] (Fig. 10). Kreupl et al. [19] succeeded in growing a single MWCNT into a 25-nm hole and measured a high resistance of 20–30 kO. Given the difficulty of growing a single large MWCNT and the difficult task of making a precise electrical measurement, there may be room for further improvement if we could grow an individual, crystalline (defect-free) MWCNT with the maximum number of walls for a given external diameter, thus maximizing the number of channels of conduction. Most of the effort on CNT synthesis to replace copper interconnects has focused on vertical growth of dense carpets of CNTs, which can be achieved by a high density of active catalyst dots. In contrast there have been fewer successful reports of horizontal growth, with less spectacular results. Many techniques have been used to achieve horizontal alignment among which we can mention high gas flow rates, electric fields, and epitaxial techniques to guide horizontal alignment of the nanotubes [10]. In 2010, Yan et al. [20] obtained an interesting horizontal growth of bundled CNTs with a density of 5  1010 CNTs/cm2, which is approaching what has been achieved for vertical CNT growth (although still over an order of magnitude lower compared to the best result for vertical CNT growth) (Fig. 11). Carbon Nanotubes for Chip Interconnections Carbon Nanotubes for Chip Interconnections, Fig. 11 Scanning electron microscope images of dense carpets of horizontally aligned CNTs grown using CVD (Figure reproduced from Yan et al. [20]) 401 a b C 200 nm Although the experimental results obtained are encouraging, there are still numerous challenges that need to be resolved for CNTs to enter the semiconductor fab: 1. Increase the CNT areal density by one or two orders of magnitude. For SWCNTs, assuming all of them are metallic, a packing density of 1013–1014 CNTs/cm2 is required to compete with copper in terms of resistance, while for MWCNTs, the required packing density is lower and depends on the number of channels of conduction (i.e., number of walls). This will require, among other considerations, adequate catalyst and underlayer materials choice and deposition, possible surface pretreatment (e.g., plasma, reduction, and etching), maximum nucleation of active catalyst dots, and optimizing the CNT growth process. Although CNT areal density is an important issue, it may not be the dealbreaker. 2. Minimize the contact resistance between the CNTs and the substrate. To achieve maximum conductivity, the choice of the appropriate underlayer is critical; specific metals (e.g., Pd) and possibly silicides are good candidates. For MWCNTs, it is also important to ensure electrical contact with all the walls. 3. When using SWCNTs, synthesize only metallic SWCNTs (on average one-third of the SWCNTs grown) which are the ones participating in the electrical conduction. This is closely linked to the issue of chirality control, for which no solution has been proposed to date. Selective catalyst choice may provide an alternative avenue to synthesize a higher fraction of metallic SWCNTs. 4. Control growth direction of CNTs. This is especially challenging for horizontal interconnects where the directionality and the packing density achieved are still lagging compared to vertical C 2 μm growth of CNTs, despite some interesting progress in this area [10, 20]. 5. Synthesize crystalline, defect-free CNTs to ensure maximum electrical conductivity in the nanotube. This is challenging, especially when combined with the requirement of growing CNTs at low temperature to achieve CMOS compatibility. 6. Synthesize CNTs at temperatures below 400– 450 C to ensure CMOS compatibility. 7. Repeatably yield the same CNT structures when the same process conditions are applied. This is a major issue since important variations in the structure and shape of the CNTs grown have been experimentally observed. Conclusions and Outlook for CNTs as Chip Interconnections In the past decade, the synthesis of CNTs and the understanding of their growth mechanisms have massively improved. However, for CNTs to enter the CMOS fab and replace copper, significant challenges still need to be resolved. In my opinion, the most significant challenge to overcome is developing a reliable and repeatable fabrication process consistent with CMOS conditions. To achieve that tall order, we need to improve our understanding of the CNT growth mechanisms. Although many simulation models and many experimentally-based insights have been achieved [10], there are still many questions related to CNT growth mechanisms that have not been fully answered. For instance: – Which precursor gases favor CNT growth and which gases hinder CNT growth? What is the role of the gases in the resulting level of crystallinity of C 402 the CNTs grown? Could we pretreat the gases to improve the CNT yield or structure? – What is the exact role of the catalyst? How does its materials properties and its lattice structure influence the resulting CNTs grown (in shape and structure)? – What is the role of the underlayer (layer below the catalyst) and its interactions with the catalyst? Why is it so challenging to grow CNTs on metallic layers? In addition to improving our mechanistic understanding of CNT growth, I believe that a parallel effort focused on developing better reactors is needed. Most researchers use standard CVD-based systems that were designed for a general purpose. A customized reactor, where the same growth conditions can be repeatably achieved with very small variations could provide the repeatability in results (CNT structures) that has eluded us so far. Carbon nanotubes are already becoming a manufacturing reality in mechanical engineering applications (e.g., CNT-based composites) and many interesting results have been obtained to develop novel CNT structures for electrical applications. Although the jury is still out, if process repeatability could be achieved, we could hope, not only that CNTs will enter the CMOS fabs and replace copper for chip interconnections, but also that they will lead to innovative ventures requiring lower investments to develop integrated circuits with radically new architectural designs using carbon nanotubes as new building blocks. Cross-References ▶ Carbon Nanotube-Metal Contact ▶ Carbon Nanotubes ▶ Chemical Vapor Deposition (CVD) ▶ CMOS-CNT Integration ▶ Nanotechnology ▶ Physical Vapor Deposition ▶ Synthesis of Carbon Nanotubes References 1. Kuhn, K.J.: Moore’s Law Past 32 nm: Future Challenges in Device Scaling. Intel Publication, Hillsboro (2009) 2. Magen, N., Kolodny, A., Weiser, U.: Interconnect-power dissipation in a microprocessor. In: Proceedings of the 2004 International Workshop, Paris, 1 Jan 2004 Carbon Nanotubes for Chip Interconnections 3. ITRS. International Technology Roadmap for Semiconductors - Interconnect 2009, International Sematech, Austin 4. Goel, A.K.: High-Speed VLSI Interconnections, 2nd edn. Wiley/IEEE, Hoboken (2007) 5. Nessim, G.D.: Carbon Nanotube Synthesis for Integrated Circuit Interconnects. Massachusetts Institute of Technology, Cambridge, MA (2009) 6. Havemann, R.H., Hutchby, J.A.: High-performance interconnects: an integration overview. Proc. IEEE 89(5), 586–601 (2001) 7. Awano, Y., Sato, S., Nihei, M., Sakai, T., Ohno, Y., Mizutani, T.: Carbon nanotubes for VLSI: interconnect and transistor applications. Proc. IEEE 98(12), 2015–2031 (2010) 8. Dresselhaus, M.S., Dresselhaus, G., Avouris, P. (eds.): Carbon Nanotubes: Synthesis, Structure, Properties, and Applications. Springer, Berlin (2001) 9. Chen, F., Joshi, A., Stojanović, V., Chandrakasan, A.: Scaling and evaluation of carbon nanotube interconnects for VLSI applications. In: Nanonets Symposium 07, Catania, 24–26 Sept 2007 10. Nessim, G.D.: Properties, synthesis, and growth mechanisms of carbon nanotubes with special focus on thermal chemical vapor deposition. Nanoscale 2(8), 1306–1323 (2010) 11. Collins, P.G., Avouris, P.: Multishell conduction in multiwalled carbon nanotubes. Appl. Phys 74(3), 329–332 (2002) 12. Li, H.J., Lu, W.G., Li, J.J., Bai, X.D., Gu, C.Z.: Multichannel ballistic transport in multiwall carbon nanotubes. Phys. Rev. Lett. 95(8), 086601 (2005) 13. Naeemi, A., Meindl, J.D.: Compact physical models for multiwall carbon-nanotube interconnects. IEEE Electr. Device Lett. 27(5), 338–340 (2006) 14. Naeemi, A., Meindl, J.D.: Design and performance modeling for single-walled carbon nanotubes as local, semiglobal, and global interconnects in gigascale integrated systems. IEEE T Electron Dev 54(1), 26–37 (2007) 15. Nessim, G.D., Seita, M., O’Brien, K.P., Hart, A.J., Bonaparte, R.K., Mitchell, R.R., Thompson, C.V.: Low temperature synthesis of vertically aligned carbon nanotubes with ohmic contact to metallic substrates enabled by thermal decomposition of the carbon feedstock. Nano Lett. 9(10), 3398–3405 (2009) 16. Awano, Y., Sato, S., Kondo, D., Ohfuti, M., Kawabata, A., Nihei, M., Yokoyama, N.: Carbon nanotube via interconnect technologies: size-classified catalyst nanoparticles and lowresistance ohmic contact formation. Phys. Status Solidi 203(14), 3611–3616 (2006) 17. Nessim, G.D., Hart, A.J., Kim, J.S., Acquaviva, D., Oh, J.H., Morgan, C.D., Seita, M., Leib, J.S., Thompson, C.V.: Tuning of vertically-aligned carbon nanotube diameter and areal density through catalyst pre-treatment. Nano Lett. 8(11), 3587–3593 (2008) 18. Thompson, C.V.: Carbon nanotubes as interconnects: emerging technology and potential reliability issues. In: 46th International Reliability Symposium; 2008: IEEE CFP08RPS-PRT, p. 368, 2008 19. Kreupl, F., Graham, A.P., Duesberg, G.S., Steinhogl, W., Liebau, M., Unger, E., Honlein, W.: Carbon nanotubes in interconnect applications. Microelectron. Eng. 64(1–4), 399–408 (2002) Cell Morphology 20. Yan, F., Zhang, C., Cott, D., Zhong, G., Robertson, J.: Highdensity growth of horizontally aligned carbon nanotubes for interconnects. Phys Status Solidi. 247(11–12), 2669–2672 (2010) 21. Jackson, R.L., Broadbent, E., Cacouris, T., Harrus, A., Biberger, M., Patton, E., Walsh, T.: Processing and integration of copper interconnects. In: Solid State Technology. Novellus Systems, San Jose (1998) 22. Graham, A.P., Duesberg, G.S., Hoenlein, W., Kreupl, F., Liebau, M., Martin, R., Rajasekharan, B., Pamler, W., Seidel, R., Steinhoegl, W., Unger, E.: How do carbon nanotubes fit into the semiconductor roadmap? Appl. Phys. 80, 1141–1151 (2005). Copyright 2005, Springer Berlin/ Heidelberg 23. Yokoyama, D., Iwasaki, T., Yoshida, T., Kawarada, H., Sato, S., Hyakushima, T., Nihei, M., Awano, Y.: Low temperature grown carbon nanotube interconnects using inner shells by chemical mechanical polishing. Appl. Phys. Lett. 91, 263101 (2007). Copyright 2007, American Institute of Physics 403 C Catalytic Bimetallic Nanorods ▶ Molecular Modeling on Artificial Molecular Motors C Catalytic Chemical Vapor Deposition (CCVD) ▶ Chemical Vapor Deposition (CVD) Catalytic Janus Particle ▶ Molecular Modeling on Artificial Molecular Motors Carbon Nanotubes for Interconnects in Integrated Circuits Cathodic Arc Deposition ▶ Carbon Nanotubes for Chip Interconnections ▶ Physical Vapor Deposition Carbon Nanotubes for Interconnects in Microprocessors Cavity Optomechanics ▶ Optomechanical Resonators ▶ Carbon Nanotubes for Chip Interconnections Cell Adhesion Carbon Nanowalls ▶ Chemical Vapor Deposition (CVD) ▶ Bioadhesion ▶ BioPatterning Carbon-Nanotubes Cell Manipulation Platform ▶ Robot-Based Automation on the Nanoscale ▶ Biological Breadboard Platform for Studies of Cellular Dynamics Catalyst Cell Morphology ▶ Chemical Vapor Deposition (CVD) ▶ Physical Vapor Deposition ▶ BioPatterning C 404 Cell Patterning toxicity and cell death. They are associated to the uptake of nanoparticles, their persistence at cellular level, their ability to release free radicals, and to induce an oxidative stress. The resulting activation of molecular pathways and transcription factors could lead to a pro-inflammatory response or, depending on the level of free radicals, apoptosis. Cell Patterning ▶ BioPatterning Cellular and Molecular Toxicity of Nanoparticles Background ▶ Cellular Mechanisms of Nanoparticle’s Toxicity Cellular Electronic Energy Transfer ▶ Micro/Nano Harvesting Transport in Microbial Energy Cellular Imaging ▶ Electrical Impedance Tomography for Single Cell Imaging Cellular Mechanisms of Nanoparticle’s Toxicity Francelyne Marano, Rina Guadagnini, Fernando Rodrigues-Lima, Jean-Marie Dupret, Armelle BaezaSquiban and Sonja Boland Laboratory of Molecular and Cellular Responses to Xenobiotics, University Paris Diderot - Paris 7, Unit of Functional and Adaptive Biology (BFA) CNRS EAC 4413, Paris cedex 13, France Synonyms Cellular and molecular toxicity of nanoparticles Definition The interaction between nanoparticles and cell triggers a cascade of molecular events which could induce The last five years have shown an increasing number of papers on the nanoparticle’s mechanisms of cytotoxicity. What are the reasons? It is likely that the specific useful properties which appear at nanoscale can also lead to adverse effects. This hypothesis is strongly supported by in vivo and in vitro studies which compare the toxicity of NPs with their fine counterparts of the same chemical composition. These results have clearly demonstrated a higher toxicity of particles at nanoscale than at microscale. Moreover, it appears from experimental studies that solid nano-sized particles could be translocated beyond the respiratory tract and could induce a systemic response. The interstitial translocation of a same mass of particles is higher for ultrafine than fine particles after intratracheal instillation in rats [1]. Surface area, which is strongly increased for nanoparticles compared to microparticles of same chemical composition, and surface reactivity are considered as the principal indicators of NP’s reactivity. It was shown that a toxic response could be observed even to apparently nontoxic substances when the exposure occurred in the nanometre size range. All these observations have lead to the development of a new field of toxicology, nanotoxicology [2]. However, the toxicological mechanisms which sustained the biological response are not yet clear and a matter of debates. The concerns about the toxicity of engineered nanoparticles, which are increasingly used for industrial and medical applications, came also from the knowledge on the toxicity of non-intentional atmospheric particles. Short-term epidemiological studies in Europe and North America have shown an association between cardio-respiratory morbidity and mortality and an increased concentration of atmospheric fine particles [3]. Moreover, long-term epidemiological studies have also demonstrated an association between exposure to atmospheric particles Cellular Mechanisms of Nanoparticle’s Toxicity (Particulate Matter or PM10 and 2.5) and increased cancer risk [3]. In parallel, in vitro and in vivo studies on fine and ultrafine airborne particles such as Diesel exhaust particles and PM2.5, gave causal explanations to these adverse health effects (reviewed in [1]). They allow to define the molecular events induced by these particles in lung cells. The major event is a pro-inflammatory response which is characterized by various cytokine releases (pro-inflammatory mediators), associated to the activation of transcription factors, and signaling pathways. This was especially demonstrated for diesel exhaust particles (DEP), a major component of urban PM in Europe. These events are mostly induced by organic components of the DEP and are probably mediated by the generation of reactive oxygen species (ROS) during the metabolism of organic compounds (for a review see [4]). These findings were used as a background for the researches on biological mechanisms induced by NPs considering that fine and ultrafine atmospheric particles have great similarities with NPs, especially Diesel exhaust particles which are of nano-size and aggregate after their release in atmosphere. It became rapidly obvious that the understanding of the cellular and molecular mechanisms leading to the biological effects of NPs is essential for the development of safe materials and accurate assays for risk assessment of engineered NPs [2] and several recent reviews were focused on demonstrated or hypothetic cellular mechanisms of these responses [4–6, 11]. The first event, when NPs enter in contact with the human body by inhalation, ingestion, dermal exposure, or intravenous application, is their interaction in the biological fluids and the cellular microenvironment with biological molecules such as proteins thus forming a protein corona [7]. Consequently, NPs do not directly interact with the cell membrane but through the protein and/or lipids of the corona. NPs-bound proteins may recognize and interact with the membraneous receptors or could bind nonspecifically cellular membranes. Whatever these interactions, they seem to play a central role which could determine further biological responses. In particular, these interactions may drive the uptake of NPs by the first target cells at the level of the biological barriers such as immune cells (macrophages, dendritic cells, and neutrophiles) or epithelial and endothelial cells. This uptake seems to be general for many NPs which are able to bind proteins at their surface and the paradigm 405 C of “Trojan horse” was developed to explain this uptake and the further biological responses. One of the first responses is the direct or indirect production of ROS which is associated to the size, the chemical composition and the surface reactivity of the NPs. This common response occurs for a large number of NPs even with different chemical patterns and different abilities to form agglomerates, so that, the paradigm of the central role of oxidative stress was developed [5]. These authors suggested that “although not all materials have electronic configurations or surface properties to allow spontaneous ROS generation, particle interactions with cellular components are capable of generating oxidative stress.” Further activation of nuclear factors and specific genetic programs are associated to the level of ROS production leading to cell death by necrosis and apoptosis or adaptive responses such as pro-inflammatory responses, antioxidant enzyme activation, repair processes, effects on cell cycle control and proliferation. Over the last years, numerous in vitro studies have confirmed this hypothesis leading to the development of assays using the detection of ROS or oxidative stress for the screening of NPs. However, new data during the last year have pointed out other specific effects of NPs which are not related to oxidative stress. For example, NPs can interact with membrane receptors, induce their aggregation and mimic sustained physiological responses through specific signaling pathways in the target cells. This type of mechanism may contribute to the development of diseases but could also be of use to develop therapeutic strategies whereby NPs activate or block specific receptors. Cellular Uptake of Nanoparticles and Their Fate at Cellular Level The uptake of particles by specialized immune cells in humans is a normal process which leads to their removal and contributes to the integrity of the body. However, depending on the level of the uptake, this process could induce an increasing release of inflammatory mediators and disturbance of phagocyte normal functions such as the clearance and the destruction of pathogens. One of the knowledge of the 50 last years on the effects of a sustained exposure to airborne particles, especially at occupational level, is the concept of overloading. If the mechanisms of C C 406 Cellular Mechanisms of Nanoparticle’s Toxicity Interactions Size Shape Diameter Agregation/Agglomeration Specific surface area Interactions with Proteins + + + − − − + − + − −−− +++ Chemical composition Coating Solubilisation Charge Interactions with Receptors Surface chemistry Cellular Mechanisms of Nanoparticle’s Toxicity, Fig. 1 Different physicochemical characteristics of the nanomaterials involved in their biological activity: size, surface area, shape, bulk chemical composition, surface chemistry including solubility as well as surface charge or coatings, interactions between particles leading to agglomeration, and aggregation as well as with proteins leading to “corona” or with receptors in the cell membrane clearance are not sufficient to eliminate the particles and if they are persistent, the particles could accumulate in the tissues, leading to a sustained inflammation and chronic pathologies. This was demonstrated for exposure to quartz, asbestos, coal, mineral dusts but also for long time exposure to heavy PM-polluted atmospheres such as in Mexico City [1]. These questions of uptake and persistence are fundamental for risk assessment evaluation of NPs. This may explain the number of papers published recently that analyze the mechanisms of uptake, the behavior and the translocation of various NPs. So far, it appears that the response depends on several parameters. Taken together, NP’s surface and its specific chemical composition resulting from the engineering processes, the capacity of NPs to form aggregates (particle comprising strongly bonded or fused NPs) or agglomerates (collection of weakly bound NPs), the methods used for dispersion and experimental preparation determine in the NP’s ability to adsorb or not specific biological compounds, such as proteins, to form the “corona” and to interact with biological membranes [7]. The amount and the structural/functional properties of the adsorbed proteins drive the interactions of these nanomaterials with the membranes and their uptake (Fig. 1). Recent studies have clearly identified a number of serum proteins such as albumin, IgG, IgM, IgA, apolipoprotein E, cytokines, or transferrin that bind to carbon black, titanium dioxide, acrylamide or polystyrene NPs [8]. Among the identified proteins, several are ligands for cellular receptors and may contribute to the biological effects of NPs. For example, receptor aggregation induced by NPs could lead to cell signaling: coated gold NPs were able to bind and cross-link IgE-Fc epsilon receptors leading to degranulation and consequent release of chemical mediators [9]. On another hand, integrins such as a5b3 are known to play a key role in cell signaling and their activation by extracellular ligands can modulate biological processes such as matrix remodeling, angiogenesis, tissue differentiation, and cell migration. These receptors were recently demonstrated as important membrane targets for carbon nanoparticles and their activation induced lung epithelial cell proliferation which was due at least in part to b1-integrin activation [6]. As far as, uptake process is concerned, it is likely that different cell types might have different uptake mechanisms, even for the same NPs. The possible pathways of cellular uptake were previously described by several authors (see [10]). It could occur through phagocytosis, macropinocytosis, clathrin-mediated endocytosis, non-clathrin, non-caveolae-mediated endocytosis, caveolae-mediated endocytosis, or diffusion (Fig. 2). These mechanisms have been described for different NPs and may occur for the same NP depending on the cell type, the medium, the level of aggregation. Therefore, uptake processes are considered as very complex and not easy to measure. Dawson et al. 2009 [12] have postulated that the uptake depends mostly on the size: NPs less than 100 nm can enter the cells and less than 40 nm in the nucleus. It was also suggested that the size of the NPs determine caveolin Cellular Mechanisms of Nanoparticle’s Toxicity 407 NPs ? Calcium Integrin Receptor Channel ? NPs EGFR C ? Endocytosis ROS ↑↑↑Calcium Lysosomal damage ROS Mitochondrial Damage Activation of apoptosis Activation of MAPK (Proteases/Cathepsin release) Activation of transcription factors ARE DNA Damage phase II enzymes (GST, NQO-1) antioxidant enzymes (HO-1) Cellular Mechanisms of Nanoparticle’s Toxicity, Fig. 2 A schematic representation of NP’s triggered cellular pathways through membrane receptors, ROS production and implication of oxidative stress in these responses. NPs could induce activation of EGF or integrin receptors can lead to apoptosis, inflammation or proliferation. ROS produced by NPs in immediate cellular environment or inside the cells lead to activation of redox-dependent signaling pathways like MAPK and the activation of transcription factors for example, AP-1, NF-kB, or Nrf2. They migrate to the nucleus and modify gene expression of cytobines, phase 2 enzymes (gluthation S transferase or GST, quinone oxydoréductase 1 or NQO-1), and antioxydant enzymes (hemeoxygenase 1 or HO-1). Oxidative stress could also results in the damage of different organelles like the mitochondria, lysosomes and nucleus resulting apoptosis. Accumulation of high intracellular calcium levels through a direct effect on calcium channel might also act as an alternative mechanism for the induction of these mechanisms (Adapted from [11]) versus clathrin dependent uptakes [13]. However, these oversimplified scenarios are refuted by obvious discrepancies in the recent literature about the optimal size, shape, and mechanisms of internalization of NPs. The surface charge of the NPs could be an important factor for uptake since the negatively charged surface membrane could favor the positively charged NPs for higher internalization. However, negatively charged NPs were also shown to have enhanced uptake as compared to unfunctionalized NPs, perhaps by their possible interactions with proteins. Endocytosis of small NPs is energy-dependent and associated to lipid rafts, dynamin and F-actin mechanisms. Phagocytosis and macropinocytosis are mostly involved in endocytosis of large particles (more than 500 nm) and also in the uptake of the aggregates or agglomerates of NPs which could be promoted by their opsonisation in the biological fluids. Macropinocytosis (which is one kind of pinocytosis) is also an important mechanism for positively charged NPs and TiO2 or carbon black aggregates internalization [14]. The behavior of the NPs after their uptake is another important question but, surprisingly, as far as now, little is known about the intracellular fate of NPs. Most of the transmission electron microscopy (TEM) observations have shown the NPs in cytoplasm vesicles limited by membranes. These vesicles could further be transported in the cytoplasm through the microtubule network. The biopersistence of nanomaterials which are resistant to degradation in the endosomal compartment could be one of the factors of further toxicity and accumulation. However, several metal oxide NPs are toxic after dissolution in the cell. Indeed, the uptake of ZnO NPs into the lysosomal acidic medium accelerates their dissolution and the release of Zn2+ ions in the cytoplasm. Their excess could induce cytokine production and cytotoxicity and the initiation of acute inflammation at the level of the target organ such as the lung. NPs such as TiO2 or carbon black NPs were also observed only in the cytoplasm of cells [14]. Two explanations may be put forward. The first one is that C C 408 NPs could directly enter by diffusion through the lipid bilayer. It has been shown that cationic NPs could pass through cell membranes by generating transient holes without membrane disruption [15]. Another possible explanation could be the release of NPs after rupture of endosomal compartment. It was described that cationic NPs, after binding to lipid groups on the surface cell membrane, could be endocysed in vesicles and accumulated into the lysosomal compartment. Within, they are able to sequestrate protons which could lead to the activation of proton pump and further rupture of the ion homeostasis and lysosomal accumulation of water. The subsequent lysosomal swelling and membrane rupture leads to the cytoplasmic release of NPs [16]. In proliferating cells, these cytoplasmic NPs associated or not to microtubules, could enter in the nucleus during the mitosis, and could explain that non soluble NPs were observed in the nucleus [14]. More rarely, NPs were also observed within the mitochondrial matrix but, so far, no explanation was given to explain this organelle localisation. The Cellular Stress Induced by Nanoparticles and Its Biological Consequences Over the last 10 years of research conducted on the mechanisms of toxicity of non- intentional as well as engineered nanoparticles, has led to the establishment of a consensus within the scientific community of toxicologists to consider the central role of oxidative stress in cellular responses to NPs leading to inflammation or apoptosis [5, 17]. The concept of oxidative stress was developed for many years to explain dysfunctions leading to pathologies. Oxidative stress could occur when reactive oxygen species (ROS) are overproduced leading to an imbalance between ROS production and antioxidant defence capacity. It could also occur when the organism shows a deficiency in antioxidant systems and, especially in antioxidant enzymatic systems (superoxide dismutase, catalase, and glutathione peroxidase). An increased concentration of ROS, exceeding the antioxidant capacity of the cells, can lead to oxidative damage at molecular or cellular level. ROS have important cellular roles either by acting as second messengers for the activation of specific pathways and gene expressions or by causing cell death. In the hierarchical oxidative stress model in Cellular Mechanisms of Nanoparticle’s Toxicity response to NPs, Nel et al. [5] propose that a minor level of oxidative stress leads to the activation of the antioxidant protection whereas, at a higher level, cell membrane and organelles injuries could lead to cell death by apoptosis or necrosis, but specific signaling pathways and gene expression are involved at each step. The induction of oxidative stress by several NPs is due to their ability to produce ROS (TiO2 for example) or to lead to their production. The surface properties of NPs modulate the production of ROS and the smaller they are; the higher is their surface area and their ability to react with biological components and to produce ROS. However, if this cellular induction appears to be general, all the NPs are not able to produce ROS and the cellular increase of the latter could be an indirect effect of the uptake. ROS interact nonspecifically with biological compounds; however, some macromolecules are more sensitive such as the unsaturated lipids, the amino acids with a sulphydril group (SH) and guanine sites in nucleic acids. When lipid bilayer is attacked by ROS, cascade peroxidation occurs leading to the disorganization of the membranes and of their functions (exchange, barriers, and information). The most sensitive proteins contain methionine or cysteine residues, especially in their active site and their oxidation could lead to modify their activity and even to their inactivation. The adaptive cellular responses to NPs are associated to the modulation of different redox-sensitive cellular pathways. Tyrosine kinases and serine/threonine kinase such as mitogen-activated protein kinases or MAP kinases were especially studied (ERK, p38, and JNK) in association with several transcription factors such as NFkB. The free radical can degrade the NFkB inhibitor IkB by the activation of the cascades leading to its proteolysis. The activation of NFkB induces its translocation within the nucleus and its link to consensus sequences in the promoter of numerous genes leading to their transcription. It is also true for other transcriptions factors such as AP1 and NrF2. The latter plays an essential role in the Antioxidant Response Element (ARE) mediated expression of phase 2 enzymes such as NQO1 (NADPH quinone oxidoreductase-1) and antioxidant enzymes such as heme-oxygenase-1 (HO-1). Indeed, HO-1 was found to be activated by CeO2 NPs exposure of human bronchial cells via the p38-Nrf-2 signaling pathway. The ability of NPs to interact with these Cellular Mechanisms of Nanoparticle’s Toxicity signaling pathways could partially explain their cytotoxicity. Recently, TiO2 and SiO2 NPs were demonstrated in vitro and in vivo to induce the release of IL1b and IL1a, two potent mediators of innate immunity, via the activation of inflammasome, a large multiprotein complex containing caspase 1 which cleaves pro IL1a and b in their active forms. These results lead to consider that these NPs could induce a potent inflammatory response. However, the mechanisms leading to this activation are not yet clear. Another important target of ROS produced by NPs is DNA. Oxidative damage of DNA could generate intra-chain adducts and strand breakage. The bond between the base and desoxyribose could also be attacked leading to an abasic site and the attack on the sugar could create a single strand-break. The genotoxicity of NPs begins to be studied and recent reviews pointed out the possible genotoxic mechanisms. However, oxidative stress appears now not sufficient to explain all the biological effects of NPs. The role of epidermal growth factor receptor (EGFR) was investigated by the group of K. Unfried with the demonstration that carbon black NPs induce apoptosis and proliferation via specific signaling pathways both using EGFR [18]. Carbon black NPs could also impair phagosome transport and cause cytoskeletal dysfunctions with a transient increase of intracellular calcium not associated with the induction of ROS since antioxidants did not suppress the response and which could be due to a direct effect on ion channels that control the calcium homeostasis in the cell [19]. Even if all the mechanisms are not completely demonstrated, it appears now that transmembrane receptors are implicated in NP-induced cell signaling and could lead to specific biological responses to NPs. Nanoparticles and Cell Death NPs have also been shown to induce either apoptotic or necrotic cell death in a variety of in vitro systems depending on the concentration and duration of exposure. This induction of cell death mechanism by NPs might act as basis of different pathologies and consequently it is important to understand NPs induced apoptosis pathways. Cells are able to undergo apoptosis through two major pathways, the extrinsic pathway with the activation of death receptors and the intrinsic 409 C pathway with the central role of mitochondria, its permeabilization and the release of Cytochrome C leading to the activation of apoptosome. Recently, the permeabilization of lysosomal membrane was also shown to initiate apoptosis with the release of cathepsins and other hydrolases from the lysosomal lumen. The molecular pathways of apoptosis induction by carbon black and titanium dioxide NPs in human bronchial epithelial cells were recently studied. It was shown that the initial phase of apoptosis induction depends upon the chemical nature of the NPs. Carbon black NPs triggered the mitochondrial pathway, with the decrease of mitochondrial potential, the activation of bax ( a pro-apoptotic protein of the Bcl2 family), and the release of cytochrome C, and the production of ROS is implicated in the downstream mitochondrial events. Whereas TiO2 NPs induced lysosomal pathway with lipid peroxidation, lysosomal membrane destabilization, and catepsin B release [20], Lysosomal permeabilization has also been shown to be important in silica NPs induced apoptosis. These results point out the necessity of a careful characterization of the molecular mechanisms involved by NPs and not just describing the final outcome. Future Directions of Research The interactions between nanomaterials and their biological target are essential to explain their biological effect and the interest of the recent researches on the cellular mechanisms induced by NPs is to take in account the specificity of the cells and of their microenvironment. The first step is the formation of the corona in biological fluids whose composition and affinity kinetics strongly depend on the characteristics of NPs and, especially, their size and surface reactivity. This coating of proteins influences the aggregation, the final size and, finally, the uptake of NPs via the interaction with the membranes, their specific receptors or lipid rafts. It could determine if the nanomaterial is bioavailable and if it induces or not adverse interactions. The central mechanism proposed to explain the biological response is the oxidative stress. However, this paradigm is debated because very similar oxidative stress effects observed in cellular models and induced by different particles could lead in vivo to different pathological effects. It is now obvious that oxidative stress is a common and nonspecific mechanism in toxicology and that the responses at the level of C C 410 the cell depend on the perturbation of the redox balance with a few number of induced signaling pathways. The different biological responses could depend on the tissue specificity which could lead to different diseases observed after occupational or environmental exposure to well known particles or fibers. Recent studies have also shown that NPs could develop a response without a direct contact with the cells but after an induction of secreted factors, it is the “bystander effect.” Small molecules such as purines could be increased at cytoplasmic level in response to NPs, transferred through the gap junctions within a tissue to activate specific receptors [10]. Moreover, NP-induced apoptosis was also demonstrated to be propagated through hydrogen peroxide mediated bystander killing in an in vitro model of human intestinal epithelium. These specific responses could explain the in vivo observed differences. Finally, the interactions of NPs with proteins, enzymes, cytokines, and growth factors, outside, or inside the cell lead to modify the functions of these proteins with a possible indirect pathological effect. The large variety of engineered NPs in the market and under development makes these studies very complex. However, the development of safe nanomaterials depends on better knowledge of these specific interactions Cross-References ▶ Ecotoxicity of Inorganic Nanoparticles: From Unicellular Organisms to Invertebrates ▶ Genotoxicity of Nanoparticles ▶ In Vivo Toxicity of Carbon Nanotubes ▶ In Vivo Toxicity of Titanium Dioxide and Gold Nanoparticles ▶ Quantum-Dot Toxicity ▶ In Vitro and In Vivo Toxicity of Silver Nanoparticles ▶ Toxicology: Plants and Nanoparticles References 1. Donaldson, K., Borm, P. (eds.): Particle Toxicology, p. 434. CRC Press, Boca Raton (2007) 2. Oberdorster, G., Oberdorster, E., Oberdorster, J.: Nanotoxicology: an emerging discipline evolving from studies of ultrafine particles. Environ. Health Perspect. 113, 823–839 (2005) Cellular Mechanisms of Nanoparticle’s Toxicity 3. Brunekreef, B., Holgate, S.T.: Air pollution and health. Lancet 360, 1233–1242 (2002) 4. Marano, F., Boland, S., Baeza-Squiban, A.: Particleassociated organics and proinflammatory signaling. In: Donaldson, K., Borm, P. (eds.) Particle Toxicology, pp. 211–226. CRC Press, Boca Raton (2007) 5. Nel, A., Xia, T., Madler, L., Li, N.: Toxic potential of materials at the nanolevel. Science 311, 622–627 (2006) 6. Unfried, K., Albrecht, C., Klotz, L.O., Mikecz, A.V., Grether-Beck, S., Schins, R.P.F.: Cellular responses to nanoparticles: target structures and mechanisms. Nanotoxicology 1, 52–71 (2007) 7. Nel, A.E., Madler, L., Velegol, D., Xia, T., Hoek, E.M., Somasundaran, P., Klaessig, F., Castranova, V., Thompson, M.: Understanding biophysicochemical interactions at the nano-bio interface. Nat. Mater. 8, 543–557 (2009) 8. Lynch, I., Salvati, A., Dawson, K.A.: Protein-nanoparticle interactions: what does the cell see? Nat. Nanotechnol. 4, 546–547 (2009) 9. Huang, Y.F., Liu, H., Xiong, X., Chen, Y., Tan, W.: Nanoparticle-mediated IgE-receptor aggregation and signaling in RBL mast cells. J. Am. Chem. Soc. 131, 17328–17334 (2009) 10. Bhabra, G., Sood, A., Fisher, B., Cartwright, L., Saunders, M., Evans, W.H., Surprenant, A., Lopez-Castejon, G., Mann, S., Davis, S.A., Hails, L.A., Ingham, E., Verkade, P., Lane, J., Heesom, K., Newson, R., Case, C.P.: Nanoparticles can cause DNA damage across a cellular barrier. Nat. Nanotechnol. 4, 876–883 (2009) 11. Marano, F., Hussain, S., Rodrigues-Lima, F., BaezaSquiban, A., Boland, S.: Nanoparticles: molecular target and cell signaling. Arch. Toxicol. 85, 733–741 (2011). Online May 10 12. Dawson, K.A., Salvati, A., Lynch, I.: Nanotoxicology: nanoparticles reconstruct lipids. Nat. Nanotechnol. 4, 84–85 (2009) 13. Rejman, J., Oberle, V., Zuhorn, I.S., Hoekstra, D.: Sizedependent internalization of particles via the pathways of clathrin- and caveolae-mediated endocytosis. Biochem. J. 377, 159–169 (2004) 14. Hussain, S., Boland, S., Baeza-Squiban, A., Hamel, R., Thomassen, L.C., Martens, J.A., Billon-Galland, M.A., Fleury-Feith, J., Moisan, F., Pairon, J.C., Marano, F.: Oxidative stress and proinflammatory effects of carbon black and titanium dioxide nanoparticles: role of particle surface area and internalized amount. Toxicology 260, 142–149 (2009) 15. Gratton, S.E., Ropp, P.A., Pohlhaus, P.D., Luft, J.C., Madden, V.J., Napier, M.E., Desimone, J.M.: The effect of particle design on cellular internalization pathways. Proc. Natl. Acad. Sci. U.S.A. 105, 11613–11618 (2008) 16. Xia, T., Kovochich, M., Liong, M., Zink, J.I., Nel, A.E.: Cationic polystyrene nanosphere toxicity depends on cell-specific endocytic and mitochondrial injury pathways. ACS Nano 2, 85–96 (2008) 17. Ayres, J.G., Borm, P., Cassee, F.R., Castranova, V., Donaldson, K., Ghio, A., Harrison, R.M., Hider, R., Kelly, F., Kooter, I.M., Marano, F., Maynard, R.L., Mudway, I., Nel, A., Sioutas, C., Smith, S., Baeza-Squiban, A., Cho, A., Duggan, S., Froines, J.: Evaluating the toxicity of airborne particulate matter and nanoparticles by measuring oxidative stress Charge Transport in Self-Assembled Monolayers potential–a workshop report and consensus statement. Inhal. Toxicol. 20, 75–99 (2008) 18. Sydlik, U., Bierhals, K., Soufi, M., Abel, J., Schins, R.P., Unfried, K.: Ultrafine carbon particles induce apoptosis and proliferation in rat lung epithelial cells via specific signaling pathways both using EGF-R. Am. J. Physiol. Lung Cell. Mol. Physiol. 291, L725–L733 (2006) 19. Moller, W., Brown, D.M., Kreyling, W.G., Stone, V.: Ultrafine particles cause cytoskeletal dysfunctions in macrophages: role of intracellular calcium. Part. Fibre Toxicol. 2, 7 (2005) 20. Hussain, S., Thomassen, L.C., Feracatu, I., Borot, M.C., Andreau, K., Fleury, J., Baeza-Squiban, A., Marano, F., Boland, S.: Carbon black and titanium oxide nanoparticles elicit distinct apoptosic pathways in bronchial epithelial cells. Part. Fibre Toxicol. 7(10), 1–17 (2010). Online Apr.16 411 C Charge Transport in Self-Assembled Monolayers Jeong Young Park Graduate School of EEWS (WCU), Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Republic of Korea Synonyms Charge transfer molecules on self-assembled monolayer Definition Cellular Toxicity ▶ Nanoparticle Cytotoxicity Characterization of DNA Molecules by Mechanical Tweezers ▶ DNA Manipulation Based on Nanotweezers Characterizations of Zinc Oxide Nanowires for Nanoelectronic Applications ▶ Fundamental Properties of Zinc Oxide Nanowires Charge Transfer on Self-Assembled Monolayer Molecules ▶ Charge Transport in Self-Assembled Monolayers Charge Transport in Carbon-Based Nanoscaled Materials ▶ Electronic Transport in Carbon Nanomaterials Charge transport in self-assembled monolayers (SAMs) is the transport of an electron or a hole through an organized molecule layer which is bound to a substrate. Overview Charge Transport Through Organic Molecules Significant studies on charge transport properties through organic molecules have been carried out in the general area of molecule-based and moleculecontrolled electronic devices, often termed “molecular electronics” [1, 2]. Self-assembled monolayers (SAMs) are composed of an organized layer of amphiphilic molecules in which one end of the molecule, the “head group,” shows a special affinity for a substrate [3]. SAMs also consist of a tail with a functional group at the terminal end, as seen in Fig. 1. Charge transport of organic molecules is usually limited by hopping processes and is therefore dominated by surface ordering. Self-assembled monolayers are a good model system of molecular electronics due to the ordered surface structure. In order to measure charge transport in a self-assembled monolayer, the substrate surface should be metallic. For example, a gold surface exhibits strong bonds with alkanethiol through S–H bonds. The other electrode should also be metallic for charge transport through the selfassembled monolayer. The measurement scheme of charge transport through a self-assembled monolayer C C 412 Charge Transport in Self-Assembled Monolayers Functional group R R R R S S S S Tail Head group Substrate Charge Transport in Self-Assembled Monolayers, Fig. 1 Schematic of a self-assembled monolayer (SAMs) showing the head group that is bound to the substrate. SAMs consist of a tail with a functional group at the terminal S Metal electrode S S S S S A S S S S S S Substrate Charge Transport in Self-Assembled Monolayers, Fig. 2 Scheme of charge transport mechanisms through selfassembled monolayers. The dominant charge transport mechanism in a molecular junction involves “through-bond” (TB) tunneling, and “through space” (TS) as illustrated in the left and right transport channel, respectively that represents a conductor-molecule-conductor junction is shown in Fig. 2. Charge Transport Mechanism For insulating molecules, such as alkane chains, electron transport occurs via tunneling mechanisms. When such molecules are placed between electrodes, the junction resistance changes exponentially: R ¼ R0 exp(bs), with electrode separation s, where R0 is the contact resistance and b a decay parameter. In most experiments, the separation s is the length of the alkane chain. However, length is not the only important parameter. Conformation and molecular orientation relative to the electrodes are also important. Other factors need to be considered as well, including energy positions of the highest occupied and lowest unoccupied molecular orbitals (HOMO, LUMO), electrode work function, and nature of the bonding to the electrodes. Charge transport mechanisms through selfassembled monolayers consist mainly of three processes [4]. The dominant charge transport mechanism in a molecular junction is “through-bond” (TB) tunneling, where the current follows the bond overlaps along the molecules (as illustrated in the left transport channel of Fig. 2). Another contribution involves the charge transport from electrode to electrode, in which the molecule plays the role of a dielectric medium that is called “through space” (TS), as illustrated in the right transport channel of Fig. 2. The last contribution of charge transport pathway involves a chain-to-chain coupling as illustrated in the middle of Fig. 2. As the molecular chains tilt, the decrease of the electron tunneling distance leads to a lateral hop between the neighboring molecular chains. Two Pathway Models If electron transport was determined purely by tunneling through the alkane chains, one would expect the value of b to equal zero, since the tunneling distance is the same for all tilt angles. The nonzero value of b indicates the existence of either intermolecular charge transfer or variations in the S-Au bonding as a function of tilt that affect the conductivity in an exponential way with angle. Slowinski et al. [4] proposed a two-pathway conductance model involving “through-bond” tunneling, and the “chain-to-chain” coupling. Assuming no effects due to changes in S-Au bonding, the first pathway is independent of tilt, while the second depends on the tilt angle. The tunneling current, thus, is given by It ¼ I0 expðbTB dÞ þ I0 ns exp½bTB ðd  dCC tan YÞ  expðbTS dCC Þ where It is the current at a specific tilt angle Y, d is the length of the molecule, ns is a statistical factor accounting for the number of pathways containing a single lateral hop as compared to those containing only through-bond hops, d is the diameter of the molecule chains, bTB and bTS are respectively through-bond and Charge Transport in Self-Assembled Monolayers 413 C C S S w1 S S w1 S S w1 Substrate S S w1 S S w2 S S w3 Substrate Charge Transport in Self-Assembled Monolayers, Fig. 3 Scheme of the measurement of the junction resistance for two different situations: (1) decreasing of the alkane chain (left part), and (2) the tilting of the alkane chain while maintaining the same number of carbon atoms (right part), which will yield the resistance (1) per unit length of molecule or (2) tilting angle of the molecules, respectively through-space decay constants. For example, in case of C16 alkanethiol molecule chains, dCC ¼ 4.3 Å, d ¼ 24 Å, and ns ¼ 16, i.e., the number of carbon atoms in the molecule. As one example, details on the preparation of an alkanethiol SAM will be described below. Gold substrates (200–300 nm of gold coating over 1–4 nm of chromium layer n glass) are prepared by butane flame annealing in air after cleaning in acetone, chloroform, methanol, and a piranha solution (1:3; H2O2:H2SO4). The resulting surface consisted of large grains with flat terraces of (111) orientation (sizes up to 400 nm) separated by monatomic steps. Flatness and cleanness were tested by the quality of the lattice-resolved images of the gold substrate. Two types of hexadecanethiol (C16) self-assembled monolayer can be formed on Au (111): complete monolayers of the molecules and islands of molecules covering only a fraction of the substrate. In the first case, the film was produced by immersing the substrate in 1 mM ethanolic solution of C16 for about 24 h, followed by rinsing with absolute ethanol and drying in a stream of nitrogen to remove weakly bound molecules. Incomplete monolayers in the form of islands were prepared by immersing the substrate in a 5 mM ethanolic solution of C16 for approximately 60 s, followed by rinsing. Samples consisting of islands facilitate the determination of the thickness of the molecular film relative to the surrounding exposed gold substrate. The molecular order of the islands improves with storage time at ambient conditions. Decay Constant upon Shortening and Tilting of Molecules The junction resistance is dependent on electrode spacing for two different situations: (1) shortening of the alkane chain [5] and maintaining the same width (w) between chains and (2) tilting of the alkane chain but changing w [6, 7]. These measurements will yield the resistance per unit length of molecule or tilting angle of the molecules, respectively. The conductance decay constant b has already been measured using SAMs with different chain lengths when the separation between electrodes decreases as a function of the alkane chain length (the left image of Fig. 3). The decay constant, b, upon tilting of molecules can be measured using deformation with an AFM tip and simultaneous measurement of current (the right image of Fig. 3). This methodology will be described in the next section. Basic Methodology Preparation of Self-Assembled Monolayer The organic molecular films on various types of substrates (conducting, semiconducting, or insulating substrates) have been prepared using techniques such as the Langmuir-Blodgett technique, dipping the substrates into solution with molecules, drop casting, or spin-coating [8]. Techniques to Measure Charge Transport in Self-Assembled Monolayers The current through a thiol SAM on a hanging Hg drop electrode can be measured in an electrochemical C 414 solution. The current was measured as a function of the monolayer thickness that can be tuned by two methods: by changing the number of carbons in the alkane chain and therefore its length; or, expansion of the Hg drop such that the monolayer surface coverage was reduced and the molecules increased their tilt angle with respect to the surface. Slowinski et al. determined the decay constants bTB ¼ 0.91/Å and bTS ¼1.31/Å by both a fit to their experimental data and by independent ab initio calculations. Mercury drop expansion experiments by Slowinsky et al. have shown a dependence of the current through the alkanethiol monolayers on surface concentration, prompting the authors to suggest the existence of additional pathways for charge transfer, like chain-to-chain tunneling. Scanning tunneling microscopy and scanning tunneling spectroscopy have been used to reveal the atomic scale surface structure and charge transport properties of SAM layers [9, 10]. STM has been used to reveal various phases of surface structure and atomic scale defects, which could play a crucial role in the electrical transport. Conductance measurements were performed with a conductive-probe atomic force microscopy (CP-AFM) system. The use of AFM with conducting tips provides the ability to vary the load on the nanocontact and also opens the way for exploring electron transfer as a function of molecular deformation. A junction is fabricated by placing a conducting AFM tip in contact with a metal-supported molecular film, such as a selfassembled monolayer (SAM) on Au, as shown in Fig. 4. The normal force feedback circuit of the AFM controls the mechanical load on the nanocontact while the current–voltage (I–V) characteristics are recorded. The possibility to control the load on the contact is an unusual characteristic of this kind of junction and provides the opportunity to establish a correlation between the mechanical deformation and electronic properties of organic molecules. The normal force exerted by the cantilever was kept constant during AFM imaging, while the current between tip and sample was recorded. It is crucial to carry out the experiment in the low load regime so that there is no damage to the surface. This can be confirmed by inspection of the images with Ångstrom depth sensitivity as well as by the reproducibility of the current and adhesion measurements. If the measured conductance did not change at constant load and did not show timedependent behavior in the elastic regime, the tip Charge Transport in Self-Assembled Monolayers Conductive probe AFM A S S S S S S Substrate Charge Transport in Self-Assembled Monolayers, Fig. 4 Scheme of conductance measurements of SAM with a conductive-probe atomic force microscopy (CP-AFM) system experiences minimal changes during subsequent contact measurements. Key Research Findings The molecular tilt induced by the pressure applied by the tip is one major factor that leads to increased film conductivity. By measuring the current between the conductive AFM tip and SAM as a function of the height of the molecules, the decay parameter (b) can be obtained [11]. Wold et al. studied the junction resistance as a function of load using AFM. The resistance was found to decrease with increasing load within two distinct power law scaling regimes [12]. Song et al. examined the dependence of the tunneling current through Au-alkanethiol-Au junctions on the tip-loading force [13]. It is found that the two-pathway model proposed by Slowinsky et al. can reasonably fit with the results, leading the authors to conclude that the tilt configuration of alkanethiol SAMs enhances the intermolecular charge transfer. Charge Transport in Self-Assembled Monolayers Charge Transport in Self-Assembled Monolayers, Fig. 5 AFM images (200 nm  200 nm) of topography, and current images obtained simultaneously for a full monolayer of C16 on Au (111) surface. Lattice-resolved images of the film (inset in the left figure) reveal a lattice image of SAM (size: 2 nm  2 nm) 415 Topography Current C Topography Charge Transport in Self-Assembled Monolayers, Fig. 6 AFM images (500 nm  500 nm) of topographic, and current images, respectively, that were acquired simultaneously on hexadecylsilane SAM islands on silicon surface C Current Silicon Figure 5 shows topography and current images obtained simultaneously for a full monolayer of C16 on an Au (111) surface. The topographic image reveals the commonly found structure of the gold film substrate, composed of triangular-shaped terraces separated by atomic steps. Lattice-resolved images of the film (inset in the left figure) reveal a (√3  √3)-R30 periodicity of the molecules relative to the gold substrate. Qi et al. measured current–voltage (I–V) characteristics on the C16 alkanethiol sample for loads varying between 20 and 120 nN, and found that the current changes in a stepwise manner and the plateaus are associated with the discrete tilt angle of the molecules. A stepwise response of the SAM film to pressure has been observed previously in other properties such as film height and friction of alkanesilanes on mica and alkanethiols on gold. In order to measure the thickness of the selfassembled monolayer upon molecular deformation, C16 alkylsilane the SAM islands that partially cover the substrate can be used. The heights of the islands can be obtained from topographical AFM images, while charge transport properties of alkanesilane SAMs on silicon surface are measured using AFM with a conducting tip. In this manner, the load applied to the tip-sample contact can be varied while simultaneously measuring electric conductance. Figure 6 shows the topographic and current images, respectively, that were acquired simultaneously on hexadecylsilane islands on a silicon surface. The image size is 500  500 nm. The hexadecylsilane islands are 100–200 nm in diameter and have a height of 1.6 nm at the applied load of 0 nN (or effective total load of 20 nN). It is also clear that the current measured on the alkanesilane island is much smaller than that measured on the silicon surface. These changes were shown to correspond to the molecules adopting specific values of tilt angle relative to the surface, and explained as the result of methylene C 416 Charge Transport in Self-Assembled Monolayers –2 Tilting ln [J (nA/nm2)] –3 –4 d1 d2 –5 –6 –7 –8 10 Linear fit (b = 0.52 Å–1) Two pathway model 15 20 Height of C16 island (Å) Charge Transport in Self-Assembled Monolayers, Fig. 7 Semilog plot of current density (nA/nm2) as a function of the height of the hexadecylsilane SAM islands on a silicon surface. A decay constant (b) ¼ 0.52 0.04 Å1 was found for the current passing through the film as a function of tip-substrate separation groups interlocking with neighboring alkane chains. In the case of complete monolayers of alkanethiol SAM, the junction resistance (R) was measured as a function of the applied load [6]. These data were converted to current versus electrode separation by assigning each step in the current to a specific molecular tilt angle, following the sequence established in previous experiments. It was found that ln(R) increases approximately linearly with tip-surface separation, with an average slope b ¼ 0.57 ( 0.03) Å1. Similar measurement of the decay parameter upon the molecular tilts was carried out with a scanning tunneling microscope and simultaneous sensing of forces. By measuring the current as a function of applied load, a tunneling decay constant b ¼ 0.53 ( 0.02) Å1 was obtained [14]. In the case of hexadecylsilane molecules, the local conductance of hexadecylsilane SAM islands on a silicon surface was measured with conductiveprobe AFM. A semilog plot of current density (nA/nm2) was obtained as a function of the height of the hexadecylsilane SAM islands on a silicon surface, as shown in Fig. 7. A decay constant (b) ¼ 0.52 0.04 Å1 was found for the current passing through the film as a function of tip-substrate separation [7]. Figure 7 shows the best fit of the two-pathway model with the experimental current measurement as a function of the heights of molecule islands by using the fitting parameters of bTB and bTS that are 0.9 and 1.1 Å1, respectively. The good fit indicates that the two-path tunneling model is a valid model to describe this observation. While saturated hydrocarbon chains mainly interact with each other via weak van der Waals forces, much stronger intermolecular p-p interactions can be present in organic films comprised of conjugated/hybrid molecules. This influences charge transport significantly [5, 15]. In a conductance AFM study of two SAM systems, Fang et al. revealed the role of p-p stacking on charge transport and nanotribological properties of SAM consisting of aromatic molecules [16]. The two model molecules chosen in this study are (4-mercaptophenyl) anthrylacetylene (MPAA) and (4-mercaptophenyl)-phenylacetylene (MPPA). In MPPA, the end group is a single benzene ring, while in MPAA it is changed to a three fused benzene ring structure. This structural difference induces different degrees of lattice ordering in these two molecular SAM systems. Lattice resolution is readily achieved in the MPAA SAM, but it is not possible for the MPPA SAM under the same imaging conditions, indicating the MPAA is lacking long-range order. However, it is important to note that even without long-range order, the stronger intermolecular p-p stacking in the MPAA SAM greatly facilitates charge transport, resulting in approximately one order of magnitude higher conductivity than in the MPPA SAM. Future Directions for Research In this contribution, the basic concept of and recent progress on charge transport studies of organic SAM films formed by saturated hydrocarbon molecules and conjugated molecules has been outlined. Several techniques, including AFM, STM, and hanging Hg drop electrode, are used to elucidate the charge transport properties of SAM layers. A number of molecular scale factors such as packing density, lattice ordering, molecular deformation, grain boundaries, annealing induced morphological evolution, and phase separation play important roles in determining charge transport through SAM films. High resolution offered by scanning probe microscopy (SPM) is a key element in identifying and studying microstructures (e.g., molecular tilt, lattice ordering, defects, vacancies, grain boundaries) in organic films and their effects on electronic properties. Other advanced surface characterization techniques, such as SAM with nano-electrodes, in Chemical Etching combination with conductive-probe atomic force microscopy, and spectroscopic techniques such as ultraviolet photoemission spectroscopy (UPS) and inverse photoemission spectroscopy (IPES), could be promising venues to explore the correlation between microstructures and electronic properties of organic films. Cross-References ▶ Atomic Force Microscopy ▶ Conduction Mechanisms in Organic Semiconductors ▶ Electrode–Organic Interface Physics ▶ Scanning Tunneling Microscopy ▶ Self-Assembly References 1. Aviram, A., Ratner, M.A.: Molecular Electronics: Science and Technology. New York Academy of Sciences, New York (1998) 2. Reed, M.A., Zhou, C., Muller, C.J., Burgin, T.P., Tour, J.M.: Conductance of a molecular junction. Science 278, 252–254 (1997) 3. Ulman, A.: An Introduction to Ultrathin Organic Films from Langmuir-Blodgett to Self-Assembly. Academic, Boston (1991) 4. Slowinski, K., Chamberlain, R.V., Miller, C.J., Majda, M.: Through-bond and chain-to-chain coupling. Two pathways in electron tunneling through liquid alkanethiol monolayers on mercury electrodes. J. Am. Chem. Soc. 119, 11910–11919 (1997) 5. Salomon, A., et al.: Comparison of electronic transport measurements on organic molecules. Adv. Mater. 15, 1881–1890 (2003). doi:10.1002/adma.200306091 6. Qi, Y.B., et al.: Mechanical and charge transport properties of alkanethiol self-assembled monolayers on a Au(111) surface: The role of molecular tilt. Langmuir 24, 2219– 2223 (2008). doi:10.1021/la703147q 7. Park, J.Y., Qi, Y.B., Ashby, P.D., Hendriksen, B.L.M., Salmeron, M.: Electrical transport and mechanical properties of alkylsilane self-assembled monolayers on silicon surfaces probed by atomic force microscopy. J. Chem. Phys. 130, 114705 (2009) 8. Barrena, E., Ocal, C., Salmeron, M.: Molecular packing changes of alkanethiols monolayers on Au(111) under applied pressure. J. Chem. Phys. 113, 2413–2418 (2000) 9. Bumm, L.A., Arnold, J.J., Dunbar, T.D., Allara, D.L., Weiss, P.S.: Electron transfer through organic molecules. J. Phys. Chem. B 103, 8122–8127 (1999) 10. Xu, B.Q., Tao, N.J.J.: Measurement of single-molecule resistance by repeated formation of molecular junctions. Science 301, 1221–1223 (2003) 417 C 11. Wang, W.Y., Lee, T., Reed, M.A.: Electron tunnelling in self-assembled monolayers. Rep. Prog. Phys. 68, 523–544 (2005) 12. Wold, D.J., Haag, R., Rampi, M.A., Frisbie, C.D.: Distance dependence of electron tunneling through self-assembled monolayers measured by conducting probe atomic force microscopy: Unsaturated versus saturated molecular junctions. J. Phys. Chem. B 106, 2813–2816 (2002). doi:10.1021/jp013476t 13. Song, H., Lee, H., Lee, T.: Intermolecular chain-to-chain tunneling in metal-alkanethiol-metal junctions. J. Am. Chem. Soc. 129, 3806 (2007) 14. Park, J.Y., Qi, Y.B., Ratera, I., Salmeron, M.: Noncontact to contact tunneling microscopy in self-assembled monolayers of alkylthiols on gold. J. Chem. Phys. 128, 234701 (2008). doi:234701 10.1063/1.2938085 15. Yamamoto, S.I., Ogawa, K.: The electrical conduction of conjugated molecular CAMs studied by a conductive atomic force microscopy. Surf. Sci. 600, 4294–4300 (2006) 16. Fang, L., Park, J.Y., Ma, H., Jen, A.K.Y., Salmeron, M.: Atomic force microscopy study of the mechanical and electrical properties of monolayer films of molecules with aromatic end groups. Langmuir 23, 11522–11525 (2007) Chem-FET ▶ Nanostructure Field Effect Transistor Biosensors Chemical Beam Epitaxial (CBE) ▶ Physical Vapor Deposition Chemical Blankening ▶ Chemical Milling and Photochemical Milling Chemical Dry Etching ▶ Dry Etching Chemical Etching ▶ Wet Etching C C 418 Chemical Milling and Photochemical Milling Agitator Chemical Milling and Photochemical Milling Seajin Oh and Marc Madou Department of Mechanical and Aerospace Engineering & Biomedical Engineering, University of California at Irvine, Irvine, CA, USA Workpiece Synonyms Chemical blankening; Photoetching; Photofabrication; Photomilling Definition Photochemical milling (PCM), also known as photochemical machining, is the process of fabricating high precision metal workpieces using photographically produced masks and etchants to corrosively remove unwanted parts. This process is called wet etching in MEMS fabrication techniques and can be also applied to nonmetal materials. Wet etching, when combined with nanolithography, is a useful process to fabricate detailed nanostructures by extremely controlled removal (Fig. 1). Overview Photochemical machining (PCM) produces threedimensional features by wet chemical etching (Fig. 2). PCM yields burr-free and stress-free metal products and allows for the machining of a wide range of materials which would not be suitable for traditional metal working techniques. PCM is also known as photoetching, photomilling, photofabrication, or chemical blankening [1]. There is a special type of photochemical milling that uses light for initiating or accelerating the wet etching process in metal or semiconductor materials. The combination of photoresists and wet etching enables the fabrication of very detailed structures with complex geometry or large arrays of variable etching profiles in thin (<2 mm) flat metal sheets. Photoresists are made of synthetic polymers having consistent properties. Liquid photoresist coats a thin Tank Maskant Heating Chemical reagent Cooling coils Chemical Milling and Photochemical Milling, Fig. 1 (a) Schematic illustration of photochemical milling process film by dipping or spin casting which enables the production of detailed patterns, but often creates pinholes in the thin layer. Thick dry photoresist films applied by hot lamination have advantages of process simplicity and reliability although the materials are expensive. The process of wet etching is based on the redox chemistry of etchant reduction and metal oxidation, which results in the formation of soluble metalcontaining ions that diffuse away from the reaction metal surface. Many metals commonly used in manufacturing industry are etched readily in aqueous solutions comprising etchant (e.g., ferric chloride). Metal oxides and virtually all materials can be etched with a proper selection of etchant regardless of different etching rates. Wet etching in the PCM process is isotropic where the etchant attacks both downward into the material and sideways under the edge of the resist layer and the ratio of the depth to the undercut is termed the etch factor (Fig. 3). In MEMS device fabrication, the undercut plays a key role in fabricating free-standing microstructure patterns (e.g., beams and cantilevers) that are necessary where microstructures have to be flexible, thermally isolated, small mass, double-sided contacts with gases or liquid surroundings. In most cases, the free-standing material is deposited as a film on a substrate surface, termed a sacrificial material. The etchant, possessing a lateral etching component, must be sufficiently selective not to attach the free-standing Chemical Milling and Photochemical Milling 419 C 3. Photo-tool 1. Artwork, drawing or CAD data Chemicaly clean surface Photographic processing Exposure to UV Developed image Metal preparation Phototool 4. Metal selection Cleaned metal Photoresist coating metal Sensitised metal 2. Coated with photo-resist 5. Photoresist processing Etchant spray Photoresist stencil on metal Etching Removal of metal Photo-resist removed Etched metal Photoresist stripping 6. Finished product Characteristic double cusped edge Chemical Milling and Photochemical Milling, Fig. 2 PCM process in metal sheet etching. 1. Chemically clean the metal surface. 2. Coat both sides of the plate with photoresist that adheres to the metal when exposed to UV light. 3. Expose plate and phototool to ensure image transfer. 4. Develop and Chemical Milling and Photochemical Milling, Fig. 3 Schematic illustration of etch detail through line openings in the patterned photoresist (blue). Etch factor ¼ Depth of etch (D)/ Undercut [½ (BA)] [2] material. Further, novel nanolithography techniques enable to create submicrometer-scaled features. Basic Methodology Photolithography An image of the profile of the flat feature is generated by computer-aided design (CAD) and electronically transferred onto a photographic film to produce a phototool, a photolithographic mask as known in MEMS. The photoresist is exposed through the create photomask image. 5. Spray metal with etchant or dip it in hot acidic solution to etch all material other than part covered with photoresist. 6. Rinse the plate to ensure photoresist and etchant removal [2]. Photoresist A D B phototool from ultraviolet source. There are two types of photoresist – negative and positive. UV lights soften the positive film, and the exposed area is released in the developing solution. The negative photoresist film has reversed pattern developing characteristics. Wet resist is applied to a metal sheet by a dipping process while spin coating is commonly used in micromachining (Fig. 4). Wet Etching In wet chemical etching, the components of the solid are changed into soluble chemical components which C C 420 Chemical Milling and Photochemical Milling Mercury Lamp Chemical Milling and Photochemical Milling, Fig. 4 Positive and negative photoresist [2] Ultraviolet Light Collimating Lens Patterned Glass Photomask Expose Develop Negative are transported by diffusion or convection away from the surface into the bulk of the solution. The solvent molecules form a shell around the dissolved solid particles that are mobile in the liquid phase. The specific interactions of the components of the liquid with the solid determine the reaction rate, which is attributed to a greater etching selectivity of the solid than dry etching methods. Water is used as the solvent in most wet etching processes [3]. 1. In a metal or a semiconductor, the dissolution of metal or semiconductor is accompanied by an electron transfer and obeys the laws of electrochemistry. The oxidation reaction of metal or semiconductor M releases metal ions into solution and produce electrons. M ¼ Mnþ þ ne The etch rate corresponds to the number of metal ions produced at the solid surface per time unit, which is proportional to the interchanged anodic partial current I over the surface A, I/A. Metals and semiconductors often dissolve as complexes (e.g., [MYx]+) where smaller molecules or ions (ligands) form a chemically bound primary shell around the central atom. The etch rate can be changed by varying the concentration of reactants, temperature, viscosity, and convection of the solution. 2. In dielectric materials, acid–base reactions take place if the material to be etched reacts with hydrogen or hydroxyl ions. The cations are solvated by water as a strong polar solvent and they can diffuse rapidly into the bulk of the solution. Etching processes are applicable with oxides and hydroxides of metals and semiconductors at low pH-values, ðanodic partial processÞ A secondary chemical redox process takes place to transfer the liberated electrons to an oxidizing agent OM in the solution. OM þ ne ¼ OMn Positive ðcathodic partial processÞ The anodic and cathodic partial processes results in etching metal. Min solid þ OMin solution ¼ Mnþ þ OMn Mx Oy þ 2yHþ ¼ xM2y=xþ þ yH2 O MðOHÞx þ xHþ ¼ Mxþ þ xH2 O M ¼ metal or semiconductor (e.g., Cu, Si) Similarly, at very high pH-values some metals and semiconductors form stable water dissolvable complexes that are easily solvated by water. Acidic anions, such as chloride and fluoride, or neutral molecules react as ligands, a typical example of which is the etching of SiO2 in HF-containing etchants. Chemical Milling and Photochemical Milling Chemical Milling and Photochemical Milling, Fig. 5 (Left) Schematic illustration of the process generating nanometer-scale lines by controlled undercutting. A pattern is produced in the photoresist by photolithography or soft lithography. An isotropic etch is applied to the substrate beneath the photoresist. Shallow undercutting of the base layer, followed by evaporation into the exposed areas, and lift-off generates
50 200 nm gaps in the thin film at the edges of the photoresist pattern. The pattern is then used as an optical filter. (Right) SEM image of a cross section of linear trenches at the edges of a 100 nm line transferred into a Si <100> substrate. The trenches are 250 nm deep [4] Metal Layer Substrate Photolithography Photoresist C Overetch 500 nm Metal Deposition Metal layer Cr/Si interface Lift-off 50–200 nm ~75 nm 100 nm Test optically Etch SiO2 þ 6HF ¼ H2 SiF6 þ H2 O Salt-like film is also dissolved by complexing agents. For example, copper chloride is etchable in neutral KCl solutions. CuCl þ 2Cl  ¼ ½CuCl3 C 421 2 Key Research Findings for Nanotechnology Edge Lithography Undercutting by isotropic wet etching is applied to transfer the edges of a photoresist pattern into a feature of the final pattern, as illustrated in Fig. 5. The process generates 50 nm scale trenches by controlled undercutting. Currently, patterning features <100 nm are possible by advanced lithography techniques – deep ultraviolet, electron beam writing, extreme ultraviolet, and x-ray photolithography – but are prohibitively expensive. In contrast, edge lithography is a convenient, inexpensive technique for patterning features with nanometer-scale dimensions. Wet Etching for Maskless Patterning In the photochemical etching process, light exposure can increase the charge carrier density in near surface areas which enhances the anodical as well as the cathodical partial processes in wet etching. For example, defect electrons in the lower band left by light exposure support the release of cations and at the same time the released electrons in the conduction band are readily accepted by an oxidizing agent in the solution. In the same process, the intensive exposure with a focused beam enables direct pattern generation without a lithographic mask. This method is specified preferentially for patterning semiconductors deposited on a nonconducting substrate (e.g., GaN on sapphire) [5]. Future Directions for Research Wet chemical etching is a simple, inexpensive, and wellunderstood process. The process plays a key role in the field of MEMS and nanotechnology. Continuous effort is being made to provide controllability, repeatability, and, most importantly, detail in fabricating microstructures. New etchant compositions have been developed to C 422 apply wet chemical etching to the materials constituting new devices. A representative example is selective removal of metal nitride films on sapphire for a new version of a light-emitting diode. At the same time, great efforts have been made to develop new lithography techniques for nanoscale patterns such as proximal probe lithography, very thin to monolayer lithography, and soft lithography. PCM equipped with the emerging lithography techniques can cause the paradigm shift in the creation of nanoscaled features [6]. Cross-References ▶ DUV Photolithography and Materials ▶ Dry Etching ▶ Electron Beam Lithography (EBL) ▶ EUV Lithography ▶ Nanoimprint Lithography ▶ Nanotechnology ▶ Stereolithography ▶ Wet Etching References 1. Abate, K.: Photochemical etching of metals. Met. Finish. 100 (6A), 448–451 (2002) 2. Allen, D.M.: Photochemical machining: from manufacturing’s best kept secret to a $6 billion rapid manufacturing process. CIRP J. Manuf. Syst. 53, 559–572 (2005) 3. Köhler, J.M.: Wet chemical etching method. In: Etching in Microsystem Technology. Wiley, Weinheim (1999) 4. Love, J.C., Paul, K.E., Whitesides, G.M.: Fabrication of nanometer-scale features by controlled isotropic wet chemical etching. Adv. Mater. 13(8), 604–607 (2001) 5. Bardwell, J.A., Webb, J.B., Tang, H., Fraser, J., Moisa, S.: Ultraviolet photoenhanced wet etching of GaN in K2S2O8 solution. J. Appl. Phys. 89(7), 4142–4149 (2001) 6. Madou, M.: Fundamentals of Microfabrication, 2nd edn. CRC Press, Boca Raton (2002) Chemical Modification ▶ Nanostructures for Surface Functionalization and Surface Properties Chemical Solution Deposition ▶ Sol-Gel Method Chemical Modification Chemical Vapor Deposition (CVD) Yoke Khin Yap Department of Physics, Michigan Technological University, Houghton, MI, USA Synonyms Aerosol-assisted chemical vapor deposition (AACVD); Atmospheric pressure chemical vapor deposition (APCVD); Atomic layer chemical vapor deposition (ALCVD); Atomic layer deposition (ALD); Atomic layer epitaxial (ALE); Boron nitride nanotubes (BNNTs); Carbon nanotubes (CNTs); Carbon nanowalls; Catalyst; Catalytic chemical vapor deposition (CCVD); Cold-wall thermal chemical vapor deposition; Dissociated adsorption; Doublewalled carbon nanotubes (DWCNTs); High-pressure carbon monoxide (HiPCO); Hot filament chemical vapor deposition (HFCVD); Hot-wall thermal chemical vapor deposition; Inductively coupled-plasma chemical vapor deposition (ICP-CVD); Low-pressure chemical vapor deposition (LPCVD); Metalorganic chemical vapor deposition (MOCVD); Multiwalled Nanobelts; carbon nanotubes (MWCNTs); Nanocombs; Nanoparticles; Nanotubes; Nanowires; Plasma-enhanced chemical vapor deposition (PECVD); Single-walled carbon nanotubes (SWCNTs); Thermal chemical vapor deposition; Ultrahigh vacuum chemical vapor deposition (UHVCVD); Vertically aligned carbon nanotubes Definition Chemical vapor deposition (CVD) is referred to as deposition process of thin films and nanostructures through chemical reactions of vapor phase precursors. Since CVD can be conducted using high purity precursors, it likely leads to thin film and nanostructures with high purity. The use of vapor phase precursors also enables better control on the composition and doping of thin films and nanostructures. For example, Si thin films can be deposited by decomposition of silane gas (SiH4) by plasma or heat as follows: SiH4 (g) ! Si (s) + 2 H2 (g). Doping of Si films with boron will lead to p-type Si films and can be achieved by the addition of B2H6 gas. Chemical Vapor Deposition (CVD) Classification Classification by Operating Pressures Chemical reactions involved in a CVD technique can be initiated by many ways leading to the classification of various types of CVD approaches. For example, CVD can be classified according to the operating pressures as follows: • Atmospheric pressure CVD (APCVD) is referred to CVD processes that conducted at atmospheric pressure. The advantage of APCVD is simple experimental setup without the need of a vacuum system. The potential drawback will be the undesired contamination. • Low-pressure CVD (LPCVD) is referred to CVD processes at pressures below and close to atmospheric pressure. One of the purposes of reducing the operation pressure is to avoid undesired reactions between precursors. LPCVD can also improve film uniformity. • Ultrahigh vacuum CVD (UHVCVD) is LPCVD processes at a very low pressure, typically below
108 torr. There are CVD processes that operate at high pressures. See details in section Classification by Excitation Techniques. Classification by Excitation Techniques In addition, CVD can be classified by the excitation techniques that initiate the chemical reactions, as follows. • Plasma-enhanced CVD (PECVD) is referred to CVD processes that employed plasmas to initiate the needed chemical reactions. In general, PECVD could reduce the growth temperatures as the chemical reactions in PECVD are not ignited by heat. There are many subclassifications of PECVD that depend on the type of AC potential used for plasma generation. For example, RF-PECVD and MWPECVD employed radio frequency (RF, typically at 13.56 MHz) and microwave (MW, 2.45 GHz) potential to dissociate gases and produce the needed plasmas. Some PECVD are classified by the configuration of plasma generation. For instance, inductively coupled plasma CVD (ICP-CVD) is actually RF-PECVD that uses an induction coil as the RF electrode outside the vacuum chamber and is sometimes called remote plasma CVD [1]. On the other hand, two RF plasmas can be used in a CVD 423 C system. For example, a dual-RF-PECVD approach was demonstrated for the growth of vertically aligned carbon nanofoils/nanowalls and carbon nanotubes (CNTs) [2, 3]. PECVD can also be obtained using DC plasma generated by a DC potential across a pair of electrodes and is simply called DC-PECVD. • Thermal CVD is referred to CVD processes that employed heat to initiate the needed chemical reactions. The most common thermal CVD technique employed external furnace to control the growth temperatures of the entire reaction zone (e.g., vacuum quarts/glass chambers). This is sometimes called hot-wall thermal CVD. The advantages of this approach are including potential of large-scale synthesis. In contrast, there are several approaches to achieve the so-called cold-wall thermal CVD. For example, hot filaments (HF) are used to heat up the temperatures of the adjacent substrates while the needed chemical reactions are ignited with higher temperatures on the filaments. This approach is called hot filament CVD (HFCVD). Other heating approaches can also be used including IR lamps, laser beams, and passing current flows through a suspended Si chip [4]. Classification by the Precursor Type and Feeding Procedure CVD can also be classified by the type of precursors or the procedures where precursors are introduced into the reaction chamber. Some of the examples are described as follows, • Aerosol-assisted CVD (AACVD) involve the use of carrier gases (usually inert gases such as Ar, He, etc.) to transfer vapors of liquid-phase precursor into the reaction chamber in the form of aerosol. This approach enables the use of liquid precursors for the CVD process. The most well-known AACVD is metalorganic CVD (MOCVD) where metalorganic solids are dissolved in organic solvents. For example, Trimethyl-gallium [TMGa, Ga(CH3)3] is often used as the source of Ga to form GaN when it reacts with ammonia (NH3) at high temperatures. Thus, AACVD can be viewed as a subclassification of thermal CVD. • Atomic layer CVD (ALCVD) is better known as atomic layer deposition (ALD) or atomic layer epitaxial (ALE) [5]. ALD allow conformal deposition of monolayer of binary (or ternary) compounds by C C 424 introducing the two (or three) reacting precursors in a pulsed-mode one after another. For example, monolayers of Zinc sulfide (ZnS) can be deposited by first exposing the growth surface with ZnCl2. Once chemisorption of ZnCl2 is completed, the reaction chamber will be purged with an inert gas. Thereafter, hydrogen sulfide (H2S) will be introduced to the chamber so that the following reaction will take place on the growth surface: ZnCl2 (g) + H2S (g) ! ZnS (s) + HCl (g). Since the chemical process is self-limiting, only one monolayer (or less) of ZnS will be formed in each cycle. The advantage of ALD is conformal coating on the full surface of the sample and the precision of film thickness control. The major drawback is the extreme low deposition rate. ALD is usually conducted at low temperatures (200–400 C) and may be viewed as a subclassification of thermal CVD. However, plasma and metalorganic precursors are sometime used in ALD. Examples of CVD Approaches for Nanotechnology Catalytic Chemical Vapor Deposition (CCVD) Catalytic chemical vapor deposition (CCVD) is the simplest and most popular technique for the synthesis of carbon nanotubes (CNTs), graphene, and nanowires (NWs). CCVD is simply thermal CVD approach with the use of catalyst that induces chemical reactions for the formation of nanomaterials. A typical experimental layout for CCVD is shown in Fig. 1. As shown, the CCVD system consists of a tube furnace and a quartz tube chamber where chemical reactions are taking place. Synthesis of Vertically Aligned Carbon Nanotubes by CCVD A typical experimental setup of CCVD for the synthesis of carbon nanotubes (CNTs) is shown in Fig. 1a. In this case, the substrates (usually oxidized Si) are first deposited with catalyst films such as Fe, Ni, or Co by electron beam evaporation, sputtering, pulsed laser deposition, etc. [6–8]. These samples were then annealed at 600 C in Ar, N2 or H2 for about 30 min. During the annealing process, the catalyst films will be converted into nanoparticles that will serve as the growth sites for CNTs. After the annealing, the growth Chemical Vapor Deposition (CVD) a Tube furnace Quartz vacuum chamber Vapors/gas b Substrates Tube furnace Quartz vacuum chamber Vapors/gas Quartz test tube Precursor powders Substrates Chemical Vapor Deposition (CVD), Fig. 1 (a) Typical set up for CCVD and (b) modified double-tube configuration temperatures (
600–800 C) and growth ambient (usually Ar, H2, or their mixtures) will be set prior to the introduction of hydrocarbon source gas (methane, CH4; ethylene C2H4, acetylene, C2H2). These hydrocarbon gases will be decomposed on the surface of the catalyst nanoparticles through the chemical process of dissociative adsorption [6]. For example, dissociated adsorption of C2H2 is summarized in Fig. 2 [6]. Figure 2a shows the adsorption of a C2H2 molecule (step A) on the surface of the Fe nanoparticle. This will lead to either the breaking of C–H bond (step B1) to form C2H and H fragments or the breaking of C¼C bond (step B2) to form two C–H fragments. The catalytic function of the Fe nanoparticle is to reduce the energy required for decomposition by a charge transfer from hydrocarbon molecules to Fe. According to a first principles calculation, the dissociation energy of the first hydrogen atom from an isolated C2H2 (step A to B1) in vacuum can be reduced from 5.58 eV to 0.96 eV. On the other hand, the energy barrier between A and B2 is 1.25 eV. The C–H bond breaking (step B1) is followed by C¼C bond breaking (step C) with a potential barrier of 1.02 eV. Whereas, C¼C bond breaking (step B2) is followed by C–H bond breaking (step C) with an energy barrier of 0.61 eV. Both modes (A to B1 to C or A to B2 to C) are possible and give one C–H fragment, one C and one H. The decomposition of C2H2 is completed after the breaking of the last C–H Chemical Vapor Deposition (CVD) Chemical Vapor Deposition (CVD), Fig. 2 (a) Sequences of dissociative adsorption of C2H2 on Fe surface. See text for detailed description. The (b) decomposed carbon atoms (c) diffused into the solid-core Fe nanoparticle until (d) supersaturation and then (e) segregate as nanotubes 425 Carbon a C Hydrogen B1 Fe Fe A Fe Fe C Fe D B2 e b bond (step D) with the need of a potential energy of 0.61 eV. The decomposed carbon atoms (Fig. 2b) will then diffuse into the subsurface of the Fe nanoparticle (Fig. 2c). At typical growth temperature (650– 800 C), these nanoparticles are not melted even after considering the eutectic point of Fe-C phase. Since dissociative adsorption is an exothermic process, the near-surface temperatures of the catalytic nanoparticles will be higher than the growth temperatures. Therefore it is possible that the near surface region of the particles is melted. This will form the gas–liquid interface between carbon and Fe solid-core nanoparticles. Due to the high diffusion rate of carbon in Fe melt, a Fe-C alloy will start to form. When these nanoparticles become supersaturated with carbon (Fig. 2d) to a value critical for growth at the solid– liquid interface, the excess carbon will segregate as carbon nanotubes (Fig. 2e). A tip-growth mode is illustrated in this figure where the nanoparticles remained at the tips of CNTs. A base-growth mode is also possible where the nanoparticles remained at the bases of CNTs. Thus, the diameters of CNTs depend, to a certain extent, on the diameters of the nanoparticles. By controlling the thickness of the catalyst films and other parameters, CCVD enables the growth of vertically aligned (VA) single-, double-, and multiwalled CNTs [8]. In fact, one of the most impressive achievements in the growth of vertically aligned single-walled CNTs (VA-SWCNTs) is the so-called “super growth” [9]. This water-assisted approach was based on CCVD with the addition of 20–500 ppm of water vapors. c d These water vapors were introduced into the growth chamber by flowing carrier gas (Ar, or He with 40% H2) through a water bubbler. Ethylene was used as the carbon source along with various catalysts [Fe (1 nm), Al (10 nm)/Fe (1 nm), Al2O3 (10 nm)/Fe (1 nm), to Al2O3 (10 nm)/Co (1 nm)]. VA-SWCNTs with the length of several mm and diameter of 1–3 nm were reported. This approach also enabled the growth of double- and multiwalled CNTs [10]. The abovementioned CCVD approaches are achievable at relatively low temperatures and lead to VASWCNTs. Historically, SWCNTs were grown by CCVD at higher temperatures in a powder form. In 1996, Dai et. al, demonstrated the growth of SWCNTs at 1,200 C by using carbon monoxide (CO) as the carbon source gas and supported MoOx powder as the catalyst [11]. In this case, powder form of SWCNTs was grown on the catalyst that was loaded on a quartz boat. Later, the growth temperature was reduced to 1,000 C by using CH4 as the sources gas and results in the growth of randomly distributed SWCNTs on powders of supported metal oxide catalysts [12]. The diameters of these SWCNTs are 1–6 nm. Later, CCVD was modified into a high-pressure mode for large-scale synthesis of SWCNTs [13]. In this approach, liquid iron pentacarbonyl, Fe(CO)5, was used as the catalyst and introduced into the CVD chamber by CO gas. At operational pressures of 1–10 atm. and temperatures of 800–1,200 C, Fe(CO)5 thermally decomposed into iron clusters in gas phase and reacted with CO gas to produce SWCNTs. The yield of SWCNTs was found to increase with temperature and pressure. The average diameter of SWCNTs was decrease from
1.0 nm at C C 426 Chemical Vapor Deposition (CVD) 1 atm. to 0.8 nm at 10 atm. This approach is now known as high-pressure carbon monoxide (HiPCO) technique, one of the major techniques for large-scale synthesis of SWCNTs with small diameters. Finally, alcohol was also used as the carbon source for the synthesis of SWCNTs by CCVD [14]. In this case, ethanol vapors (5 torr) were supplied to the reaction chamber that contained Fe/Co catalyst supported with zeolite at 700–800 C. Synthesis of ZnO Nanostructures by CCVD On the other hand, CCVD can be modified into a double-tube configuration as shown in Fig. 1b. This approach will enable the use of solid precursors to generate the needed growth vapors. For example, various ZnO nanostructures can be grown by such a setup by using ZnO and graphite powders as the precursors [15, 16]. As shown, these powders were loaded on a ceramic combustion boat which is contained at the end of a closed-end quartz tube. The substrates can be placed a distance away from the boat. The following reactions will occur at 1,100 C when oxygen gas (O2) is introduced: ZnOðsÞ þ CðsÞ ! ZnðgÞ þ COðgÞ 2ZnðgÞ þ O2 ðgÞ ! 2ZnOðsÞ Based on this approach, various ZnO nanostructures can be grown with and without the use of catalyst (gold films, Au), including nanotubes [15], nanowires, nanobelts, nanocombs [16], and nanosquids [17]. Synthesis of Boron Nitride Nanotubes by CCVD The double-tube CCVD configuration in Fig. 1b was also used for the growth of boron nitride nanotubes (BNNTs) [18–20]. In this case, boron, magnesium oxide, and iron oxide were used as the precursors and loaded on the ceramic boat. The possible chemical reaction at 1,100–1,200 C is 4B (s) + MgO (s) + FeO (s) ! 2B2Ox (g) + MgOy (s) + FeOz (s), where x, y, and z are yet to be determined. When ammonia gas (NH3) is introduced into the chamber, the generated boron oxide vapors (B2Ox) will react with NH3 to form BNNTs. The partially reduced MgOy and FeOz are the possible catalysts for the formation of BNNTs [18]. It is noted that the synthesis of BNNTs by these chemical processes usually require temperatures above 1,350 C. The key success for the low temperature growth discussed here is due to the so-called “growth vapor trapping” approach obtained by placing the bare substrates directly on the ceramic boat. These substrates trapped the growth vapors from the precursors and enhanced the nucleation rate of BNNTs at low temperatures. In recent experiments, Fe, Ni, or MgO films were coated on the substrates used as the catalysts [19, 20]. Such an approach leads to patterned growth of BNNTs. The possible chemical reactions are: B2 O2 ðgÞ þ MgOðsÞ þ 2NH3 ðgÞ ! 2BNðBNNTsÞ þ MgOðsÞ þ 2H2 OðgÞ þ H2 ðgÞ; or B2 O2 ðgÞ þ NiðsÞ or FeðsÞ þ 2NH3 ðgÞ ! 2BNðs; BNNTsÞ þ NiðsÞ or FeðsÞ þ 2H2 OðgÞ þ H2 ðgÞ Cross-References ▶ Atomic Layer Deposition ▶ Carbon Nanotube-Metal Contact ▶ Carbon Nanotubes for Chip Interconnections ▶ Carbon-Nanotubes ▶ Focused-Ion-Beam Chemical-Vapor-Deposition (FIB-CVD) ▶ Synthesis of Carbon Nanotubes References 1. Tsu, D.V., Lucovsky, G., Dvidson, B.N.: Effects of the nearest neighbors and the alloy matrix on SiH stretching vibrations in the amorphous SiOr:H (0<r<2) alloy system. Phys. Rev. B 40, 1795–1805 (1989) 2. Menda, J., et al.: Structural control of vertically aligned multiwalled carbon nanotubes by radio-frequency plasmas. Appl. Phys. Lett. 87(173106), 3 (2005) 3. Hirao, T., et al.: Formation of vertically aligned carbon nanotubes by dual-RF-plasma chemical vapor deposition. Jpn. J. Appl. Phys. 40, L631–L634 (2001) 4. van Laake, L., Hart, A.J., Slocum, A.H.: Suspended heated silicon platform for rapid thermal control of surface reactions with application to carbon nanotube synthesis. Rev. Sci. Instrum. 78(083901), 9 (2007) 5. Leskel€a, M., Ritala, M.: Atomic layer deposition chemistry: Recent developments and future challenges. Angew. Chem. Int. Ed. 42, 5548–5554 (2003) Chitosan Nanoparticles 6. Kayastha, V.K., et al.: Controlling dissociative adsorption for effective growth of carbon nanotubes. Appl. Phys. Lett. 85, 3265–3267 (2004) 7. Kayastha, V.K., et al.: High-density vertically aligned multiwalled carbon nanotubes with tubular structures. Appl. Phys. Lett. 86(253105), 3 (2005) 8. Kayastha, V.K., et al.: Synthesis of vertically aligned singleand double- walled carbon nanotubes without etching agents. J. Phys. Chem. C 111, 10158–10161 (2007) 9. Hata, K., et al.: Water-assisted highly efficient synthesis of impurity-free single-walled carbon nanotubes. Science 306, 1362–1364 (2004) 10. Yamada, T., et al.: Size-selective growth of double-walled carbon nanotube forests from engineered iron catalysts. Nat. Nanotechnol. 1, 131–136 (2006) 11. Dai, H., et al.: Single-wall nanotubes produced by metalcatalyzed disproportionation of carbon monoxide. Chem. Phys. Lett. 260, 471–475 (1996) 12. Kong, J., Cassell, A.M., Dai, H.: Chemical vapor deposition of methane for single-walled carbon nanotubes. Chem. Phys. Lett. 292, 567–574 (1998) 13. Nikolaev, P., et al.: Gas-phase catalytic growth of singlewalled carbon nanotubes from carbon monoxide. Chem. Phys. Lett. 313, 91–97 (1999) 14. Maruyama, S., et al.: Low-temperature synthesis of highpurity single-walled1 carbon nanotubes from alcohol. Chem. Phys. Lett. 360, 229–234 (2002) 15. Mensah, S.L., et al.: Formation of single crystalline ZnO nanotubes without catalysts and templates. Appl. Phys. Lett. 90, 113108 (2007) 16. Mensah, S.L., et al.: Selective growth of pure and long ZnO nanowires by controlled vapor concentration gradients. J. Phys. Chem. C 111, 16092–16095 (2007) 17. Mensah, S.L., et al.: ZnO nnosquids: banching nnowires from nnotubes and nnorods. J. Nanosci. Nanotechnol. 8, 233–236 (2008) 18. Lee, C.H., et al.: Effective growth of boron nitride nanotubes by thermal chemical vapor deposition. Nanotechnology 19(455605), 5 (2008) 19. Lee, C.H., et al.: Patterned growth of boron nitride nanotubes by catalytic chemical vapor deposition. Chem. Mater. 22, 1782–1787 (2010) 20. Wang, J., Lee, C.H., Yap, Y.K.: Recent advancements in boron nitride nanotubes. Nanoscale. 2, 2028–2034 (2010) Chemistry of Carbon Nanotubes ▶ Functionalization of Carbon Nanotubes Chitosan ▶ Chitosan Nanoparticles 427 C Chitosan Nanoparticles Burcu Aslan1, Hee Dong Han2,4, Gabriel Lopez-Berestein1,3,4,5 and Anil K. Sood2,3,4,5 1 Department of Experimental Therapeutics, M.D. Anderson Cancer Center, The University of Texas, Houston, TX, USA 2 Gynecologic Oncology, M.D. Anderson Cancer Center, The University of Texas, Houston, TX, USA 3 Cancer Biology, M.D. Anderson Cancer Center, The University of Texas, Houston, TX, USA 4 Center for RNA Interference and Non–coding RNA, M.D. Anderson Cancer Center, The University of Texas, Houston, TX, USA 5 The Department of Nanomedicine and Bioengineering, UTHealth, Houston, TX, USA Synonyms Chitosan; Drug delivery system; Nanoparticles Definition Chitosan nanoparticles are biodegradable, nontoxic carriers for nucleotides and drugs with the potential for broad applications in human disease. Overview Characteristics of Chitosan Chitosan is a natural cationic polysaccharide composed of randomly distributed N-acetyl-D-glucosamine and b-(1,4)-linked D-glucosamine. Chitosan can be chemically synthesized via alkaline deacetylation from chitin, which is the principal component of the protective cuticles of crustaceans [1]. Chitosan is biodegradable in vivo by enzymes such as lysozyme, which is endogenous and nontoxic [2]. In addition, biodegradation of chitosan is highly associated with the degree of deacetylation. These properties render chitosan particularly attractive for clinical and biological applications as a highly biocompatible material with low toxicity and immunogenicity. Several techniques have been designed to assemble chitosan nanoparticles (CH-NPs) as a drug delivery C C 428 Chitosan Nanoparticles system including emulsions, ionotropic gelation, micelles, and spray drying. A variety of therapeutic agents can be loaded into CH-NPs with high efficiency, which can then be injected intravenously, intraperitoneally, or intrathecally. incorporation of low molecular drugs, proteins, DNA/ siRNA have been demonstrated [5]. CH-NPs are rapidly formed through ionic interactions between the negatively charged phosphates of TPP and positively charged amino groups of chitosan. Chemical Modification of Chitosan The abundant amine and hydroxyl groups present in chitosan offer a unique opportunity to attach targeting ligands or imaging agents. Numerous derivatives of chitosan have been designed and tailored to improve the physicochemical and adhesive properties of nanoparticles such as size, shape, charge, density, and solubility. Quaternized chitosan, N,N,N-trimethyl chitosan, thiolated chitosan, carboxyalkyl chitosan, sugar-bearing chitosan, bile acid-modified chitosan, and cyclodextrin-linked chitosan are among the modifications frequently utilized in chitosan-based drug delivery systems. Each modification offers unique properties and characteristics. For instance, trimethyl chitosan is soluble over a wide pH range and enhances the condensation capacity of plasmid DNA at neutral pH due to fixed positive charges on its backbone. Thiolation of chitosan provides free sulphydryl groups on its side chains and forms disulfide bonds with cysteine-rich subdomains of muco-glycoproteins on cell membranes and increases cellular uptake. In addition, both modifications have been used as nonviral carrier systems to combine the advantages of trimethyl chitosan and thiolated chitosan while minimizing their shortcomings [3]. Hydrophobic moieties can also be attached to chitosan to facilitate the incorporation of insoluble drugs, i.e., hydrophobic glycol chitosan nanoparticles. Chemical conjugation of hydrophobic 5b-cholanic acid to the hydrophilic glycol chitosan backbone allows for the incorporation of the water-insoluble drug camptothecin [4]. Drug Delivery Ringsdorf first reported the concept of polymer–drug conjugates for delivering small molecule drugs [6]. The concept of polymer–drug conjugates allows chemical conjugation of a drug using a biodegradable spacer. The spacer is usually stable in the bloodstream, but cleaved at the target site by hydrolysis or enzymatic degradation. Based on this concept, several polymer–drug conjugates have been developed such as glycol chitosan conjugated with doxorubicin, which forms self-assembled nanoparticles in an aqueous condition. A paclitaxel-chitosan conjugate that can be cleaved at physiological conditions was developed for oral delivery of paclitaxel. N-succinyl-chitosanmitomycin conjugates demonstrated high antitumor efficacy against a variety of murine tumor models of leukemia, melanoma, and primary and metastatic liver tumors. Key Research Findings Preparation of Chitosan Nanoparticles The amino groups of chitosan backbone can interact with anionic molecules such as tripolyphosphate (TPP). The ionic cross-linking of chitosan is advantageous since the method is easy and mostly performed under mild conditions without using organic solvents. Ionotropic gelation of chitosan using TPP for the Gene Delivery Recently, plasmid DNA, siRNA, and oligonucleotideloaded CH-NPs have been used for targeted gene silencing. Positively charged chitosan can easily form polyelectrolyte complexes with negatively charged nucleotides based on electrostatic interaction, incorporation, or adsorption, as illustrated in Fig. 1 [7]. The positive charge on the surface of CH-NPs is desirable to prevent aggregation due to electrostatic repulsion and increase binding efficiency with the negatively charged cell membrane by enhancing electrostatic interactions. However, other cellular uptake mechanisms such as clathrin-mediated endocytosis, caveolae-mediated endocytosis, and macropinocytosis may also be involved [8]. Moreover, the configuration of chitosan is modified under acidic pH by triggering the opening of tight junctions. Acidic pH results in an increase in the number of protonated amines on the chitosan leading to a further disruption on membrane organization. It has also been reported that chitosan can swirl across the membrane lipid bilayer and facilitate the cellular uptake of the polyplex due to increase in mole fraction of chitosan, leading to reduction in the polymeric chains which results in decreased molecular Chitosan Nanoparticles 429 C Chitosan Nanoparticles, Fig. 1 Preparation of chitosan-based DNA/siRNA nanoparticles based on different mechanisms [21] Chitosan and its derivatives Electrostatic interaction weight [7]. In addition, Lu et al. have recently reported the biodistribution of intravenously administered siRNA-CH-NPs in tumor-bearing mice. They demonstrated that CH-NPs allowed for a higher localization of siRNA in tumor tissues compared to other organs (Fig. 2) [9]. Electrostatic interactions between protonated amines of chitosan and negative charge of DNA or siRNA leads to spontaneous formation of highly compact encapsulation of either DNA or siRNA into CH-NPs [10]. Gel electrophoresis has been used to assess hydrogen bonding and hydrophobic interactions between chitosan and DNA [11]. However, a major limitation of siRNA delivery is its rapid degradation in plasma and cytoplasm. It has been reported that stable chitosan/siRNA complexes can protect siRNA degradation in circulation of CH-NPs in bloodstream to overcome extracellular and intracellular barriers. On the other hand, disassembly is also needed to allow release of siRNA. This emphasizes the importance of an appropriate balance between protection and release of siRNA for biological functionality [12]. Positively charged CH-NPs can bind to negatively charged cell surfaces with high affinity. CH-NPs are known to overcome endosomal escape via its “proton sponge effect.” Once these nanoparticles penetrate into an acidifying lysosomal compartment, the unsaturated amino groups of chitosan distrain protons that are delivered by proton pumps (vATPase) which is called Encapsulation DNA/siRNA Adsorption the “proton sponge mechanism.” Subsequent lysosomal swelling and rupture leads to endo-lysosomal escape of nanoparticles [13]. Targeted Delivery An ideal delivery system should lead to enhanced concentrations of therapeutic payloads at disease sites, and minimize potential non-desirable off-target effects, and ultimately raise the therapeutic index. Differences between tumor and normal tissue microenvironment and architecture, such as vascularization, overexpressed receptors, pH, temperature, ionic strength, and metabolites, can be exploited for selective targeting. Targeted delivery systems have been designed to increase and/or facilitate uptake into target cells, and to protect therapeutic payloads. Recent work comparing non-targeted and targeted nanoparticles have shown that the primary role of the targeting ligands is to enhance selective binding efficiency to receptor in cell surface and cellular uptake into target cells, and to minimize accumulation in normal tissues. The addition of targeting ligands that provide specific ligandreceptor binding on nanoparticle-cell surface interactions can play a vital role in the ultimate location of nanoparticles. For example, nanoparticles decorated with specific moieties such as peptides, proteins, or antibodies can be targeted to cancer cells via cellsurface receptor proteins such as transferrin or folate C C 430 Chitosan Nanoparticles a b 7 c Cy5.5 siRNA/CH 6 5 Unlabeled siRNA/CH 4 3 n Br ai s ng Lu ey dn Ki en le Sp ve r Li or Tu m H ea rt Untreated control 2 p/s/cm2/Sr Chitosan Nanoparticles, Fig. 2 (a) Fluorescent siRNA distribution in tumor tissue. Hematoxylin and eosin, original magnification 2003 (left); tumor tissues were stained with anti-CD31 (green) antibody to detect endothelial cells (right). The scale bar represents 50 mm. (b) Fluorescent siRNA distribution in tumor tissue. Sections (8 mm thick) were stained with Sytox green and examined with confocal microscopy (scale bar represents 20 mm) (left); lateral view (right). Photographs taken every 1 mm were stacked and examined from the lateral view. Nuclei were labeled with Sytox green and fluorescent siRNA (red) was seen throughout the section. At all time points, punctated emissions of the siRNA were noted in the perinuclear regions of individual cells, and siRNA was seen in >80% of fields examined. (c) Shows fluorescence intensity overlaid on white light images of different mouse organs and tumor receptors (known to be increased on a wide range of cancer cells). These targeting ligands enable nanoparticles to bind on to cell-surface receptors and penetrate cells by receptor-mediated endocytosis. Target selective ligand-labeled CH-NPs can enhance receptor-mediated endocytosis. Various receptors on the tumor cell surface have been established as a target-binding site to achieve selective delivery. The overexpression of transferrin and folate in certain tumors has been exploited to deliver CH-NPs conjugated with these receptor’s ligands [14, 15]. Another example is the anb3 integrin, which is overexpressed in a wide range of tumors, and is largely absent in normal tissues. Han et al. [16] have recently reported that the administration of RGD peptidelabeled CH-NPs led to increased tumor delivery of siRNA-CH-NP and enhanced antitumor activity in ovarian carcinoma models (Figs. 3 and 4). Chitosan-Based Environmental-Responsive Particles Physiological alterations such as pH, temperature, ionic strength, and metabolites in the microenvironment of tumor have gained increased interest in terms Chitosan Nanoparticles 431 C SKOV3ip1 Control A2780ip2 150 SKOV3ip1 C 100 50 0 C on tro C l R G H-N D -C P H -N C P on tro R CH l G D -NP -C H -N P RGD-CH-NP CH-NP Binding efficiency (%) (NPs/cells) A2780ip2 Chitosan Nanoparticles, Fig. 3 Binding of Alexa555 siRNA/RGD-CH-NPs and Alexa555 siRNA-CH-NP in SKOV3ip1 or A2780ip2 cells by fluorescence microscopy SKOV3ip1 Chitosan Nanoparticles, Fig. 4 Binding of siRNA/ RGD-CH-NPs in SKOV3ip1 or A2780ip2 cells by transmission electron microscopy against ovarian cancer cells in vitro RGD-CH-NP Control A2780ip2 of targeted therapy. Potential differences in these parameters between tumor and normal tissue can be used for enhanced targeting. A novel, pH responsive NIPAAm/CH-NPs containing camptothecin and paclitaxel was successfully used to enhance tumor uptake and antitumor activity [17]. On the other hand, thermosensitive CH-g-poly(N-vinylcaprolactam) composite has been developed by an ionic crosslinking method and incorporated with 5-FU(5- fluorouracil). This study showed that 40% of drug is released from the particles when the temperature is above a lower critical solution temperature of 38 C while only 5% of drug is released below 38 C, which confirms the drug release mechanism of this polymeric carrier system is based on temperature. These nanoparticles were more toxic to cancer cells while devoid of toxicity to normal cells, leading to enhance antitumor efficacy [18]. C 432 Chitosan-Based Magnetic Nanoparticles Chitosan-based magnetic nanoparticles have been developed for magnetic resonance (MR) imaging via passive and tissue-specific targeting. The lowoxidizing ferromagnetic materials are the most commonly used compounds for nanoparticle formulations and provide a stable magnetic response. The accumulation of magnetic nanoparticles when injected intravenously can be induced when the tissue or organ is subjected to an external high-gradient magnetic field [19]. Drugs, DNA plasmids, or bioactive molecules are released into target tissues and effectively taken up by tumor cells after accumulation of these carriers. On the other hand, magnetic nanoparticles such as iron oxide nanoparticles (Fe3O4) are applied to oscillating magnetic fields, which results in the generation of heat and holds potential for rapid heating of tumor tissue. When magnetic nanoparticles are administered systemically, they are rapidly coated with plasma proteins. These particles are taken up by the reticuloendothelial system, leading to decreased circulation time. Magnetic nanoparticles coated with hydrophilic materials such as chitosan could provide longer circulation time. In addition, chitosan-coated magnetic nanoparticles, which are used in magnetic resonance imaging, can also be functionalized and used as theranostic carriers due to amino and hydroxyl groups of chitosan. Application of Chitosan Nanoparticles for SiRNA Delivery Recently, siRNA targeted to EZH2 (a critical component of the polycomb repressive complex 2 [PRC2]), loaded CH-NPs were shown to enhance the delivery of siRNA to tumors, leading to downregulation of the target protein and subsequently enhanced antitumor activity. Lu et al. demonstrated target gene silencing using EZH2 siRNA loaded CH-NPs, leading to increased antitumor efficacy in animal tumor models. Moreover, Zhang et al. reported the use of intrathecal administration of siRNA targeted against specific muscarinic receptor subtypes loaded in CH-NPs in a rat model of pain. CH-NP/siRNA was distributed to the spinal cord and the dorsal root ganglion [20]. The administration of Chitosan-M2-siRNA caused a large reduction in the inhibitory effect of muscarine on the rat paw withdrawal threshold from a heat stimulus. These studies support the use of CH-NPs as a delivery system for siRNA into neuronal tissues in vivo. Chitosan Nanoparticles Future Directions for Research Chitosan nanoparticles offer a unique potential for clinical and biological applications due to low immunogenicity, low toxicity, and high biocompatibility. In addition to its advantages such as protonated amine groups, chitosan can increase binding efficiency with cells because of electrostatic interactions. Therefore, CH-NPs may be used for broad applications in human disease. Moreover, chitosan allow modifications that will exploit the inherent physicochemical properties by conjugation of selective ligands. The highly desirable specific targeting of drugs has been elusive to date; however, CH-NPs can bring us closer to this goal. Acknowledgments Portions of this work were supported by the NIH (CA 110793, 109298, P50 CA083639, P50 CA098258, CA128797, RC2GM092599, U54 CA151668), the Ovarian Cancer Research Fund, Inc. (Program Project Development Grant), the DOD (OC073399, W81XWH-10-1-0158, BC085265), the Zarrow Foundation, the Marcus Foundation, the Kim Medlin Fund, the Laura and John Arnold Foundation, the Estate of C. G. Johnson, Jr., the RGK Foundation, and the Betty Anne Asche Murray Distinguished Professorship. Cross-References ▶ Effect of Surface Modification on Toxicity of Nanoparticles ▶ Nanomedicine ▶ Nanoparticle Cytotoxicity ▶ Nanoparticles References 1. Muzzarelli, R.A.A.: Chitin (1–37). Pergamon, Elmsford (1977) 2. Khor, E.: Chitin: Fulfilling a Biomaterials Promise. Elsevier, Oxford, UK (2001) 3. Zhao, X., Yin, L., Ding, J., Tang, C., Gu, S., Yin, C., Mao, Y.: Thiolated trimethyl chitosan nanocomplexes as gene carriers with high in vitro and in vivo transfection efficiency. J. Control. Release 144, 46–54 (2010) 4. Min, K.H., Park, K., Kim, Y., Bae, S.M., Lee, S., Jo, H.G., Park, R.W., In-San Kim, I.S., Jeong, S.Y., Kim, K., Kwon, I.C.: Hydrophobically modified glycol chitosan nanoparticles-encapsulated camptothecin enhance the drug stability and tumor targeting in cancer therapy. J. Control. Release 127, 208–218 (2008) 5. Park, J.H., Saravanakumar, G., Kim, K., Kwon, I.C.: Targeted delivery of low molecular drugs using chitosan and its derivatives. Adv. Drug Deliv. Rev. 62, 28–41 (2010) CMOS MEMS Biosensors 6. Ringsdorf, H.: Structure and properties of pharmacologically active polymers. J. Polym. Sci. Polym. Symp. 51, 135–153 (1975) 7. Lai, W.F., Lin, M.C.M.: Nucleic acid delivery with chitosan and its derivatives. J. Control. Release 134, 158–168 (2009) 8. Nam, H.Y., Kwon, S.M., Chung, H., Lee, S.Y., Kwon, S.H., Jeon, H., Kim, Y., Park, J.H., Kim, J., Her, S., Oh, Y.K., Kwon, I.C., Kim, K., Jeong, S.Y.: Cellular uptake mechanism and intracellular fate of hydrophobically modified glycol chitosan nanoparticles. J. Control. Release 135, 259–267 (2010) 9. Lu, C., Han, H.D., Mangala, L.S., Ali-Fehmi, R., Newton, C.S., Ozbun, L., Armaiz-Pena, G.N., Hu, W., Stone, R.L., Munkarah, A., Ravoori, M.K., Shahzad, M.M.K., Lee, J.W., Mora, E., Langley, R.R., Carroll, A.R., Matsuo, K., Spannuth, W.A., Schmandt, R., Jennings, N.J., Goodman, B.W., Jaffe, R.B., Nick, A.M., Kim, H.S., Guven, E.O., Chen, Y.H., Li, L.Y., Hsu, M.C., Coleman, R.L., Calin, G. A., Denkbas, E.B., Lim, J.Y., Lee, J.S., Kundra, V., Birrer, M.J., Hung, M.C., Lopez-Berestein, G., Sood, A.K.: Regulation of tumor angiogenesis by EZH2. Cancer Cell 18, 185–197 (2010) 10. Mao, S., Sun, W., Kissel, T.: Chitosan-based formulations for delivery of DNA and siRNA. Adv. Drug Deliv. Rev. 62, 12–27 (2010) 11. Messai, I., Lamalle, D., Munier, S., Verrier, B., Ataman-Onal, Y., Delair, T.: Poly(D, Llactic acid) and chitosan complexes: interactions with plasmid DNA. Colloids Surf. A Physicochem. Eng. Asp. 255, 65–72 (2005) 12. Liu, X., Howard, K.A., Dong, M., Andersen, M.O., Rahbek, U.L., Johnsen, M.G., Hansen, O.C., Besenbacher, F., Kjems, J.: The influence of polymeric properties on chitosan/siRNA nanoparticle formulation and gene silencing. Biomaterials 28, 1280–1288 (2007) 13. Nel, A.E., M€adler, L., Velegol, D., Xia, T., Hoek, E.M.V., Somasundaran, P., Klaessig, F., Castranova, C., Thompson, M.: Understanding biophysicochemical interactions at the nano–bio interface. Nat. Mater. 8, 543–557 (2009) 14. Mao, H.Q., Roy, K., Troung-Le, V.L., Janes, K.A., Lin, K.Y., Wang, Y., August, J.T., Leong, K.W.: Chitosan–DNA nanoparticles as gene carriers: synthesis, characterization and transfection efficiency. J. Control. Release 70(3), 399–421 (2001) 15. Fernandes, J.C., Wang, H., Jreyssaty, C., Benderdour, M., Lavigne, P., Qiu, X., Winnik, F.M., Zhang, X., Dai, K., Shi, Q.: Bone-protective effects of nonviral gene therapy with Folate–Chitosan DNA nanoparticle containing Interleukin1 receptor antagonist gene in rats with adjuvant-induced arthritis. Mol. Ther. 16(7), 1243–1251 (2008) 16. Han, H.D., Mangala, L.S., Lee, J.W., Shahzad, M.M.K., Kim, H.S., Shen, D., Nam, E.J., Mora, E.M., Stone, R.L., Lu, C., Lee, S.J., Roh, J.W., Nick, A.M., Lopez-Berestein, G., Sood, A.K.: Targeted gene silencing using RGDLabeled chitosan nanoparticles. Clin. Cancer Res. 16(15), 3910–3922 (2010) 17. Li, F., Wu, H., Zhang, H., Gu, C.H., Yang, Q.: Antitumor drug paclitaxel loaded pH-sensitive nanoparticles targeting tumor extracellular pH. Carbohydr. Polym. 77(4), 773–778 (2009) 18. Rejinold, N.S., Chennazhi, K.P., Nair, S.V., Tamura, H., Jayakumar, R.: Carbohydr. Polym (2010). doi:10.1016/j. carbpol.2010.08.052 433 C 19. Veiseh, O., Gunn, J.W., Zhang, M.: Design and fabrication of magnetic nanoparticles for targeted drug delivery and imaging. Adv. Drug Deliv. Rev. 62, 284–304 (2010) 20. Zhang, H.M., Chen, S.R., Cai, Y.Q., Richardson, T.E., Driver, L.C., Lopez-Berestein, G., Pan, H.L.: Signaling mechanisms mediating muscarinic enhancement of GABAergic synaptic transmission in the spinal cord. Neuroscience 158, 1577–1588 (2009) 21. Lai, W.F., Lin, M.C.M.: Nucleic acid delivery with chitosan and its derivatives. J. Control. Release 134, 158–168 (2009) Clastogenicity or/and Aneugenicity ▶ Genotoxicity of Nanoparticles Clinical Adhesives ▶ Bioadhesives Cluster ▶ Synthesis of Subnanometric Metal Nanoparticles CMOS (Complementary Metal-OxideSemiconductor) ▶ CMOS MEMS Fabrication Technologies CMOS MEMS Biosensors Michael S.-C. Lu Department of Electrical Engineering, Institute of Electronics Engineering, and Institute of NanoEngineering and MicroSystems, National Tsing Hua University, Hsinchu, Taiwan, Republic of China Synonyms Integrated biosensors C C 434 Definition CMOS MEMS biosensors are miniaturized biosensors fabricated on CMOS (complementary metal-oxide semiconductor) chips by the MEMS (microelectromechanical systems) technology. Overview A biosensor is a device designed to detect a biochemical molecule such as a particular deoxyribonucleic acid (DNA) sequence or particular protein. Many biosensors are affinity-based, meaning that they use an immobilized capture probe that selectively binds to the target molecule being sensed. Most biosensors require a label attached to the target, and the amount of detected label is assumed to correspond to the number of bound targets. Labels can be fluorophores, magnetic beads, gold nanoparticles, enzymes, or anything else allowing convenient binding and detection; however, labeling a biomolecule can change the associated binding properties, especially for protein targets. An electrical biosensor is capable of detecting a binding event by producing an electrical current and/ or a voltage. Conventional optical detection methods require external instruments that are expensive and not amenable to miniaturization. Electrical bioassays hold great promise for numerous decentralized clinical applications ranging from emergency-room screening to point-of-care diagnostics due to their low cost, high sensitivity, specificity, speed, and portability. Miniaturization of an electrical biosensor can be achieved through microfabrication – widely known as the MEMS technology; in addition, the sensing circuits can be embedded on the same chip through integrated-circuit (IC) processes, among which the CMOS technology is the most popular choice for implementing various analog and digital circuits. A fully integrated CMOS MEMS biosensor array is capable of providing real-time high-throughput detection of multiple samples. CMOS biosensors can be implemented based on the electrochemical, impedimetric, ion-sensitive, magnetic, optical, and micromechanical approaches, which require different MEMS processes to construct the sensing interfaces. Some of the methods require labeling and some are label-free for bio-signal transduction. More details are provided in the following sections. CMOS MEMS Biosensors Sensing Principles and Key Research Findings Electrochemical Biosensors Electrochemical biosensors have been the subject of basic as well as applied research for many years. In 1970, Dr. Leland C. Clark demonstrated that glucose could be measured in whole blood with the presence of the glucose oxidase enzyme. Commercial glucose sensors based on electrochemical detection have been developed since then. Electrochemical biosensors typically depend on the presence of a suitable enzyme in the biorecognition layer to catalyze reaction of electroactive substances. Affinity-based electrochemical sensors use enzymes as labels that bind to antibodies, antigens, or oligonucleotides with a specific sequence. Electrochemical biosensors can employ potentiometric, amperometric, and impedimetric sensing principles to convert the chemical information into a measurable electrical signal. In potentiometric devices, ion selective electrodes (ISE) are commonly used to transduce a biorecognition event into a potential signal between the working and the reference electrodes whose value, depending on the concentration of the analyte, can be predicted by the Nernst equation. The current flowing through the electrode is equal to or near zero. Amperometric biosensors operate by applying a constant potential and monitoring the current associated with the reduction or oxidation of an electroactive species. The sensors can work in three- or fourelectrode configurations. The former case consists of a reference, a working, and a counter electrode. The four-electrode setup has an additional working electrode such that oxidation and reduction take place at anode and cathode simultaneously. Working electrodes are normally made of noble metals which are critical to the operation of a sensor. Since these metals are not CMOS compatible, the electrodes must be formed after fabrication of CMOS circuits. Silver/silver chloride (Ag/AgCl) is commonly used as the reference electrode in many electrochemical biosensors. Since not all particles oxidized at the anode reach the cathode, a potentiostat, whose input and output are connected to a reference and to a counter electrode, respectively, is required to provide the difference current to the electrolyte and regulate the potential of the electrolyte to a constant value. CMOS MEMS Biosensors CMOS MEMS Biosensors, Fig. 1 Schematic of the CMOS interdigitated microelectrode for electrochemical DNA detection. On the right: schematic illustration of the redox-cycling process (Schienle et al. # 2004 IEEE) C 435 counter electrode generator electrode DNA target substrate DNA probe Igen Vgen C enzyme label Icol red Vcol potentiostat ox collector electrode Interdigitated microelectrodes have been adopted in many electrochemical biosensors for quantitative analysis. The width and spacing of microelectrodes are reduced by microfabrication. The main advantage is the enhanced redox current due to fast redox recycling, as the chemical products produced at one side of the electrodes are readily collected at the other side of adjacent electrodes and regenerated to the original states. The relationship between the produced redox current and the analyte concentration has been derived by Aoki et al. [1]. Electrochemical DNA hybridization biosensors rely on the conversion of the DNA base-pair recognition event into an electrical signal. Schienle et al. [2] reported a CMOS electrochemical DNA sensor array with each sensing element consisting of interdigitated gold electrodes separated by 1 mm. As depicted in Fig. 1, probe molecules were immobilized on the gold surface through thiol coupling and the target molecules were tagged by an enzyme label (alkaline phosphatase). A chemical substrate (paraaminophenyl phosphate) was applied to the chip after hybridization. The enzyme label on the matched DNA strands cleaved the phosphate group and generated an electroactive compound (para-aminophenol), which was subsequently oxidized and reduced as the indicator of successful DNA hybridization. Levine et al. [3] also reported a CMOS electrochemical DNA sensor array which was operated based on conventional cyclic voltammetry (CV). During the measurement the electrode potential was scanned up and down in order to reference electrode produce the redox reactions associated with the ferrocene labels attached to DNA strands. In addition to DNA detection, CMOS electrochemical sensors have been realized to allow high-throughput detection of dopamine and catecholamine release from adrenal chromaffin cells [4], since the release of neurotransmitters from secretory vesicles of biological cells is closely related to the function of a nervous system. Impedimetric Biosensors Changes in the electrical properties of a sensing interface (e.g., capacitance, resistance) can occur when a target biomolecule interacts with a probe-functionalized surface. A conventional impedimetric biosensor measures the electrical impedance of an electrode-solution interface in a.c. steady state with constant d.c. bias conditions. This approach, known as electrochemical impedance spectroscopy (EIS), is accomplished by imposing a small sinusoidal voltage over a range of frequencies and measuring the resulting current. The current–voltage ratio gives the impedance, which consists of both energy dissipation (resistor) and energy storage (capacitor) elements. Results obtained by EIS are often graphically represented by a Bode plot or a Nyquist plot. EIS reveals information about the reaction mechanism of an electrochemical process since different reaction steps can dominate at certain frequencies. Impedimetric biosensors can detect a variety of target analytes by simply varying the probe used. Changes in the impedance can be correlated to DNA hybridization, antigen–antibody reaction, or be used to detect biological cells. C 436 CMOS MEMS Biosensors CMOS MEMS Biosensors, Fig. 2 Schematic of the impedance extraction method (Adopted by Lee et al. # 2010 Elsevier) silicon substrate PBS PH7.4 (Phosphate Buffer saline) peak iin Vin RSENSE CSENSE valley iC Vin iR iR iC iin Current Analysis Low Frequency Response A B A ∝ CSENSE B ∝ 1 / RSENSE The interface impedance is commonly represented by an equivalent circuit model for analysis. E. Warburg [5] first proposed that the interface impedance can be represented by a polarization resistance in series with a polarization capacitor. The interface capacitance possesses a frequency dependency and is commonly represented by the Gouy-Chapman-Stern model as the series combination of the double-layer capacitance (Helmholtz capacitance) and the diffuse layer capacitance (Gouy-Chapman capacitance). In addition to the frequency-domain measuring method, interface impedance changes can be measured by the potentiostatic step method where small potential steps are applied to the working electrode and the transient current responses, as determined by the time constant of the interface resistance and capacitance, are measured accordingly. Lee et al. [6] reported a fully integrated CMOS impedimetric sensor array for label-free detection of DNA hybridization. The changes in the reactive capacitance and the chargetransfer resistance on the gold sensing electrodes were extracted by applying a triangular voltage waveform and monitoring the produced currents. As illustrated in Fig. 2, the current flowing through the capacitor is associated with the slope of the applied triangular wave, while the current flowing through the resistor is in proportion to the magnitude of the triangular wave. The reported detection limit was 10 nanomolar (nM). It is important to distinguish the differences between faradaic and non-faradaic biosensors for impedance detection. In electrochemical terminology, a faradaic process involves charge transfer across an interface, while a transient current can flow without charge transfer in a non-faradaic process by charging a capacitor. The faradaic EIS requires the addition of a redox species which is alternately oxidized and reduced by the transfer of charges to and from the metal electrode. In contrast, no additional reagent is required for non-faradaic impedance spectroscopy. The associated impedance change is predominantly capacitive with the charge transfer resistance being omitted. Miniaturized CMOS capacitive sensors have been developed for numerous biosensing applications. As the sensor size is small, monolithic integration provides the benefit of enhancing the signal-to-noise ratio by reducing the parasitic capacitance observed at the sensing node, which would otherwise negatively impact the detection limit during direct capacitance measurement. Stagni et al. [7] reported an 8  16 CMOS MEMS Biosensors CMOS MEMS Biosensors, Fig. 3 (a) Schematic of the floating-gate ISFET structure. (b) Schematic of the open-gate ISFET structure 437 a b Dielectric thin film Liquid (Gate) Drain Source Source Drain n+ n+ gate oxide Ion-Sensitive Field Effect Transistors (ISFET) ISFETs were first developed in the early 1970s and have been utilized in various biosensing applications which depend on the type of receptor used for analyte recognition or how a signal is generated. Immunologically modified FETs and DNA-modified FETs detect the change of surface charges by monitoring the associated current–voltage relationship. Cell-based FETs detect the potential changes produced by biological cells due to the flow of ions across the cell membranes upon stimulations. For applications where it is required ions n+ Gate silicon CMOS DNA sensor array where the bindings of complementary DNA strands on gold microelectrodes reduced the dielectric constant of the electrode-analyte impedance and the associated change was used to modify the charging and discharging transients of the detection circuit. Capacitive sensitivity can be enhanced by use of interdigitated microelectrodes with the minimum gap defined by the adopted CMOS process. Lu et al. [8] reported very sensitive detection of the neurotransmitter dopamine in the subfemtomolar (fM) range. CMOS capacitive sensors have also been used for monitoring cellular activity since the morphological and physiological states of biological cells have correlations with their electrical properties [9]. The impedance change of a biosensing event can be made purely resistive with additional chemical modifications. Li et al. [10] reported a DNA array detection method in which the binding of immobilized DNA probes functionalized with gold nanoparticles produced a conductivity change between adjacent metal electrodes, which were made of gold metal in order to withstand the chemical in the clean processing. The detected signal can be further enhanced by additional silver deposition. Silicon dioxide was used as the underlying material under the gap for immobilization of the DNA probes. Detection limit of the target DNA concentration was 1 picomolar (pM). C n+ silicon to build an ISFET array to provide detection of multiple samples at different locations, monolithic integration is then preferred in order to reduce wiring complexity and noise interference. CMOS ISFET sensors can be built by using a floating-gate or an open-gate structure as depicted in Fig. 3. The floating-gate structure is easier to fabricate as it requires minimal post-processing after completion of a conventional CMOS process. Silicon dioxide or silicon nitride in the CMOS passivation thin films can be used as the material for surface functionalization. The produced signal due to charges on sensor surface is capacitively coupled to the floating gate through a relatively thick dielectric layer which reduces the signal-coupling efficiency. The gate in an open-gate ISFET is removed and replaced by the aqueous solution whose potential is commonly set via a reference electrode. The exposed gate oxide is used for surface functionalization. Since the gate oxide thickness is only tens of angstroms in a conventional sub-mm CMOS process, sensitivity is thus significantly enhanced as compared to a floating-gate ISFET. Accumulated charges on the ISFET surface modulate the threshold voltage of a MOS transistor, leading to a channel current change under fixed voltage biases of the drain, source, and gate terminals. The threshold voltage of an open-gate ISFET is expressed as: Vth ¼ Eref  C þ wsol  fsi Qox þ Qss þ QB  þ 2fF q Cox where Eref is the reference electrode potential, C is a chemical input parameter as a function of solution pH value, wsol is the surface dipole potential of the solvent, fsi is the work function of silicon, q is the electron charge, Cox is the gate oxide capacitance per unit area, Qox and Qss are the charges in the oxide and at the oxide-silicon interface, QB is the depletion charges in silicon and fF is the difference between the Fermi potential of the substrate and intrinsic silicon. C C 438 Kim et al. [11] reported the use of p-type ISFET fabricated in a standard CMOS process for detection of DNA immobilization and hybridization. Gold was deposited by post-processing as the gate material for immobilizing DNA due to its chemical affinity with thiols. The channel current increased during hybridization due to the negative charges present in the phosphate groups of DNA strands. Li et al. [12] presented a post-CMOS fabrication method to make open-gate ISFETs and demonstrated ultrasensitive dopamine detection in the fM range. Magnetic Biosensors Some of the magnetic biosensors require the use of magnetic beads as labels attached to the samples in order to induce a measurable electrical signal when specific binding on sensor surface occurs. Magnetic beads are made of small ferromagnetic or ferrimagnetic nanoparticles that exhibit a unique quality referred to as superparamagnetism in the presence of an externally applied magnetic field. This phenomenon, as discovered by Louis Néel (Nobel Physics Prize winner in 1970), has been used in numerous applications such as magnetic data storage and magnetic resonance imaging (MRI). Several methods can be used to electronically detect the existence of magnetic beads, such as the GMR (giant magnetoresistance) effect discovered by the 2007 Nobel Physics Prize winners Albert Fert and unberg. The effect appears in thin film strucPeter Gr€ tures composed of alternating ferromagnetic and nonmagnetic layers. A significant change in the electrical resistance is observed depending on whether the magnetization of adjacent ferromagnetic layers is in a parallel alignment (the low-resistance state) due to the applied magnetic field or an anti-parallel alignment (the high-resistance state) in the absence of the magnetic field. Han et al. [13] reported a CMOS DNA sensor array that adopted the spin valve structure to observe the GMR effect. Magnetic thin films of nanometers in thickness were deposited and patterned after the conventional CMOS process. The biotinylated analyte DNA was captured by complementary probes immobilized on the sensor surface. Then streptavidincoated magnetic labels were added and produced specific binding to the hybridized DNA. The stray magnetic field of the magnetic labels was detected as a resistance change in the sensor. In general, signal CMOS MEMS Biosensors modulation is required in the sensing scheme in order to separate the true bio-signals from the false ones caused by drifts or ionic solution interference. Other than the GMR principle, detection of magnetic beads can be achieved based on the Hall effect of a CMOS sensor. As discovered by Edwin Hall in 1879, the effect produces an electric field perpendicular to the magnetic induction vector and the original current direction. Detection of a single magnetic bead has been demonstrated [14]. To eliminate the needs of externally applied magnetic fields and post-CMOS fabrication for a magnetoresistive biosensor, Wang et al. [15] presented an inductive approach that can detect existence of a single magnetic bead on a CMOS chip. The sensing scheme used a highly stable integrated oscillator with an on-chip LC resonator. An a.c. electrical current through the on-chip inductor produced a magnetic field that polarized the magnetic particles present in its vicinity, leading to an increased effective inductance and therefore a reduced oscillation frequency. As the frequency shift due to a single micron-size magnetic bead is typically a few parts per million (ppm) of the resonant frequency, the sensing oscillator needs to have small phase noises at small offset frequencies to achieve a stable frequency behavior. A miniaturized CMOS NMR (nuclear magnetic resonance) system has been reported by Sun et al. [16] for applications in biomolecular sensing. NMR was first discovered by Isidor Rabi who was awarded Nobel Prize in Physics in 1944. Magnetic nuclei, like 1 H and 31P, could absorb radio frequency (RF) energy when placed in a magnetic field of a strength specific to the identity of the nuclei. The nucleus is described as being in resonance when the absorption occurs. Different atomic nuclei within a molecule exhibit different resonant frequencies for the same magnetic field strength. Essential chemical and structural information about a molecule can thus be studied by observing such magnetic resonant frequencies. The reported CMOS NMR system consists of a magnet (static magnetic field), an RF coil surrounding a sample, and an RF transceiver linked to the coil as shown in Fig. 4. The RF magnetic field produced through the coil at the right frequency oo can excite nuclei spins within the sample. Once the RF excitation is stopped and the receiver is connected to the coil, the detected NMR signal displays an exponential relaxation in the precession of the net magnetic moment. CMOS MEMS Biosensors CMOS MEMS Biosensors, Fig. 4 Operation of the CMOS NMR system (Sun et al. # 2009 IEEE) C 439 S Tx B0 N Rx Tx t 0 C 2π/ωo S NMR signal Tx B0 N Rx T2 Rx t 0 2π/ωo Both the resonant frequency and the relaxation’s characteristic time are specific for the sample to be studied. Optical Biosensors Optical detection methods such as luminometry and fluorometry can be utilized to make CMOS-based biosensors. Luminometric methods, such as luciferasebased assays, involve the detection and quantification of light emission as a result of a chemical reaction. Luciferase is a generic term for the class of oxidative enzymes used in bioluminescence. Such methods have been used to detect pathogens and proteins, perform gene expression and regulation studies, and sequence DNA. Fluorometric methods require an excitation source to stimulate photoemission of the fluorescent-tagged species and optical filters to separate the generated photoemission from the high background interference. Luminometry is more amenable to miniaturization and integration than the fluorometric methods since no filters or excitation sources are required. Eltoukhy et al. [17] reported a CMOS photodetector array which was directly integrated with a fiber-optic faceplate with immobilized luminescent reporters/probes for DNA synthesis by pyrosequencing. Pyrosequencing is a “sequencing by synthesis” technique which involves taking a single strand of the DNA to be sequenced and then synthesizing its complementary strand enzymatically. Synthesis of the complementary strand is achieved with one base pair at a time by monitoring the activity of the DNA synthesizing enzyme with another chemiluminescent enzyme. The challenge of such a detection system is to achieve high sensitivity despite the presence of relatively high dark currents. The reported sensor array was able to detect emission rates below 106 lux over a long integration time (30 s). Huang et al. [18] reported a CMOS microarray chip that leveraged low-cost integration of solid-state circuits for fluorescence-based diagnostics. The featured timegated fluorescence detection was able to significantly reduce interferences from the external excitation source, eliminating the need for additional optical filters. Direct immobilization of DNA capture strands on the CMOS chip allows placement of optical detectors in close vicinity to the fluorescent labels, leading to improved collection efficiency and providing imaging resolution limited by pixel dimensions rather than diffraction optics. In addition to the aforementioned approaches, a method based on detection of the incident light intensity using a normal light source has been developed [19]. The technique uses gold nanoparticles with silver enhancement to induce opacity on top of CMOS photodetectors when specific bindings occur. It does not require an external optical scanner and a specific light source as needed in the fluorescence-based method. Micromechanical Cantilevers Microcantilever-based biosensors have attracted considerable attention as a means of label-free detection of C 440 biomolecules. Intermolecular forces arising from adsorption of small molecules are known to induce surface stress. The specific binding between ligands and receptors on the surface of a microcantilever beam thus produces physical bending of the beam. Optical detection of the beam deflection due to hybridization of complementary oligonucleotides has been demonstrated [20]; however, the method requires external instruments that are not amenable to monolithic integration. Piezoresistive detection is a viable alternative for CMOS integration because it is compatible with aqueous media. The piezoresistive effect is associated with the resistivity change of a semiconductor subject to a mechanical strain. By placing a MOS transistor into the base of a cantilever, modulation of the channel current underneath the gate region can be measured as a result of adsorption-induced surface stress [21]. Sensing resolution of the static beam deflection is limited by the flicker noise – the dominant source of noise in MOS transistors at low frequencies. Summary A miniaturized biosensing platform can be achieved through monolithic integration of sensing devices and detection circuits by the CMOS MEMS technology. Sensing devices that can be directly fabricated on a CMOS chip are introduced, including those based on electrochemical, impedimetric, ion-sensitive, magnetic, optical, and micromechanical approaches. Some of the methods require labeling (e.g., magnetic beads, nanoparticles) and some are label-free for biosignal transduction. Some approaches use the devices (e.g., transistors) or the materials (e.g., metal electrodes) in a CMOS process for sensing, such that sensor performances can be enhanced in accordance with the scaling of CMOS technologies. Arrays of sensors can be conveniently fabricated on a CMOS chip such that sensing resolution and accuracy can be enhanced through statistical analysis of the collected data. Cross-References ▶ Biosensors ▶ Nanogap Biosensors CMOS MEMS Biosensors References 1. Aoki, K., Morita, M., Niwa, O., Tabei, H.: Quantitative analysis of reversible diffusion-controlled currents of redox soluable species at interdigitated array electrodes under steady-state conditions. J. Electroanal. Chem. 256, 269–282 (1988) 2. Schienle, M., Paulus, C., Frey, A., Hofmann, F., Holzapfl, B., Schindler-Bauer, P., Thewes, R.: A fully electronic DNA sensor with 128 positions and in-pixel A/D conversion. IEEE J. Solid State Circuits 39, 2438–2445 (2004) 3. Levine, P.M., Gong, P., Levicky, R., Shepard, K.L.: Active CMOS sensor array for electrochemical biomolecular detection. IEEE J. Solid State Circuits 43, 1859–1871 (2008) 4. Ayers, S., Berberian, K., Gillis, K.D., Lindau, M., Minch, B.A.: Post-CMOS fabrication of working electrodes for on-chip recordings of transmitter release. IEEE Trans. Biomed. Circuits Syst. 4, 86–92 (2010) 5. Warburg, E.: Ueber das Verhalten sogenannter unpolarisbarer Elektroden gegen Wechselstrom. Ann. Phys. Chem. 67, 493–499 (1899) 6. Lee, K., Lee, J., Sohn, M., Lee, B., Choi, S., Kim, S.K., Yoon, J., Cho, G.: One-chip electronic detection of DNA hybridization using precision impedance-based CMOS array sensor. Biosens. Bioelectron. 26, 1373–1379 (2010) 7. Stagni, C., Guiducci, C., Benini, L., Riccò, B., Carrara, S., Samorı́, B., Paulus, C., Schienle, M., Augustyniak, M., Thewes, R.: CMOS DNA sensor array with integrated A/D conversion based on label-free capacitance measurement. IEEE J. Solid State Circuits 41, 2956–2963 (2006) 8. Lu, M.S.-C., Chen, Y.C., Huang, P.C.: 5  5 CMOS capacitive sensor array for detection of the neurotransmitter dopamine. Biosens. Bioelectron. 26, 1093–1097 (2010) 9. Ghafar-Zadeh, E., Sawan, M., Chodavarapu, V.P., HosseiniNia, T.: Bacteria growth monitoring through a differential CMOS capacitive sensor. IEEE Trans. Biomed. Circuits Syst. 4, 232–238 (2010) 10. Li, J., Xue, M., Lu, Z., Zhang, Z., Feng, C., Chan, M.: A high-density conduction-based micro-DNA identification array fabricated with a CMOS compatible process. IEEE Trans. Electron Devices 50, 2165–2170 (2003) 11. Kim, D.S., Jeong, Y.T., Park, H.J., Shin, J.K., Choi, P., Lee, J.H., Lim, G.: An FET-type charge sensor for highly sensitive detection of DNA sequence. Biosens. Bioelectron. 20, 69–74 (2004) 12. Li, D.C., Yang, P.H., Lu, M.S.-C.: CMOS open-gate ionsensitive field-effect transistors for ultrasensitive dopamine detection. IEEE Trans. Electron Devices 57, 2761–2767 (2010) 13. Han, S., Yu, H., Murmann, B., Pourmand, N., Wang, S.X.: A high-density magnetoresistive biosensor array with drift-compensation mechanism. In: IEEE International Solid-State Circuits Conference (ISSCC) Digest of Technical Papers, pp. 168–169, San Francisco, 11–15 Feb 2007 14. Besse, P., Boero, G., Demierre, M., Pott, V., Popovic, R.: Detection of a single magnetic microbead using a miniaturized silicon Hall sensor. Appl. Phys. Lett. 80, 4199–4201 (2002) CMOS MEMS Fabrication Technologies 15. Wang, H., Chen, Y., Hassibi, A., Scherer, A., Hajimiri, A.: A frequency-shift CMOS magnetic biosensor array with single-bead sensitivity and no external magnet. In: IEEE International Solid-State Circuits Conference (ISSCC) Digest of Technical Papers, pp. 438–439 (2009) 16. Sun, N., Liu, Y., Lee, H., Weissleder, R., Ham, D.: CMOS RF biosensor utilizing nuclear magnetic resonance. IEEE J. Solid State Circuits 44, 1629–1643 (2009) 17. Eltoukhy, H., Salama, K., El Gamal, A.: A 0.18-mm CMOS bioluminescence detection lab-on-chip. IEEE J. Solid State Circuits 41, 651–662 (2006) 18. Huang, T.D., Sorgenfrei, S., Gong, P., Levicky, R., Shepard, K.L.: A 0.18-mm CMOS array sensor for integrated timeresolved fluorescence detection. IEEE J. Solid State Circuits 44, 1644–1654 (2009) 19. Xu, C., Li, J., Wang, Y., Cheng, L., Lu, Z., Chan, M.: A CMOS-compatible DNA microarray using optical detection together with a highly sensitive nanometallic particle protocol. IEEE Electron Device Lett. 26, 240–242 (2005) 20. Fritz, J., Baller, M.K., Lang, H.P., Rothuizen, H., Vettiger, untherodt, H.-J., Gerber, Ch., Gimzewski, P., Meyer, E., G€ J.K.: Translating biomolecular recognition into nanomechanics. Science 288, 316–318 (2000) 21. Shekhawat, G., Tark, S.H., Dravid, V.P.: MOSFETembedded microcantilevers for measuring deflection in biomolecular sensors. Science 311, 1592–1595 (2006) CMOS MEMS Fabrication Technologies Hongwei Qu1 and Huikai Xie2 1 Department of Electrical and Computer Engineering, Oakland University, Rochester, MI, USA 2 Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL, USA Synonyms CMOS (complementary metal-oxide-semiconductor); CMOS-MEMS; Integration; MEMS (micro-electromechanical systems) Definition CMOS-MEMS are micromachined systems in which MEMS devices are integrated with CMOS circuitry on a single chip to enable miniaturization and performance improvement. CMOS-MEMS also refers to microfabrication technologies that are compatible with CMOS fabrication processes. 441 C Overview Microelectromechanical systems (MEMS) leverage semiconductor fabrication technologies to manufacture various miniature sensors and actuators. Due to their low cost and small size as well as their much improved reliability, MEMS devices have been widely used even in our daily life, e.g., MEMS accelerometers for automobiles’ airbags, MEMS gyroscopes for electronic stability program (ESP) in automobile braking systems, MEMS tire pressure sensors; digital micromirror device (DMD)-enabled portable projectors, MEMS inkjet printers, MEMS resonators as frequency references, etc. Moreover, smart cell phones are now equipped with MEMS gyroscopes and accelerometers for motion actuated functions. They are also installed with surfacemounted MEMS microphones for even smaller size. The worldwide MEMS market reached 6.5 billion US dollars in 2010 [1]. Continuous miniaturization, expanded functionalities, lower cost, and improved performance are the ultimate goals of MEMS. The nature of MEMS strongly suggests direct integration of mechanical structures with electronics whose fabrication is dominated by CMOS technologies. In the last couple of decades, great efforts have been made in the integration of MEMS structures with ICs on a single CMOS substrate. The pioneering work for CMOSMEMS transducers was done by H. Baltes and his coworkers at the Swiss Federal Institute of Technology Zurich (ETH) [2]. They employed both wet bulk silicon micromachining and surface micromachining techniques in the fabrication of integrated CMOSMEMS devices. With the great advances in IC and MEMS technologies, the current focus of CMOS-MEMS integration technology is on the modification and standardization of CMOS technology to accommodate MEMS technology. One of the best-known commercial CMOS-MEMS devices is the digital micromirror device (DMD) manufactured by Texas Instruments. Recently, some CMOS foundries, such as TSMC, X-Fab and Global Foundries, have begun to offer CMOS-MEMS services for research and product developments. This entry summarizes a variety of CMOS-MEMS technologies and devices that have been developed. Particular materials needed in associated CMOS-MEMS will also be introduced. Typical MEMS devices, including inertial sensors, resonators, C C 442 and actuators, are exemplified in featuring the respective technologies. Classification of CMOS-MEMS Technologies MEMS can be integrated with CMOS electronics in many different ways. One common way to categorize CMOS-MEMS technologies is from the perspective of manufacturing processes. Based on the process sequence, CMOS-MEMS technologies can be classified into three categories: Pre-CMOS, Intra-CMOS, and PostCMOS [3]. Due to its popularity and accessibility, postCMOS will be described in more detail in this entry. Pre-CMOS It is widely accepted that pre-CMOS technologies are represented by the modular integration process originally developed at Sandia National Laboratories. As suggested by the name, in pre-CMOS technology, MEMS structures are pre-defined and embedded in a recess in silicon wafer; and the recess is then filled with oxide or other dielectrics. The wafer is then planarized prior to the following process steps for CMOS electronics [4]. In this “MEMS first” process, although MEMS structures are pre-defined, a wet etch after the completion of the standard CMOS processes is required to release the pre-defined MEMS structures. Due to the involvement of photolithography process needed for patterning the MEMS in the recess, the thickness of the MEMS structures is constrained by the lithographical limit. Other methods for the formation of MEMS structures in pre-CMOS technologies, including wafer bonding and thinning for epitaxial and SOI wafers in which MEMS are prefabricated, have also been reported in the fabrication of a variety of MEMS devices. Inter-CMOS In early 1990s, Analog Devices, Inc (ADI) specifically developed a MEMS technology based on its BiCMOS process. This “iMEMS” technology, originally dedicated to manufacturing CMOS-MEMS accelerometers and gyroscopes, is an intermediate-CMOS-MEMS, or Inter-CMOS-MEMS technology in which the CMOS CMOS MEMS Fabrication Technologies process steps are mixed with additional polysilicon thin-film deposition and micromachining steps to form the sensor structures [5]. Infinion’s pressure sensors are also fabricated using this kind of Inter-CMOS-MEMS technology. To reduce the residual stress in structural polysilicon, high temperature annealing is normally required in the InterCMOS-MEMS, which could pose a potential risk to CMOS interconnect and active layers. Thus, the thermal budget should be carefully designed. Moreover, it is almost impractical to perform intermediate CMOS and MEMS processed in separate foundries due to possible contaminations in the wafer transfer and processes. Therefore, a dedicated foundry used for both CMOS and MEMS is necessary for InterCMOS-MEMS technology, which may not take full advantages of mainstream technologies in either area. Limitations of Pre- and Inter-CMOS-MEMS Since surface micromachining and polysilicon are typically used, most of the Pre- and Inter-CMOSMEMS technologies suffer from the limitations of thin-film structural materials. (1) Structural curling and cost associated with stress compensation: Due to the residual stress in the deposited thin films in the device, polysilicon structures often exhibit curling after release, resulting in reduced sensitivity, lower mechanical robustness, and increased temperature dependence. Although stress compensation can be realized via multiple controlled process steps, the associated cost is quite high. (2) Small size and/or mass: The curling of thin-film structures in turn limits the size of the overall microstructure. The mass is further reduced due to the small thickness of thin-film polysilicon. For inertial sensors, the smaller the structure mass, the lower performance of the device. (3) Parasitics: In a surface micromachined polysilicon accelerometer, depending on polysilicon wiring path, the parasitic impedance may considerably lower the static and dynamic performance of the device. (4) Cost and suboptimal processes of the dedicated foundry needed: The dedicated foundry combining CMOS and MEMS fabrications needed for pre- and inter-CMOS-MEMS is normally expensive and suboptimal for either fabrication. It is against the modern trends in which flexible accessibility of optimal and cost-effective processes are preferable. CMOS MEMS Fabrication Technologies 443 C CMOS MEMS Fabrication Technologies, Table 1 Representative thin-film deposition additive post-CMOS-MEMS technologies Authors and references Hornbeck [7] Yun et al. [8] Franke et al. [9] Sedky, Van Hoof et al. [10] Institute Texas instruments UC-Berkeley UC-Berkeley IMEC Structural material Al Polysilicon Poly-SiGe Poly-SiGe Post-CMOS Post-CMOS-MEMS refer to the CMOS-MEMS processes in which all MEMS process steps are performed after the completion of the CMOS fabrication. The advantages of post-CMOS-MEMS over Pre- and Inter-CMOS-MEMS include process flexibility and accessibility and low cost. In contrast to both PreCMOS-MEMS and Inter-CMOS-MEMS, for PostCMOS-MEMS technology, the fabrications of CMOS circuitry and MEMS structures are performed independently. The flexibility of foundry access makes it possible to take advantages of both advanced CMOS technologies and optimal MEMS fabrication. This is particularly attractive to research community in exploration of state-of-the-art in MEMS. Some design rules may need to be changed to accommodate MEMS structure design in the CMOS design stage. Meanwhile, post-CMOS microfabrication should be carefully designed, particularly considering the thermal budget, so as not to affect the on-chip CMOS electronics. According to how MEMS structures are formed, post-CMOS-MEMS technologies fall into two categories: additive and subtractive. In additive post-CMOS-MEMS, structural materials are deposited on a CMOS substrate. In subtractive post-CMOSMEMS, MEMS structures are created by selectively etching CMOS layers. Apparently, additive post-CMOS-MEMS methods require more stringent material compatibility with the CMOS technologies used. Thus, they are less utilized than subtractive postCMOS-MEMS. The following introduction will focus more on subtractive post-CMOS-MEMS. Additive MEMS Structures on CMOS Substrate In additive post-CMOS-MEMS, metals, dielectrics, or polymers are deposited and patterned to form MEMS structures normally on top of the CMOS layers. Some commercial MEMS products are fabricated using additive post-CMOS-MEMS approaches. In Sacrificial material Photoresist SiO2 Ge or SiO2 Ge Interconnect material Al W/TiN Al Al Year 1989 (invented in 1987) 1992 1999 1998 this category, the best-known product is probably the digital mirror device (DMD), the core of the digital light processing (DLP) technology developed by Texas Instruments. In a DMD, tilting mirror plates and their driving electrodes are fabricated directly on top of CMOS circuits. Three sputtered aluminum layers are used to form the top mirror plate and the two parallel-plate electrodes for electrostatic actuation, respectively. The driving electrodes are addressed via CMOS memory cell. To release the mirror plate and top electrodes in the post-CMOSMEMS fabrication of the mirrors, deep-UV hardened photoresist is used as the sacrificial layer. In some circumstances where CMOS protection is well designed, electroplating can also be used to grow microstructures on top of CMOS electronics. Other structural and sacrificial materials, such as polycrystalline SiGe and Ge, have been used to create CMOS-MEMS as well [6]. Additive post-CMOS-MEMS processes, along with their respective materials, are summarized in the following table (Table 1). In addition to the approaches of forming MEMS structures on top of the CMOS substrate by thin-film deposition, wafer bonding provides another method to directly integrate MEMS structures on CMOS substrate [11]. For instance, a prefabricated polysilicon capacitive acceleration sensor wafer is bonded to a CMOS wafer with read-out electronics. In a wafer-bonded piezoresistive accelerometer, the micromachined bulk silicon proof mass was sandwiched by a bottom glass cap and a top CMOS chip on which the conditioning circuit was integrated. SOI-CMOS-MEMS has also been attempted for monolithic integration of electronics with bulk MEMS structures. With 3-dimensional packaging enabled by technological breakthroughs such as through-silicon vias (TSVs), this integration method promises to be further developed in manufacturing complex microsystems. MEMS suppliers, including STMicroelectronics and InvenSense, C C 444 have adopted wafer-to-wafer or chip-to-wafer bonding CMOS-MEMS integration. Subtractive Post-CMOS-MEMS In these devices, MEMS structures are formed from built-in CMOS thin-film stacks including metals and SiO2, or from the silicon substrate. These materials are patterned and removed partially by wet or dry etching to release the MEMS structures. This section describes the thin-film and bulk CMOS-MEMS formed by such subtractive processes. Subtractive CMOS-MEMS by Wet Etching The first generation of CMOS-MEMS sensors was fabricated using a post-CMOS subtractive process in which silicon substrate was completely or partially removed using a wet etching method, leaving behind thin-film or bulk MEMS structures [2]. For thermal sensors in which beams or membranes consisting of dielectric layers, the substrate silicon is normally etched away completely to obtain thermally isolated structures. The silicon dioxide membrane can act as an intrinsic etch stop layer in backside silicon anisotropic wet etch using KOH, ethylene diamine-pyrocatechol (EDP), or Tetramethylammonium hydroxide (TMAH). A high-Q RF MEMS filter with an inter-metal dielectric layer as structural material was reported by IBM. A medical tactile sensor array was also reported in which the aluminum sacrificial layer was etched from the backside of the wafer after the CMOS substrate was etched through [12]. The silicon substrate can also be included in the MEMS structures using a wet etch process. The first method is to perform a time-controlled backside etch with a well-calibrated etching rate. A uniform single crystal silicon membrane with a desired thickness can be created. This method has been widely used in industry for fabrication of large volume products such as integrated pressure sensors. In cases where the silicon membrane thickness is not critical, even mechanical processing such as grinding can be used to create the backside cavity. The second method involves the utilization of an automatic etch stop technique to create silicon membranes or MEMS structures. In this case, an anisotropic etch stops at the electrochemically biased p-n junction formed between the n-well and p-type substrate in CMOS [13]. Although the electrochemical electrode design and implementation are complicated, CMOS MEMS Fabrication Technologies this process can be specifically used in the fabrication of highly sensitive pressure/force and thermal sensors. The anisotropic etch stop can also occur at highly doped p regions in the substrate. This method has been used in fabrication of many suspended structures including neural probes [14]. Note that the p++ doping process may not be available in a standard CMOS process. In the case where only a small portion of the silicon substrate needs to be removed to reduce the circuit-substrate coupling, a wet silicon etch can be performed from the front side. In wet silicon etching, either silicon nitride or additional polymers or both can be used to protect the front CMOS and pads. Polymers sensitive to analytes can be coated on finished CMOS-MEMS structures for chemical and biological sensing. For example, the first CMOS-MEMS electronic nose was demonstrated by forming polymer-coated CMOS thin-film cantilevers on a CMOS chip [15]. Table 2 summarizes some representative devices that were fabricated using wet etching when this technology was dominant in post-CMOS micromachining. Bibliographies of these efforts can be found in the above citations in this section. Subtractive Post-CMOS-MEMS by Dry Etching Dry etching processes have quickly become popular in microfabrication for both MEMS research and industry. Particularly, the deep reactive ion etching (DRIE) technology, or Bosch process, has revolutionized subtractive post-CMOS microfabrication [23]. This section describes thin-film and bulk CMOS-MEMS devices fabricated using dry etching processes. Most dry etching processes are based on plasma processes, such as reactive ion etch (RIE) and DRIE, while etchants in the vapor phase can also be used for dry etching. For example, vapor XeF2 provides good isotropic etching of silicon, which has been used for releasing CMOS thin-film MEMS structures [24]. The combination of RIE and DRIE, performed from the front or back side, or both sides, has been used to fabricate a large variety of CMOS-MEMS devices. Depending on the structural materials and etching methods employed, subtractive post-CMOS can be divided into two types: thin-film processes and bulk processes. Thin-Film Post-CMOS-MEMS Dry Processes In thin-film processes, structural materials are composed of CMOS thin films. Figure 1 depicts the process flow CMOS MEMS Fabrication Technologies 445 C CMOS MEMS Fabrication Technologies, Table 2 CMOS-MEMS devices enabled by subtractive process wet etching Authors and references Institutions Wise et al. [16] U. of Michigan U. of Michigan Wise et al. [14] Year 1979 1985 Device Pressure sensor Neuron probe array Yoon and Wise [17] Baltes et al. [2] Haberli et al. [18] U. of Michigan ETH Zurich ETH Zurich 1990 1996 1996 Mass flow sensor Thermal capacitor Pressure sensor Schneider et al. [19] ETH Zurich 1997 Thermal sensor 2000 AFM probe 2001 2006 Infrared imager RF MEMS U. of Neuchatel, ETH Zurich Schaufelbuhl et al. [21] ETH Zurich U. of Barcelona Verd et al. [22] Akiyama et al. [20] of a thin-film post-CMOS-MEMS process, which was originally developed at Carnegie Mellon University [25]. A sequenced process consisting of an isotropic SiO2 etching, a silicon DRIE and an isotropic Si RIE releases the MEMS structure. In these process steps, the top metal layer acts as a mask to form the MEMS structures and to protect the CMOS circuitry, as seen in Fig. 1a, b. Anisotropic and isotropic silicon etching complete the process flow, as seen in Fig. 1c, d. Various inertial sensors have been fabricated using this thin-film technology. In all these inertial sensors, mechanical springs and proof masses are formed by the multiple-layer CMOS stacks consisting of SiO2 and metals. The sensing capacitance is formed from sidewall capacitance between comb fingers. The multiple CMOS metal layers inside the comb fingers and other mechanical structures allow very flexible electrical wiring, facilitating different sensing schemes including vertical comb-drive sensing. Akustica, Inc. has commercialized digital microphones using a modified version of this process. Other sensors have also been demonstrated using similar thin-film technology, such as humidity sensors and chemical sensors. All these thin-film post-CMOS dry etching processes have excellent CMOS compatibility and accessibility as well as design flexibility. However, a major issue is the large vertical curling and lateral buckling of suspended MEMS structures, which is caused by the residual stress in the stacked thin-film CMOS layers. Although structural curling can be tolerated for some small devices such as RF MEMS, for devices such as inertial sensors that need relatively Device structure Silicon diaphragm CMOS thin films and Si substrate CMOS thin films CMOS thin films CMOS thin films CMOS thin films and suspended substrate CMOS thin films and silicon substrate CMOS thin films CMOS thin films Etching method Backside EDP etching EDP etching, p++ etching stop Backside, SiO2 etching stop Front side etching Front side etching of aluminum as sacrificial layer PN junction electrochemical etch stop N-well electrochemical etch stop Backside KOH Front side SiO2 etching large size, the impact of structural curling can be severe. Bulk CMOS-MEMS Dry Process In order to overcome the structural curling and to increase the mass, flatness, and robustness of MEMS structures, single crystal silicon (SCS) may be included underneath the CMOS thin-film stacks. The SCS silicon structures are formed directly from the silicon substrate using DRIE. Figure 2 illustrates the process flow [26]. The process starts with the backside silicon DRIE to define the MEMS structure thickness by leaving a 10–100 mm-thick SCS membrane (Fig. 2a). Next, the same anisotropic SiO2 etch as in the thin-film process is performed on the front side of wafer (chip) to expose the SCS to be removed (Fig. 2b). The following step differs from the thin-film process in that an anisotropic DRIE, instead of isotropic etch, finalizes the structure release by etching through the remaining SCS diaphragm, as shown in Fig. 2c. With the SCS underneath the CMOS interconnect layers included, large and flat MEMS microstructures can be obtained. If necessary, an optional time-controlled isotropic silicon etch can be added. This step will undercut the SCS underneath the designed narrow CMOS stacks to create thin-film structures (Fig. 2d). This step is particularly useful for fabricating capacitive inertial sensors. It can be used to form the electrical isolation structures between sensing fingers and silicon substrate. The DRIE CMOS-MEMS technology has shown great advantages in the fabrication of relatively large MEMS devices such as micromirrors. Large flat mirror C C 446 CMOS MEMS Fabrication Technologies CMOS MEMS Fabrication Technologies, Fig. 1 CMU post-CMOS fabrication process for MEMS structures made of CMOS thin films a CMOS region MEMS region SiO2 Si substrate Polysilicon Metal 4 b c d MEMS structures Metal 3 Metal 2 Metal 1 a CMOS MEMS Metal 4 Surface SiO2 Metal 3 b Metal 2 Metal 1 SCS membrane c CMOS d SCS CMOS MEMS Fabrication Technologies, Fig. 2 DRIE bulk CMOS-MEMS process flow can be obtained by including portion of silicon substrate underneath the aluminum mirror surface. A CMOS-MEMS gyroscope with a low noise floor of 0.02 degree/s/sqrtHZ has also been demonstrated using this technology [27]. By attaching SCS underneath the CMOS stack comb fingers, the sensing capacitance can be considerably increased for larger signal-to-noise ratio (SNR). Although CMOS thin films are still used in some microstructures for electrical isolation, the length of the thin-film portion is minimal to reduce the temperature effect. Compared to the thin-film dry CMOS-MEMS process, a backside silicon DRIE step is added. This requires an additional backside lithography step to define the region for MEMS structures. The maximum thickness of the MEMS structures is limited by the aspect ratio that the silicon DRIE can achieve. An Improved Bulk CMOS-MEMS Process The bulk CMOS-MEMS process depicted in Fig. 2 is useful in fabrication of many devices where SCS structures are desired to improve both mechanical and electrical performance of the devices. However, for some devices, very fine structures are formed in step (c) in Fig. 2; so the damage caused by the step (d) to these CMOS MEMS Fabrication Technologies CMOS MEMS Fabrication Technologies, Fig. 3 The modified bulk CMOS-MEMS process for separate etching of CMOS beams and SCS microstructures. Backside photoresist coating effectively reduces temperature in the device release, reducing deleterious non-uniform etching 447 C d a Silicon diaphragm Cavity Photoresist for external thermal path b e c f fine structures may be severe. This is particularly true for the fabrication of capacitive inertial sensors where narrow-gap sensing comb fingers are needed. For instance, in performing the isotropic silicon undercut to form the narrow CMOS beams for electrical isolation, the SCS in the comb fingers is also undercut. The sensing gap increases due to the undesired undercut greatly reduce sensitivity and signal-to-noise ratio (SNR). If the undercut occurs in mechanical structures such as suspension springs, the characteristics of the device will also be affected. Another issue is related to the thermal effect in the plasma etch for the SCS undercut. Upon completion of the silicon undercut, the greatly reduced thermal conductance from the isolated structure to the substrate can cause a temperature rise on the released structures. Slight over-etch is often necessary to accommodate process variations, but this will generate a large temperature rise on the suspended structures which in turn dramatically increases the SCS etching rate, resulting in uncontrollable and damaging results [28]. A modified dry bulk CMOS-MEMS process has been demonstrated to effectively address the issues caused by the undesired SCS undercut [28]. In the refined process illustrated in Fig. 3, the etching of the CMOS connection beams is performed separately from the etching of the microstructures where SCS is needed. The top metal layer is specifically used to define the connection beams. After their formation, the top metal layer is removed using a plasma or a wet etch. Then other microstructures are exposed after a SiO2 etch. The direct etch-through of the remaining silicon on the microstructures will complete the release process. To reduce the thermal effect described above, a thick photoresist layer is patterned on the backside of the cavity. In the release step, the applied photoresist provides a thermal path that reduces the temperature rise on the etched-through structures. The removal of the photoresist using O2 plasma etching completes the entire microfabrication process. Owing to the monolithic integration and large proof mass enabled by the inclusion of SCS, bulk CMOS-MEMS inertial sensors have demonstrated better performance than their thin-film counterparts [29]. The photoresist coating can also be replaced by sputtering a layer of metal such as aluminum. Combined Wet/Dry Processes In addition to the integration methods described above, efforts have been continuously made to integrate CMOS with MEMS using the combination of different microfabrication technologies. By combing silicon anisotropic wet etch with DRIE, some sophisticated surface and bulk MEMS structures such as bridges and cantilever arrays can be created. A multi-sensor system was demonstrated using a combined etch process [30]. In the accelerometers reported in [31], isotropic wet C C 448 etching is used to remove metal layers in CMOS thin stacks to create parallel-plate-like vertical capacitors for gap-closing sensing. A silicon RIE follows to release the MEMS devices and break the coupling between the sensing thin films and the substrate. Sensitivities are largely increased with the gap-closing sensing compared to comb-finger sensing. Summary CMOS-MEMS technologies have been placed in preCMOS, intra-CMOS, and post-CMOS categories. Both pre-CMOS and intra-CMOS have issues such as dedicated foundries with suboptimal and less cost-effective processes. So it is normally impractical for academic research community to access these dedicated facilities. Post-CMOS provides excellent CMOS compatibility, foundry accessibility, and design flexibility, and the cost is also relatively low. While the process standardization and industrialization of CMOS-MEMS technologies are in continuous progress, innovative processing technologies have opened up new pathways for integration. Wafer bonding–based integration has blurred the boundary between pre- and post-CMOS-MEMS integrations. SOI-CMOS-MEMS have also been aggressively explored. The technologies involved in this new exploration have emerged as enabling means for three-dimensional and systems-in-package integrations. More recently, technologies to co-fabricate many subsystems including nano-systems are being pursued enthusiastically. CMOS-MEMS integration will continually evolve with the emergence of new fabrication technologies and new materials. While some companies have demonstrated promising CMOS-MEMS products, more joint efforts from research community and industries are needed for new process transfer and standardization to allow large volume fabrication of new products. Cross-References ▶ Integration ▶ MEMS ▶ Nanofabrication ▶ Sensors CMOS MEMS Fabrication Technologies References 1. Johnson, R.C.: MEMS market projected to hit double-digit growth, again. www.eetimes.com. Accessed 14 July 2011 2. Baltes, H., Paul, O., et al.: IC MEMS microtransducers. In: Proceedings of International Electronic Device Meeting IEDM ’96, San Francisco, pp. 521–524 (1996) 3. Baltes, H., Brand, O., Fedder, G.K., Hierold, C., Korvink, J. G., Tabata, O.: CMOS-MEMS: Advanced Micro and Nanosys, 1st edn. Wiley, Weinheim (2005) 4. Smith, J.H., Montague, S., Sniegowski, J.J., Murray, J.R., McWhorter, P.J., Smith J.H.: Embedded micromechanical devices for the monolithic integration of MEMS with CMOS. In: Proceedings of International Electronic Device Meeting, IEDM ’95, Washington D.C., pp. 609– 612 (1995) 5. Kuehnel, W., Sherman, S.: A surface micromachined silicon accelerometer with on-chip detection circuitry. Sens. Actu. A Phys. 45, 7–16 (1994) 6. Franke, A.E., Heck, J.M., King, T.J., Howe, R.T.: Polycrystalline silicon-germanium films for integrated microsystems. J. Micromech. Sys. 12, 160–171 (2003) 7. Hornbeck, L.: Deformable-mirror spatial light modulators and applications. SPIE Crit. Rev. 1150, 86–102 (1989) 8. Yun, W., Howe, R.T., Gray, P.R.: Surface micromachined, digitally force-balanced accelerometer with integrated CMOS detection circuitry. Technical Digest of Solid State Sensors and Actuators Workshop, Hilton Head Island, pp. 126–131 (1992) 9. Franke, A.E., Bilic, D., Chang, D.T., Jones, P.T., King, T.J., Howe, R.T., Johnson, G.C.: Post-CMOS integration of germanium microstructures. In: The 12th IEEE International Conference on Micro Electro Mechanical Systems, Orlando, pp. 630–637 (1999) 10. Sedky, S., Fiorini, P., Caymax, M., Loreti, S., Baert, K., Hermans, L., Mertens, R.: Structural and mechanical properties of polycrystalline silicon germanium for micromachining applications. J. MEMS 7, 365–372 (1998) 11. Fedder, G.K., Howe, R.T., Liu, T.J., Quevy, E.P.: Technologies for cofabricating MEMS and electronics. Proc. IEEE 96, 306–322 (2008) 12. Salo, T., Vancura, T., Brand, O., Baltes, H.: CMOS-based sealed membranes for medical tactile sensor arrays. In: Proceedings of International Conference on Micro Electro Mechanical Systems, Kyoto, pp. 590–593 (2003) 13. Muller, T., Brandl, M., Brand, O., Baltes, H.T.: An industrial CMOS process family adapted for the fabrication of smart silicon sensors. Sen. Actu. A Phys. 84, 126–133 (2000) 14. Najafi, K., Wise, K.D., Mochizuki, T.K.: A high-yield IC-compatible multichannel recording array. IEEE T. Electron. Dev. 32, 1206–1211 (1985) 15. Baltes, H., Koll, A., Lange, D.H.: The CMOS MEMS nosefact or fiction? In: Proceedings of IEEE International Symposium on Industrial Electronics ISIE ’97, Guimaraes, vol. 1, pp. SS152–SS157 (1997) 16. Borky, J.M., Wise, K.D.: Integrated signal conditioning for silicon pressure sensors. IEEE T. Electron. Dev. ED-27, 927–930 (1979) CMOS-CNT Integration 17. Yoon, E., Wise, K.D.: A multi-element monolithic mass flowmeter with on-chip CMOS readout electronics. Technical Digest of Solid State Sensors and Actuators Workshop, Hilton Head, pp. 161–164 (1990) 18. Haberli, A., Paul, O., Malcovati, P., Faccio, M., Maloberti, F., Baltes, H.: CMOS integration of a thermal pressure sensor system. In: IEEE International Symposium on Circuits and Systems, ISCAS ’96, Atlanta, vol. 1, pp. 377–380 (1996) 19. Schneider, M., Muller, T., Haberli, A., Hornung, M., Baltes, H.: Integrated micromachined decoupled CMOS chip on chip. In: Proceedings of 10th IEEE International Workshop on MEMS, Nagoya, pp. 512–517 (1997) 20. Akiyama, T., Akiyama, T., Staufer, U., de Rooij, N.F., Lange, D., Hagleitner, C., Brand, O., Baltes, H., Tonin, A., Hidber, H.R.: Integrated atomic force microscopy array probe with metal-oxide-semiconductor field effect transistor stress sensor, thermal bimorph actuator, and on-chip complementary metal-oxide-semiconductor electronics. J. Vac. Sci. Technol. B 18, 2669–2675 (2000) 21. Schaufelbuhl, A., Schneeberger, N., Munch, U., Waelti, M., Paul, O., Brand, O., Baltes, H., Menolfi, C., Huang, Q., Doering, E., Loepfe, M.: Uncooled low-cost thermal imager based on micromachined CMOS integrated sensor array. J. MEMS 10, 503–510 (2001) 22. Verd, J., Uranga, A., Teva, J., Lopez, J.L., Torres, F., Esteve, J., Abadal, G., Perez-Murano, F., Barniol, N.: Integrated CMOS-MEMS with on-chip readout electronics for highfrequency applications. IEEE Electron. Dev. Lett. 27, 495–497 (2006) 23. Laermer, F., Schilp, A.: Method of anisotropically etching silicon. US Patent 5,501,893, Robert Bosch Gmbh (1992) 24. Kruglick, E.J.J., Warneke, B.A., Pister, K.S.: CMOS 3-axis accelerometers with integrated amplifier. In: Proceedings of International Conference on Micro Electro Mechanical System, MEMS-98, Heidelberg, pp. 631–636 (1998) 25. Fedder, G.K., Santhanam, S., Reed, M.L., Eagle, S.C., Guillou, D.F., Lu, M.S.C., Carley, L.R.: Laminated highaspect-ratio microstructures in a conventional CMOS process. Proceedings of International Conference on Micro Electro Mechanical Systems, MEMS-96, San Diego, CA, pp. 13–18 (1996) 26. Xie, H., Erdmann, L., Zhu, X., Gabriel, K.J., Fedder, G.K.: Post-CMOS processing for high-aspect-ratio integrated silicon microstructures. J. Microelectromech. Syst. 11, 93–101 (2002) 27. Xie, H., Fedder, G.K.: Fabrication, characterization, and analysis of a DRIE CMOS-MEMS gyroscope. IEEE Sens. J. 3, 622–631 (2003) 28. Qu, H., Xie, H.: Process development for CMOS-MEMS sensors with robust electrically isolated bulk silicon microstructures. J. Microelectromech. Syst. 16, 1152–1161 (2007) 29. Qu, H., Fang, D., Xie, H.: A monolithic CMOS-MEMS 3-axis accelerometer with a low-noise, low-power dual-chopper amplifier. IEEE Sens. J. 8, 1511–1518 (2008) 30. Hagleitner, C., Lange, D., Hierlemann, A., Brand, O., Baltes, H.: CMOS single-chip gas detection system comprising capacitive, calorimetric and mass-sensitive microsensors. IEEE J. Solid-St. Circc 37, 1867–1878 (2002) 449 C 31. Tsai, M.H., Sun, C.M., Liu, Y.C., Wang, C.W., Fang, W.L.: Design and application of a metal wet-etching post-process for the improvement of CMOS-MEMS capacitive sensors. J. Micromech. Microeng. 19, 105017 (2009) CMOS-CNT Integration Huikai Xie and Ying Zhou Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL, USA Synonyms Monolithic integration of carbon nanotubes Definition Carbon nanotubes can be directly grown on CMOS substrate without degrading the performance of CMOS electronics. Introduction With numerous outstanding electrical, mechanical, and chemical properties, carbon nanotubes (CNTs) have been explored for various applications with great success. As CMOS circuits possess powerful interfacing, signal amplification, conditioning, and processing capabilities, it is also highly desired to integrate CNTs with CMOS. CNTs may be either used as part of CMOS electronics or as sensing elements to form functioning nano-electromechanical systems (NEMS). Figure 1 shows a nanotube random-access memory (NRAM) which uses CNT ribbons as switches [1]. Many sensors based on CMOS-CNT hybrid systems have also been demonstrated, including mechanical, thermal, and chemical sensors [3, 4]. The integration of CMOS circuits with CNT sensors can increase signal-to-noise ratio and dynamic range, lower power consumption, and provide various controls and automations. Other efforts have been made to use multiwall carbon nanotubes (MWNTs) as CMOS interconnect for high frequency applications [5], or to apply CNT-based nano-electromechanical switches C C 450 CMOS-CNT Integration, Fig. 1 The structure of NRAM at (a) on and (b) off states [2] CMOS-CNT Integration a b CNT – + Interconnects ON for leakage reduction in CMOS logic and memory circuits [6]. However, monolithic integration of CMOS and CNTs is still very challenging. Most CMOS-CNT systems have been realized either by a two-chip solution or low-throughput CNT manipulations. In this entry, CMOS-CNT integration approaches are reviewed, with a particular focus on a localized heating CNT synthesis method that can grow CNTs on foundry CMOS. – + OFF growth temperature, but it is still too high for direct CNT growth on CMOS substrates. In addition, during or after CNT growth, electrical contacts need to be formed for functional CNT-based devices. It is reported that Mo provides good ohmic contacts with nanotubes and shows excellent conductivity after growth, with resistance ranging from 20 kO to 1 MO per tube [11]. Several other metals, such as palladium, gold, titanium, tantalum, and tungsten, have also been investigated as possible electrode materials. CNT Synthesis CMOS-CNT Integration There are three main methods for carbon nanotube synthesis: arc-discharge [7], laser ablation [8], and chemical vapor deposition (CVD) [9]. The first two methods involve evaporation of solid-state carbon precursors and condensation of carbon atoms to form nanotubes, where high annealing temperature, typically over 1,000 C, is required to remove defects and thus produce high-quality nanotubes. However, they tend to produce a mixture of nanotubes and other by-products such as catalytic metals, so the nanotubes must be selectively separated from the by-products. This requires post-growth purification and manipulation. In contrast, the CVD method employs a hydrocarbon gas as the carbon source and involves heating metal catalysts in a tube furnace to synthesize nanotubes. Nanotubes can grow either on the top (tip growth) or from the bottom (base growth). The diameters and locations of the grown CNTs can be controlled via catalyst size and catalyst patterning, and the orientation can be guided by an external electric field. Suitable catalysts that have been reported include Fe, Co, Mo, and Ni [10]. Compared to the arc-discharge and laser ablation methods, CVD uses much lower To integrate CNT on CMOS, there are several factors that must be taken into account: temperature budget, material compatibility, CNT type, CNT quality, and contamination. Depending on when CNTs are made, CMOS-CNT integration technology can be categorized as follows: • Pre-CMOS: CMOS processes will be performed after CNTs are synthesized in place • Intra-CMOS: CNT growth steps are inserted into CMOS fabrication steps • Post-CMOS: CNTs are introduced after all CMOS processes have been done For pre-CMOS, CNTs must go through standard CMOS process steps. This is very difficult to realize. There are temperature constraints, material compatibility, and contamination issues. There is no report about pre-CMOS CNT integration yet. For intraCMOS, CNTs will be introduced at a later stage in the CMOS fabrication sequence, so it is easier to protect CNTs than in the pre-CMOS case. But temperature and contamination issues still must be considered. Post-CMOS, on the hand, completely eliminates CMOS-CNT Integration 451 C C CMOS-CNT Integration, Fig. 2 SEM images of vertically aligned CNFs grown by PECVD deposition at (a) 500 C, (b) 270 C, and (c) 120 C (scale bars: (a) and (b) 1 mm and (c) 500 nm) [14] CMOS contamination issues. It has potential to achieve mass production and low cost, but the temperature remains a limiting factor. Intra-CMOS CNT Integration Intra-CMOS (High Temperature CNTs) Thermal CVD has been used to grow CNTs directly on CMOS substrate. For example, Tseng et al. demonstrated, for the first time, a process that monolithically integrates SWNTs with n-channel metal oxide semiconductor (NMOS) FET in a CVD furnace at 875 C [12]. However, the high synthesis temperature (typically 800–1,000 C for SWNT growth) may damage the aluminum metallization layers and change the characteristics of the on-chip transistors as well. Ghavanini et al. assessed the deterioration level of CMOS transistors with certain CNT CVD synthesis conditions applied, and they reported that one PMOS transistor lost its functions after the thermal CVD treatment (610 C, 22 min) [13]. As a result, the integrated circuits in Tseng’s thermal CVD CNT synthesis can only consist of NMOS and use n+ polysilicon and molybdenum as interconnects, which make it incompatible with foundry CMOS processes. Intra-CMOS (Low Temperature CNTs) Some other attempts have been made to develop low temperature growth using various CVD methods. Hofmann et al. reported vertically aligned carbon nanotubes grown at temperature as low as 120 C by plasmaenhanced chemical vapor deposition (PECVD) [14]. However, the decrease in growth temperature jeopardizes both the quality and yield of the CNTs, as shown in Fig. 2. The synthesized products are actually defectrich carbon nanofibers rather than MWNTs or SWNTs. Intra-CMOS (Localized Heating) Localized heating: To accommodate both the high temperature requirement (800–1,000 C) for highquality SWNT synthesis and the temperature limitation of CMOS processing (<450 C), CNT synthesis based on localized heating has drawn great interest recently. Englander et al. demonstrated, for the first time, the localized synthesis of silicon nanowires and carbon nanotubes based on resistive heating [15]. The fabrication processes are shown in Fig. 3. Operated inside a room temperature chamber, the suspended micro-electromechanical system (MEMS) structures serve as resistive heaters to provide high temperature at predefined regions for optimal nanotube growth, leaving the rest of the chip area at low temperature. Using the localized heating concept, direct integration of nanotubes at specific areas can be potentially achieved in a CMOS compatible manner, and there is no need for additional assembly steps. However, the devices typically have large sizes and their fabrication processes are not fully compatible with the standard foundry CMOS processes. Although this concept has solved the temperature incompatibility problem between CNT synthesis and CMOS circuit protection, the fabrication processes of microheater structures still have to be well designed to fit into standard CMOS foundry processes and the resistor materials must be selected to meet the CMOS compatibility criteria. Using the localized heating technique described above, on-chip growth using CMOS micro-hotplates was demonstrated by Haque et al. [16]. As shown in Fig. 4, tungsten was used for both the micro-hotplates (as the heating source) and interdigitated electrodes for nanotubes contacts. MWNTs have been successfully synthesized on the membrane, and simultaneously C 452 CMOS-CNT Integration CMOS-CNT Integration, Fig. 3 Fabrication process and localized heating concept [15] a b Silicon Oxide Substrate c Etched Silicon d Oxide Etch e Catalyst – + g Vacuum Pump V f C2H4 or SiH4 V Vacuum Chamber @ Room Temperature MFC Gas Supply A Vacuum Feedthrough a Heat Spreading Layer CNT Growth Area b Hotplate Area Buried Silicon Silicon Dioxide Passivation Dioxide Silicomheat spreading layer Substrate Tungsten Heater Hotplate Area PMOS NMOS P+ N P+ N+ P N+ 150 μm CMOS Control Circuits CMOS Area CMOS-CNT Integration, Fig. 4 (a) Schematic of the cross-sectional layout of the chip. (b) Optical image of the device top view showing the tungsten interdigitated electrodes on top of the membranes, heater radius ¼ 75 mm, membrane radius ¼ 280 mm [16] connected to CMOS circuits through tungsten metallization. Although tungsten can survive the high temperature growth process, and has high connectivity and conductivity, Franklin et al. reported that no SWMTs were found to grow from catalyst particles on the tungsten electrodes, presumably due to the high catalytic activity of tungsten toward hydrocarbons [11]. Further, although the monolithic integration has been achieved, the utilization of tungsten, a refractory metal, as interconnect metal is limited in foundry CMOS, especially for mixed-signal CMOS processes. Moreover, this approach requires a backside bulk CMOS-CNT Integration 453 C 24 Vpp@ 500 kHz a – dielectrophoresis electrode (Au) 1 2 CNT 3 metal clamp (Au) 0.25 μm CMOS chip C 6 via (n) 5 via hole 4 Al wire Ti b Au CNT 5 μm via hole CMOS topmost metal (Al) CMOS-CNT Integration, Fig. 5 (a) Process flow to integrate MWNT interconnects on CMOS substrate. (b) SEM image of one MWNT interconnect (wire and via) [5] micromachining process and is limited to SOI CMOS substrates. Post-CMOS CNT Integration Post-CMOS (CNT Transfer and Assembly) To overcome the temperature limitation, one possible solution is to grow nanotubes at high temperature first and then transfer them to the desired locations on CMOS substrates at low temperature. However, handling, maneuvering, and integrating these nanostructures with CMOS chips/wafers to form a complete system are very challenging. In the early stage, an atomic force microscope (AFM) tip was used to manipulate and position nanotubes into a predetermined location under the guide of scanning electron microscope (SEM) imaging [17]. Although this nanorobotic manipulation realized precise control over both the type and location of CNTs, its low throughput makes large scale assembly prohibitive. Other post-growth CNT assembly methods include surface functionalization [18], liquid-crystalline processing [19], dielectrophoresis (DEP) [20], and large scale transfer of aligned nanotubes grown on quartz [21]. A 1 GHz CMOS circuit with CNT interconnects has been demonstrated using a DEP-assisted assembly technique [5]. The fabrication process flow and the assembled MWNT interconnect are shown in Fig. 5. The DEP process provides the capability of precisely positioning the nanotubes in a noncontact manner, which minimizes the parasitic capacitances and allows the circuits to operate at more than 1 GHz. However, to immobilize the DEP-trapped CNTs in place and to improve the electrical contact between CNTs and the electrodes, metal clamps must be selectively deposited at both ends of the CNTs (Fig. 5a, step 3). The process complexity and low yield (
8%, due to the MWNT DEP assembly limitation) are still the major concerns. Post-CMOS (Localized Heating) Monolithic CMOS-CNT integration is desirable to fully utilize the potentials of nanotubes for emerging C 454 CMOS-CNT Integration CMOS-CNT Integration, Fig. 6 (a) The 3D schematic showing the concept of the CMOS-integrated CNTs. The CVD chamber is kept at room temperature all the time. The red part represents the hot microheater that has been activated for high temperature nanotube synthesis. (b) Crosssectional view of the device. (c) The schematic 3D microheater showing the local synthesis from the hotspot and self-assembly on the cold landing wall under the local electric field nanotechnology applications, but the approaches introduced above still cannot meet all the requirements and realize complete compatibility with CMOS processes. To solve the problem, a simple and scalable monolithic CMOS-CNT integration technique using a novel maskless post-CMOS surface micromachining processing has been proposed. This approach is fully compatible with commercial foundry CMOS processes and has no specific requirements on the type of metallization layers and substrates. As illustrated in Fig. 6, the basic idea of the monolithic integration approach is to use maskless postCMOS MEMS processing to form micro-cavities for thermal isolation and use the gate polysilicon to form resistors for localized heating as well as the nanotube-toCMOS interconnect. The microheaters, made of the gate polysilicon, are deposited and patterned along with the gates of the transistors in the standard CMOS foundry processes. One of the top metal layers (i.e., the metal-3 layer as shown in Fig. 6b) is also patterned during the CMOS fabrication. It is used as an etching mask in the following post-CMOS microfabrication process for creating the micro-cavities. Finally, the polysilicon microheaters are exposed and suspended in a microcavity on a CMOS substrate. The circuits are covered under the metallization and passivation layers, as illustrated in Fig. 6b. Unlike the traditional thermal CVD synthesis in which the whole chamber is heated to above 800 C, the CVD chamber is kept at room temperature all the time, with only the microheaters activated to provide the local high temperature for CNT growth (Fig. 6a, the red part represents the hot microheater). The top view of a microheater design is shown in Fig. 6c. There are two polysilicon bridges: one as the microheater for generating high temperature to initiate CNT growth and the other for CNT landing. With the cold wall grounded, an E-field perpendicular to the surface of the two bridges will be induced during CNT growth. Activated by localized heating, the nanotubes will start to grow from the hotspot (i.e., the center of the microheater) and will eventually reach the secondary cold bridge under the guide of the local E-field. Since both the microheater bridge and the landing bridge are made of the gate polysilicon layer and have been interconnected with the metal layers in CMOS foundry process, the as-grown CNTs can be electrically connected to the CMOS circuitry on the same chip without any post-growth clamping or connection steps. This technology has been verified at the chip level. The CMOS chips were fabricated in the AMI 0.5 mm CMOS-CNT Integration 455 C C CMOS-CNT Integration, Fig. 7 (a) The CMOS chip photograph (1.5  1.5 mm2) after foundry process; (b) The CMOS chip photograph after post-CMOS process (before final DRIE step); (c) Close-up optical image of one microheater and nearby circuit. CMOS circuit area, although visible, is protected under silicon dioxide layer. Only the microheater and secondary cold wall within the micro-cavity are exposed to synthesis gases. Polysilicon heater and metal wire are connected by via. (d) and (e) Closed-up SEM images of two microheaters CMOS-CNT Integration, Fig. 8 Localized synthesis of carbon nanotubes grown from the 3  3 mm microheater, suspended across the trench and landed on the secondary polysilicon tip 3-metal CMOS process. Optical microscope images of a CMOS chip before and after MEMs fabrication are shown in Fig. 7a, b. The total chip area is 1.5  1.5 mm2, including test circuits and 13 embedded microheaters. SEMs of two microheaters are shown in Fig. 7d, e, with resistances of 97 and 117 O, respectively. At about 2.5 V, red glowing was observed for the design in Fig. 7(e). This voltage was also used for the CNT growth. Figure 8 shows one device with successful CNT growth, where individual suspended carbon nanotubes were grown from the 3  3 mm microheater shown in Fig. 7e and landed on the near polysilicon tip. The overall resistance of the CNTs is measured between the microheater and the cold polysilicon wall at room temperature. The typical resistances of in situ synthesized CNTs range from 5 to15 MO. The resistance variation from device to device is mainly due to the variation of the CNT quantity grown on each microheater. Junction effects of Schottky contacts were observed for self-assembled polysilicon/CNTs/ polysilicon heterojunctions. After successful synthesis of carbon nanotubes, the influence of the localized heating on nearby CMOS circuits was evaluated. Simple circuits, such as inverters, were tested and proved working properly. There was no change to the rising and falling time after the CNT growth. The dc electrical characteristics of individual transistors had no considerable change after CNT growth, demonstrating the CMOS compatibility of this integration approach. Summary CMOS-CNT integration has been demonstrated by using both intra- and post- CMOS processes. Several methods have been developed to overcome the C 456 temperature conflict between CNT growth and CMOS, including using high temperature refractory metals for interconnect, low temperature CVD, transferring/ assembling CNTs prepared off site, and localized heating. Among these techniques, localized heating is very promising. Truly monolithic CNT-CMOS integration has been demonstrated on foundry CMOS substrate by employing MEMS and localized heating. This post-CMOS microfabrication is maskless, and the CNT growth does not affect the characteristics of the transistors on the same chip. Cross-References ▶ Carbon Nanotube-Metal Contact ▶ Carbon Nanotubes for Chip Interconnections ▶ Carbon Nanotubes ▶ Synthesis of Carbon Nanotubes References 1. Zhang, W., Jha, M., Shang, L.: NATURE: A hybrid nanotube/CMOS dynamically reconfigurable architecture. Design Automation Conference, 2006 43 rd ACM/IEEE, pp. 711–716 (2006) 2. Nantero, I.: “NRAM ®,” in http://www.nantero.com/ mission.html, 2000–2009 3. Agarwal, V., Chen, C.-L., Dokmeci, M. R., Sonkusale, S.: A CMOS integrated thermal sensor based on single-walled carbon nanotubes. IEEE Sensors 2008 Conference, pp. 748– 751 (2008) 4. Cho, T.S., Lee, K.-J., Kong, J., Chandrakasan, A.P.: A 32uW 1.83-kS/s carbon nanotube chemical sensor system. IEEE J. Soild-State Circuits 44, 659–669 (2009) 5. Close, G.F., Yasuda, S., Paul, B., Fujita, S., Wong, H.-S.P.: A 1 GHz integrated circuit with carbon nanotube interconnects and silicon transistors. Nano Lett. 8, 706–709 (2008) 6. Chakraborty, R.S., Narasimhan, S., Bhunia, S.: Hybridization of CMOS with CNT-based nano-electromechanical switch for low leakage and Robust circuit design. IEEE Trans. Circuits Syst 54, 2480–2488 (2007) 7. Journet, C., Maser, W.K., Bernier, P., Loiseau, A., Lamy de la Chapelle, M., Lefrant, S., Deniard, P., Lee, R., Fischerk, J.E.: Large-scale production of single-walled carbon nanotubes by the electric-arc technique. Nature 388, 756–758 (1997) 8. Guo, T., Nikolaev, P., Thess, A., Colbert, D.T., Smalley, R.E.: Catalytic growth of single-walled nanotubes by laser vaporization. Chem. Phys. Lett. 243, 49–54 (1995) 9. Cassell, A.M., Raymakers, J.A., Kong, J., Dai, H.: Large Scale CVD Synthesis of Single-Walled Carbon Nanotubes. J. Phys. Chem. B 103, 6484–6492 (1999) 10. Meyyappan, M.: Carbon Nanotubes: Science and Applications. New York, CRC Press (2005) CMOS-MEMS 11. Franklin, N.R., Wang, Q., Tombler, T.W., Javey, A., Shim, M., Dai, H.: Integration of suspended carbon nanotube arrays into electronic devices and electromechanical systems. Appl. Phys. Lett. 81, 913–915 (2002) 12. Tseng, Y.-C., Xuan, P., Javey, A., Malloy, R., Wang, Q., Bokor, J., Dai, H.: Monolithic integration of carbon nanotube devices with silicon MOS technology. Nano Lett. 4, 123–127 (2004) 13. Ghavanini, F.A., Poche, H.L., Berg, J., Saleem, A.M., Kabir, M.S., Lundgren, P., Enoksson, P.: Compatibility assessment of CVD growth of carbon nanofibers on bulk CMOS devices. Nano Lett. 8, 2437–2441 (2008) 14. Hofmann, S., Ducati, C., Robertson, J., Kleinsorge, B.: Low-temperature growth of carbon nanotubes by plasmaenhanced chemical vapor deposition. Appl. Phys. Lett. 83, 135–137 (2003) 15. Englander, O., Christensen, D., Lin, L.: Local synthesis of silicon nanowires and carbon nanotubes on microbridges. Appl. Phys. Lett. 82, 4797–4799 (2003) 16. Haque, M.S., Teo, K.B.K., Rupensinghe, N.L., Ali, S.Z., Haneef, I., Maeng, S., Park, J., Udrea, F., Milne, W.I.: Onchip deposition of carbon nanotubes using CMOS microhotplates. Nanotechnology 19, 025607 (2008) 17. Huang, X.M.H., Caldwell, R., Huang, L., Jun, S.C., Huang, M., Sfeir, M.Y., O’Brien, S.P., Hone, J.: Controlled placement of individual carbon nanotubes. Nano Lett. 5, 1515–1518 (2005) 18. Liu, J., Casavant, M.J., Cox, M., Walters, D.A., Boul, P., Lu, W., Rimberg, A.J., Smith, K.A., Colbert, D.T., Smalley, R.E.: Controlled deposition of individual singlewalled carbon nanotubes on chemically functionalized templates. Chem. Phys. Lett. 303, 125–129 (1999) 19. Ko, H., Tsukruk, V.V.: Liquid-crystalline processing of highly oriented carbon nanotube arrays for thin-film transistors. Nano Lett. 6, 1443–1448 (2006) 20. Schwamb, T., Schirmer, N.C., Burg, B.R., Poulikakos, D.: Fountain-pen controlled dielectrophoresis for carbon nanotube-integration in device assembly. Appl. Phys. Lett. 93, 193104 (2008) 21. Ryu, K., Badmaev, A., Wang, C., Lin, A., Patil, N., Gomez, L., Kumar, A., Mitra, S., Wong, H.-S.P., Zhou, C.: CMOSanalogous wafer-scale nanotube-on-insulator approach for submicrometer devices and integrated circuits using aligned nanotubes. Nano Lett. 9, 189–197 (2009) CMOS-MEMS ▶ CMOS MEMS Fabrication Technologies CNT Arrays ▶ Vertically Aligned Carbon Nanotubes, Collective Mechanical Behavior Compliant Mechanisms 457 C CNT Biosensor CNT-FET ▶ Nanostructure Field Effect Transistor Biosensors ▶ Nanostructure Field Effect Transistor Biosensors C Cob Web CNT Brushes ▶ Spider Silk ▶ Vertically Aligned Carbon Nanotubes, Collective Mechanical Behavior Cochlea Implant CNT Bundles ▶ Bio-inspired CMOS Cochlea ▶ Vertically Aligned Carbon Nanotubes, Collective Mechanical Behavior Cold-Wall Thermal Chemical Vapor Deposition ▶ Chemical Vapor Deposition (CVD) CNT Foams ▶ Vertically Aligned Carbon Nanotubes, Collective Mechanical Behavior Compliant Mechanisms CNT Forests Larry L. Howell Department of Mechanical Engineering, Brigham Young University, Provo, UT, USA ▶ Vertically Aligned Carbon Nanotubes, Collective Mechanical Behavior Synonyms Compliant systems; Flexures; Flexure mechanisms CNT Mats Definition ▶ Vertically Aligned Carbon Nanotubes, Collective Mechanical Behavior Compliant mechanisms gain their motion from the deflection of elastic members. Main Text CNT Turfs ▶ Vertically Aligned Carbon Nanotubes, Collective Mechanical Behavior Compliant mechanisms offer an opportunity to achieve complex motions within the limitations of micro- and nano-fabrication. Because compliant mechanisms gain C 458 Compliant Mechanisms, Fig. 1 A folded-beam suspension is an example of a widely used compliant mechanism in microelectromechanical systems (MEMS) applications Compliant Mechanisms F Compliant legs Anchor their motion from the constrained bending of flexible parts, they can achieve complex motion from simple topologies. Traditional mechanisms use rigid parts connected at articulating joints (such as hinges, axles, or bearings), which usually requires assembly of components and results in friction at the connecting surfaces [1–3]. Because traditional bearings are not practical and lubrication is problematic, friction and wear present major difficulties. Nature provides an example of how to effectively address problems with motion at small scales. Most moving components in nature are flexible instead of stiff, and the motion comes from bending the flexible parts instead of rigid parts connected with hinges (for example, consider hearts, elephant trunks, and bee wings). The smaller the specimen, the more likely it is to use the deflection of flexible components to obtain its motion. And so it is with man-made systems as well, the smaller the device, the greater the advantages for using compliance [1]. Advantages of Compliant Mechanisms Some of the advantages of compliant mechanisms at the micro- and nanoscales include the following: Can be made from one layer of material. Compliant mechanisms can be fabricated from a single layer. This makes them compatible with many common microelectromechanical system (MEMS) fabrication methods, such as surface micromachining, bulk micromachining, and LIGA. For example, consider the folded beam suspension shown in Fig. 1. This device is often used as a suspension element in MEMS systems. It offers a simple approach for Suspended proof mass Compliant Mechanisms, Fig. 2 This scanning electron micrograph shows a thermal actuator that uses multiple layers of compliant elements to achieve large amplification with a small footprint constrained linear motion, and also integrates a return spring function. The device can achieve large deflections with reasonable off-axis stiffness. The compliant mechanism makes it possible to do these functions with a single layer of material. No assembly required. Compliant mechanisms that gain all of their motion from the deflection of flexible components are “fully compliant mechanisms,” where devices that combine both traditional and compliant elements are called “partially compliant mechanisms.” Fully compliant mechanism can usually be fabricated without assembly of different components. Small footprint. Some compliant mechanisms can also be designed to have a small footprint on the substrate on which they are built. Various strategies can be used to decrease the size of a mechanism. Figure 2 shows a thermal actuator that uses multiple layers to achieve a small footprint. Compliant Mechanisms 459 C Challenges of Compliant Mechanisms Flexible cantilever Tip Sample surface Compliant Mechanisms, Fig. 3 The cantilever of an atomic force microscope (AFM) is an example of compliance employed in high-precision instruments Friction-free motion. Because compliant mechanisms gain their motion from deflection of flexible members rather than from traditional articulating joints, it is possible to reduce or eliminate the friction associated with rubbing surfaces. This results in reduced wear and eliminates the need for lubrication, as described next. Wear-free motion. Wear can be particularly problematic at small scales, and the elimination of friction can result in the elimination of wear at the connecting surfaces of joints. For devices that are intended to undergo many cycles of motion, eliminating friction can dramatically increase the life of the system. No need for lubrication. Another consequence of eliminating friction is that lubricants are not needed for the motion. This is particularly important at small scales where lubrication can be problematic. High precision. Flexures have long been used in high precision instruments because of the repeatability of their motion. Some reasons for compliant mechanisms’ precision are the backlash-free motion inherent in compliant mechanisms and the wear-free and friction-free motion described above. The cantilever associated with an atomic force microscope (Fig. 3) is an example application. Integrated functions. Like similar systems in nature, compliant mechanisms have the ability to integrate multiple functions into few components. For example, compliant mechanisms often provide both the motion function and a return-spring function. Thermal actuators are another example of integration of functions, as described later. High reliability. The combination of highly constrained motion of compliant mechanisms, the relative purity of materials used in micro/nanofabrication, and wear-free motion result in high reliability of compliant mechanisms at the micro/nanoscale. Compliant mechanisms have many advantages, but they also have some significant challenges. A few of these are discussed below [1]: Limited rotation. One clear drawback of compliant mechanisms is the general inability to undergo continuous rotation. Also, if a fully compliant mechanism is constructed from a single layer of material, then special care has to be taken to ensure that moving segments of the compliant mechanism do not collide with other segments of the same mechanism. Dependence on material properties. The performance of compliant mechanisms is highly dependent on the material properties, which are not always well known. Nonlinear motions. The deflections experienced by compliant mechanisms often extend beyond the range of linearized beam equations. This can make their analysis and design more complicated. Fatigue analysis. Because most compliant mechanisms undergo repeated loading, it is important to consider the fatigue life of the device. Interestingly, because of the types of materials used and their purity, many MEMS compliant mechanisms will either fail on their first loading cycle or will have infinite fatigue life. Because of the low inertia of MEMS devices, it is often easy to quickly test a MEMS device to many millions of cycles. Factors such as stress concentrations, the operating temperature, and other environment conditions can affect the fatigue life. Difficult design. Integration of functions into fewer components, nonlinear displacements, dependence on material properties, the need to avoid self-collisions during motion, and designing for appropriate fatigue life, all combine to make the design of compliant mechanisms nontrivial and often difficult. Example Applications of Compliant Mechanisms Examples of MEMS compliant mechanisms are shown here to further illustrate their properties and to demonstrate a few applications. Digital Micromirrors. One of the most visible commercially available microelectromechanical systems is Texas Instruments’ Digital Micromirror Device (DMD™), which is used in applications such as portable projectors. The DMD is a rectangular array of C C 460 Compliant Mechanisms Pressure Mirror Compliant diaphragm Etched cavity Mirror support post Landing tips Torsion hinge Backside port Address electrode Yoke Hinge support post Electrode support post Metal 3 address pads Landing sites Bias/reset bus To SRAM Compliant Mechanisms, Fig. 4 Texas Instrument’s Digital Micromirror Device (DMD™) uses compliant torsion hinges to facilitate mirror motion (Illustration courtesy of Texas Instruments) moving micromirrors that is combined with a light source, optics, and electronics to project high-quality color images. Figure 4 shows the architecture of a single DMD pixel. A 16-mm-square aluminum mirror is rigidly attached to a platform (the “yoke”). Flexible torsion hinges are used to connect the yoke to rigid posts. An applied voltage creates an electrostatic force that causes the mirror to rotate about the torsion hinges. When tilted in the on position, the mirror directs light from the light source to the projection optics and the pixel appears bright. When the mirror is tilted in the off position, the light is directed away from the projection optics and the pixel appears dark. The micromirrors can be combined in an array on a chip, and each micromirror is associated with the pixel of a projected Compliant Mechanisms, Fig. 5 The strain on a compliant diaphragm of a piezoresistive pressure sensors results in a detectable change in resistance, which is correlated with the pressure image. The torsion hinges use compliance to obtain motion while avoiding rubbing parts that cause friction and wear. The hinges can be deflected thousands of times per second and infinite fatigue life is essential. Piezoresistive pressure sensors. A sensor is a device that responds to a physical input (such as motion, radiation, heat, pressure, magnetic field), and transmits a resulting signal that is usually used for detection, measurement, or control. Advantages of MEMS sensors are their size and their ability to be more closely integrated with their associated electronics. Piezoresistive sensing methods are among the most commonly employed sensing methods in MEMS. Piezoresistance is the change in resistivity caused by mechanical stresses applied to a material. Bulk micromachined pressure sensors have been commercially available since the 1970s. A typical design is illustrated in Fig. 5. A cavity is etched to create a compliant diaphragm that deflects under pressure. Piezoresistive elements on the diaphragm change resistance as the pressure increases; this change in resistance is measured and is correlated with the corresponding pressure. Capacitive acceleration sensors. Accelerometers are another example of commercially successful MEMS sensors. Applications include automotive airbag safety systems, mobile electronics, hard drive protection, gaming, and others. Figure 6 illustrates an example of a surface micromachined capacitive accelerometer. Acceleration causes a displacement of the inertial mass connected to the compliant suspension, and the capacitance change between the comb fingers is detected. Compliant Mechanisms 461 Stationary fingers C Contact pad Anchor to substrate + Flexible spring legs Exaggerated leg deflection C Direction of motion Inertial mass Translating shuttle Expansion legs – Compliant Mechanisms, Fig. 6 This accelerometer makes use of compliant legs that deflect under inertial loads. The deflection results in a detectable change in capacitance and is correlated with the corresponding acceleration Thermal actuators. A change in temperature causes an object to undergo a change in length, where the change is proportional to the material’s coefficient of thermal expansion [4]. This length change is usually too small to be useful in most actuation purposes. Therefore, compliant mechanisms can be used to amplify the displacement of thermal actuators. Figure 7 illustrates an example of using compliant mechanisms to amplify thermal expansion in microactuators. Figure 8 shows a scanning electron micrograph of a thermomechanical in-plane microactuator (TIM) illustrated in Fig. 7. It consists of thin legs connecting both sides of a center shuttle. The leg ends not connected to the shuttle are anchored to bond pads on the substrate and are fabricated at a slight angle to bias motion in the desired direction. As voltage is applied across the bond pads, electric current flows through the thin legs. The legs have a small cross-sectional area and thus have a high electrical resistance, which causes the legs to heat up as the current passes through them. The shuttle moves forward to accommodate the resulting thermal expansion. Advantages of this device include its ability to obtain high deflections and large forces, as well as its ability to provide a wide range of output forces by changing the number of legs in the design. Compliant Mechanisms, Fig. 7 A schematic of a thermomechanical in-plane microactuor (TIM) that uses compliant expansion legs to amplify the motion caused by thermal expansion Analysis and Design of Compliant Mechanisms Multiple approaches are available for the analysis and design of compliant mechanisms. Three of the most developed approaches are described below. Finite element analysis. Finite element methods are the most powerful and general methods available to analyze compliant mechanisms. Commercial software is currently available that has the capability of analyzing the large, nonlinear deflections often associated with compliant mechanisms. The general nature of the method makes it applicable for a wide range of geometries, materials, and applications. Increasingly powerful computational hardware has made it possible to analyze even very complex compliant mechanisms. It is also possible to use finite element methods in the design of compliant mechanisms, particularly once a preliminary design has been determined. But in the early phases of design, other methods (or hybrid methods) are often preferred so that many design iterations can be quickly analyzed. Pseudo-rigid-body model. The pseudo-rigid-body model is used to model compliant mechanisms as traditional rigid-body mechanisms, which opens up the possibility of using the design and analysis C 462 Compliant Mechanisms methods developed for rigid-body mechanisms in the design of compliant mechanisms [1]. With the pseudorigid-body model approach, flexible parts are modeled as rigid links connected at appropriately placed pins, with springs to represent the compliant mechanism’s resistance to motion. Extensive work has been done to develop pseudo-rigid-body models for a wide range of geometries and loading conditions. Consider a simple Compliant Mechanisms, Fig. 8 A scanning electron micrograph of a thermal actuator illustrated in Fig. 7 example. The micromechanism shown in Fig. 9 has a rigid shuttle that is guided by two flexible legs. (Note that the folded-beam suspension in Fig. 1 has four of these devices connected in series and parallel.) The pseudo-rigid-body model of the mechanism models the flexible legs as rigid links connected at pin joints with torsional springs. Using appropriately located joints and appropriately sized springs, this model is very accurate well into the nonlinear range. For example, if the flexible legs are single-walled carbon nanotubes, comparisons to molecular simulations have shown the pseudo-rigid-body model to provide accurate results [5]. The advantages of the pseudorigid-body model are realized during the early phases of design where many design iterations can be quickly evaluated, traditional mechanism design approaches can be employed, and motions can be easily visualized. Topology optimization. Suppose that all that is known about a design is the desired performance and design domain. Topology optimization shows promise for designing compliant mechanisms under such conditions. The advantage is that very little prior knowledge about the resulting compliant mechanism is needed, and any biases of the designer are eliminated [6]. Topology optimization is often integrated with finite element methods to consider many possible ways of distributing material with the design domain. This has the potential to find designs that would not otherwise be discovered by other methods. Infinite possible topologies are possible and finite element methods can be employed to evaluate the different possibilities. The resolution of the design domain mesh can be a limiting factor, but once a desirable topology is identified, it can be further refined using other approaches. Rigid shuttle Compliant Mechanisms, Fig. 9 The pseudo-rigidbody model of the compliant parallel-guiding mechanism consists of appropriately located pin joints and torsional springs (This device is a building block of other devices, such as the foldedbeam suspension) Pseudo-rigidbody model Deflected position Compliant leg Torsional spring Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering 463 C Conclusion Compliant Systems Compliant mechanisms provide significant benefits for micro- and nano-motion applications. They can be compatible with many fabrication methods, do not require assembly, have friction-free and wear-free motion, provide high precision and high reliability, and they can integrate multiple functions into fewer components. The major challenges associated with compliant mechanisms come from the difficulty associated with their design, limited rotation, and the need to ensure adequate fatigue life. It is likely that compliant mechanisms will see increasing use in micro- and nano-mechanical systems as more people understand their advantages and have tools available for their development. Cross-References ▶ AFM ▶ Basic MEMS Actuators ▶ Biomimetics ▶ Finite Element Methods for Computational Nanooptics ▶ Insect flight and Micro Air Vehicles (MAVs) ▶ MEMS on Flexible Substrates ▶ Nanogrippers ▶ Piezoresistivity ▶ Thermal Actuators References 1. Howell, L.L.: Compliant Mechanisms. Wiley, New York (2001) 2. Lobontiu, N.: Compliant Mechanisms: Design of Flexure Hinges. CRC Press, Boca Raton (2003) 3. Smith, S.T.: Flexures: Elements of Elastic Mechanisms. Taylor & Francis, London (2000) 4. Howell, L.L., McLain, T.W., Baker, M.S., Lott, C.D.: Techniques in the design of thermomechanical microactuators. In: Leondes, C.T. (ed.) MEMS/NEMS Handbook, Techniques and Applications, pp. 187–200. Springer, New York (2006) 5. Howell, L.L., DiBiasio, C.M., Cullinan, M.A., Panas, R., Culpepper, M.L.: A pseudo-rigid-body model for large deflections of fixed-clamped carbon nanotubes. J. Mech. Robot. 2, 034501 (2010) 6. Frecker, M.I., Ananthasuresh, G.K., Nishiwaki, S., Kikuchi, N., Kota, S.: Topological synthesis of compliant mechanisms using multi-criteria optimization. J. Mech. Des. 119, 238–245 (1997) ▶ Compliant Mechanisms C Composite Materials ▶ Theory of Optical Metamaterials Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering A. T. Conlisk Department of Mechanical Engineering, The Ohio State University, Columbus, OH, USA Synonyms Microscale mechanics fluid mechanics; Nanoscale fluid Definition Because of the small scale of the fluid conduits, electric fields must often be used to transport fluids especially at the nanoscale. This means that the fluids must be electrically conducting, and so microfluidics and nanofluidics require the user to be literate in fluid mechanics, heat and mass transfer, electrostatics, electrokinetics, electrochemistry, and if biomolecules are involved, molecular biology. Introduction The term microfluidics refers generally to internal flow in a tube or channel whose smallest dimension is under 100 mm. Nanofluidics refers to the same phenomenon in a conduit whose smallest dimension is less than 100 nm. Microchannels and nanochannels have large surface-to-volume ratio, so that surface properties become enormously important. In fully developed C 464 Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering channel flow, the pressure drop Dp
h13 , where h is the small dimension, and so the pressure drop is prohibitively large for a nanoscale channel. Thus, a solvent fluid such as water, proteins and other biomolecules, and other colloidal particles are most often transported electrokinetically. This means that the art of designing micro and nanodevices requires a significant amount of knowledge of fluid flow and mass transfer (biofluids are usually multicomponent mixtures) and often heat transfer, electrostatics, electrokinetics, electrochemistry, and molecular biology. Details of the character of these fields of study can be found in the recently published book by the author [1]. The study of micro/nanofluidics requires knowledge of all of the above-mentioned fields and so has a unifying effect. Moreover, nanofluidics opens the door to the discovery of the structure and conformation of biomaterials such as proteins and polysacchrides through molecular simulation. In dealing with devices with small-scale features, microscale and below, there are three activities that normally comprise the design process; these are: • Modeling: computational and theoretical • Fabrication • Experimental methods Modeling is often done prior to the fabrication process as a guide as to what can be done. Experimental methods are usually used to assess the performance of a device, among other purposes. This means that surface properties become very important at the microscale and nanoscale and surfaces are routinely engineered to achieve a desired objective. In most devices the nanoscale features interface directly with microscale features. A typical channel geometry is depicted on Fig. 1. Surface-to-Volume Ratio Fluid Mechanics Consider a channel of rectangular cross-section having dimensions in the (x, y, z) coordinate system of (L, h, W) with the primary direction of fluid motion being in the x direction. Then the surface-to-volume ratio is given by Micro and nanofluidics generally involve the flow of electrically conducting fluids, electrolyte solutions that are assumed to be incompressible, having a constant density. Generally, the flows are internal, bounded on each side by walls, and the flows are assumed to be fully developed. In this case, referring to Fig. 1, the governing equation for the velocity u in a channel is given by   S 1 1 1 ¼2 þ þ ¼ 6 m1 V L h W (1) for a channel having all three dimensions L ¼ h ¼ W ¼ 1 m. On the other hand, a channel having dimensions (10 mm, 102 mm, 10 mm), S
2  108 m1 V (2) Side View L h y,v x,u Flow L >>h z,w End View W z,w W >>h Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering, Fig. 1 Geometry of a typical channel. In applications h << W,L where W is the width of the channel and L its length in the primary flow direction. u,v,w are the fluid velocities in the x,y,z directions # A.T. Conlisk used with permission m @ 2 u @p  Bx ¼ @y2 @x (3) where p is the pressure and Bx is a body force. The no-slip condition is applied at each wall: u ¼ 0 at y ¼ 0, h. Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering Mass Transfer The molar flux of species A for a dilute electrically conducting mixture is ! ! ! N A ¼ DAB =cA þ mA zA cA E þ cA V cA ¼ cA0 ezA f f¼ W¼ F q N ¼ 0 2 q 4pee r C (6) and is directed outward from the body of charge q and toward the body having a charge q0 if q > 0 and the electric field is in the opposite direction if q < 0. In general, the electric field is a vector. This formula is called Coulomb’s Law and ee is called the electrical permittivity. The electrical permittivity is a transport property like the viscosity and thermal conductivity of a fluid. The electric field due to a flat wall having a surface charge density s in Coulomb on one side is directed m2 normal to the surface and has magnitude E¼ s 2ee (7) a ! ! E ds (8) Z b a ! ! F ds (9) In differential form, the electrical potential is given by ! E ¼ =f (10) For a single charge Gauss’s Law is given by The Electric Field E¼ b The units of the electric potential are ¼ 1Volt ¼ 1 V. This formula is similar to the formula for mechanical work given by ZZ An electric field is set up around any charged body and is defined as the force per charge on a surface. Electrical charges are either positive or negative and like charges repel and opposite charges attract. For two bodies of charge q and q0 , the electric field is defined by Z Nm C (5) and this is termed the Boltzmann distribution for the concentration of species A. C A wire is characterized as having a line charge density and if charges are distributed over a volume, . charge density is defined and called re in Coulomb m3 The electrical potential is defined as the work done in moving a unit of charge and mathematically (4) Here DAB is the diffusion coefficient, R is the universal gas constant, T is the temperature, ZAmA is called the AB ionic mobility with mA ¼ FD RT ; zA is the valence, ! F ¼ 96500 Coul mole is Faraday’s constant, and E is the electric field. Equation 4 is called the Nernst-Planck equation and the electric field term in the flux equation is called electrical migration. The boundary condition of interest here is that the solid walls in Fig. 1 are impermeable to species A, or NAv ¼ 0 at y ¼ 0, h. In one dimension, Eq. 4 can be integrated to give 465 S ! ! ee E d A ¼ q (11) For a volume that contains a continuous distribution of charge, re, summing over all the charges using the definition of the integral, Gauss’s Law becomes ZZ S ! ! ee E d A ¼ ZZ redV (12) V Using Eq. 12 and the differential form of the definition of the electrical potential, it follows that =2f ¼  re ee (13) This is a Poisson equation for the potential given the volume charge density. The combination of Eqs. 4 and 13 is called the Poisson-Nernst-Planck system of equations. Electrochemistry Electrochemistry may be broadly defined as the study of the electrical properties of chemical and biological C C 466 Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering Computational Micro/ Nanofluidics: Unifier of Physical and Natural Sciences and Engineering, Fig. 2 The electric double layer (EDL) consists of a layer of counter ions pinned to the wall, the Stern layer, and a diffuse layer of mobile ions outside that layer. The wall is shown as being negatively charged and the z–potential is defined as the electrical potential at the Stern plane. # A.T. Conlisk used with permission Stern plane + + + - + + + + + + + + + + + + + + ++ ++ + + + + z + + + + + + + + + + + ++ + f U + + - + Diffuse +layer surface of shear material [2]. In particular much of electrochemistry pertinent to micro and nanofluidics involves the study of the behavior of ionic solutions and the electrical double layer (EDL). Electrochemistry of electrodes is important to understand the operation of a battery. An ionic or electrolyte solution is a mixture of ions, or charged species immersed in a solvent, often water. It is the charged nature of ionic solutions that allows the fluid to move under the action of an electric field, provided by electrodes placed upstream and downstream of a channel in a nanopore membrane. The term membrane is used to mean a thin sheet of porous material that allows fluid to flow in channels that make up the porous part of the membrane. Those channels are often like those channels depicted on Fig. 1. Because the surface-to-volume ratio is so large in a nanoscale channel, the properties of the surface are extremely important. Fluid can be moved by an electric field if the surfaces of a channel are charged. If the surface is negatively charged, a surplus of positive ions will arrange themselves near the wall. This is shown in Fig. 2. It is this excess charge that allows fluid to be transported by an externally applied electric field. The nominal length scale associated with the EDL is the Debye length defined by l¼ pffiffiffiffiffiffiffiffiffiffi ee RT FI 1=2 (14) where F is Faraday’s constant, ee is the electrical permittivity of the medium, I is the ionic strength P I ¼ i z2i ci , ci the concentrations of the electrolyte constituents at some reference location, R is the universal gas constant, zi is the valence of species i and T is the temperature. The ion distribution within the EDL can be described by using the number density, concentration, or mole fraction. Engineers usually prefer the dimensionless mole fraction whereas chemists usually use concentration or number density. There are two views of the ion distribution within the electrical double layer that are generally thought to be valid and have been verified by numerical solutions of the governing equations (see the section on Electrokinetic Phenomena below). The Gouy– Chapman [3, 4] model of the electric double layer allows counterions to collect near the surface in much greater numbers than coions. This model as numerical solutions suggest [1] occurs at higher surface charge densities. The Debye–H€uckel picture assumes that coions and counterions collect near the surface in roughly equal amounts, above and below a mean value. These pictures are depicted on Fig. 3. Molecular Biology The nanoscale is the scale of biology since many proteins and other biomolecules have nanoscale dimensions. Many of the applications of nanofluidics such as rapid molecular analysis, drug delivery, and biochemical sensing have a biological entity as an integral part of their operation. Moreover, using nanofluidic tools, DNA sequencing is now possible. The book by Alberts [6] is a useful tool for learning molecular biology. Nucleic acids are polymers consisting of nucleotides. Those based on a sugar called ribose are called ribo nucleic acids (RNA) and those based on deoxyribose are called deoxyribonucleic acids (DNA). RNA is single stranded while DNA is usually double stranded although single-stranded DNA (ss-DNA) does exist. Nucleotides contain five-carbon sugars attached to one or more phosphate groups (a phosphorus central atom surrounded by four oxygens) and a base which can be either adenine (A), cytosine (C), guanine (G), or thymine (T). Two nucleotides connected by a hydrogen bond is called a base pair Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering a b g g 467 C C f f y Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering, Fig. 3 (a) Debye– H€uckel [5] picture of the electric double layer. Here g denotes the cation mole fraction and f denotes the anion mole fraction. The Debye–H€uckel model assumes the cation and anion wall (bp). Protein synthesis begins at a gene on a particular strand of a DNA molecule in a cell [6]. There are seven basic types of proteins classified according to their function although different authors use different terms to describe each class; see for example Alberts [6], panel 5–1. Enzymes are catalysts in biological reactions within the cell. For example, the immune system responds to foreign bacteria and viruses by producing antibodies that destroy or bind to the antigen, the foreign agent. The antigen is the catalyst, or reaction enhancer for inducing the immune response: the production of the antibodies. Proteins are responsible for many of the essential functions of the body, including moving material into and out of cells, regulating metabolism, managing temperature and pH, and muscle operation, among other functions. Proteins are large and complex molecules, polymers made up of a total of 20 amino acids, and held together by peptide bonds. The 20 amino acids have side chains that can be basic, acidic, polar or nonpolar. Because they are so large they cannot described easily in a single chemical formula or picture. Thus, molecular biologists depict proteins and other macromolecules in distinct levels of structure. The primary structure is the amino acid sequence, the order in which the 20 amino acids appear. The secondary structure depicts the folding properties of a protein as depicted on Fig. 4. Proteins are further described by more complex folding of the secondary structure (tertiary structure) and a quaternary structure if the protein has more than one backbone. y mole fractions are symmetric about a mean value which occurs for low surface charge densities. (b) Gouy–Chapman model [3, 4] of the EDL allows many more counterions than coions to collect near the charged surface and is valid at higher surface charge densities. # A.T. Conlisk used with permission Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering, Fig. 4 Ribbon view of the protein albumin depicting its folding pattern, the secondary structure of a protein. From the European Bioinformatics Institute, public domain, www.ebi.ac.uk Proteins are usually negatively charged and thus nanopore membranes for rapid molecular analysis can be used to separate different types of proteins and other biomolecules based on different values of size and charge. Biomolecules are what is termed soft material in that they are porous and deform under stress. Indeed, recent measurements of the conformation of albumin show that it may take the shape of a wedge, looking like a piece of pie. With the explosive growth of computer capability, conformations of biomolecules are actually being computed using molecular simulation tools like molecular dynamics and Monte Carlo schemes. C 468 Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering Ion channels are natural conical nanopores whose walls are made of proteins that play a crucial role in the transport of biofluids to and from cells. The basic units of all living organisms are cells. In order to keep the cells functioning properly there needs to be a continuous flux of ions in and out of the cell and the cell components. The cell and many of its components are surrounded by a plasma membrane which provides selective transfer of ions. The membrane is made up of a double layer of lipid molecules (lipid bilayer) in which proteins are embedded. Ion channels are of two categories: carrier and pore. The carrier protein channel is based on the binding of the transport ion to a larger macromolecule, which brings it through the channel. A pore ion channel is a narrow, water-filled tunnel, permeable to the few ions and molecules small enough to fit through the tunnel (approximately 10 A˚ in diameter). Electrokinetic Phenomena As the scale of the channels in a nanopore membrane become smaller, pressure, the normal means for driving fluids through pipes and channels at macroscale (Fig. 1), becomes very difficult [1] since the pressure drop requires scales as h3 where h is the (nanoscale) channel height. Since in many applications the fluids used are electrically conducting, electric fields can be used to effectively pump fluid. Moreover electrically charged particles can move relative to the bulk fluid motion and thus species of particles can be separated. These electrokinetic phenomena are generally grouped into four classes [1]: 1. Electroosmosis (electroosmotic flow): the bulk motion of a fluid caused by an electric field 2. Electrophoresis: the motion of a charged particle in an otherwise motionless fluid or the motion of a charged particle relative to a bulk motion 3. Streaming potential or streaming current: the potential induced by a pressure gradient at zero current flow of an electrolyte mixture 4. Sedimentation potential: the electric field induced when charged particles move relative to a liquid under a gravitational or centrifugal or other force field By far the two most important of these phenomena are electroosmosis and electrophoresis and for the purposes of the theme of this entry, electroosmosis is discussed exclusively. Cathode Anode Electric Double Layer − + − − − − − − − − − − − − − − − − − − − − + + + + + + + + + + + + + + + + + ++ + − + + + + + − + + + + + − + − − +− + + + − + + − + − − + + − + − − − + − + + + − + + + + + + + + + + − − − − − − − − − − − + − + + − − +− + + + − + − + + + + + + + + + + + ++ + − − − − − − − − − − − − Electric Double Layer Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering, Fig. 5 The combination of electrodes in the regions upstream and downstream of a charged channel or membrane, usually fluid reservoirs, causes electroosmotic flow. # A.T. Conlisk used with permission The dimensionless form of the streamwise momentum equation in the fully developed flow region in the absence of a pressure gradient is e2 X @2u ¼ b zi X i @y2 i (15) and the Poisson equation for the potential in dimensionless form is e2 X @2f ¼ b z i Xi 2 @y i (16) where the partial derivatives in this one-dimensional fully developed analysis are really total derivatives, e ¼ lh and b ¼ cI where c is the total concentration including the solvent and I is the ionic strength. Here Xi is the mole fraction, but if the electrolyte concentrations are scaled on the ionic strength, P I ¼ lh i zi ci ; b ¼ 1. It is seen from Eq. 15 that the combination of the electrodes that create an electric field and the excess charge in the electrical double layers produces the electrical force that balances the viscous force causing the electrolyte to move and this is depicted on Fig. 5. Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering a 4.5 Rescaled Mole Fractions Xi/X0i 1.4 Potential and Velocity C b 1.6 1.2 1 0.8 0.6 0.4 ε = 0.01 ε = 0.05 ε = 0.1 ε = 0.5 0.2 0 0 469 0.2 0.4 y 0.6 ε = 0.01 ε = 0.05 ε = 0.1 ε = 0.5 4 3.5 3 C 2.5 2 1.5 1 0.5 0.8 1 0 0.2 0.4 y 0.6 0.8 1 Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering, Fig. 6 (a) Potential, velocity, and rescaled mole fractions for a 1:1 electrolyte for various values of e. Here the dimensional potential on both walls is z* ¼ 40 mV. In (b) the mole fractions are rescaled based on the upstream reservoir mole fractions as XX0i . The cations are i plotted in black lines and the anions are plotted in gray lines. # A.T. Conlisk used with permission The fluid velocity satisfies the no-slip condition at the wall and the electric potential satisfies f(0) ¼ f(1) ¼ 0. Then both the equations and the boundary conditions are identical and on a dimensionless basis u(y) ¼ f (y). In reality, the potential does not vanish at the wall but if a Dirichlet boundary condition holds and the potential satisfies f ¼ z at y ¼ 0,1 and thus u ¼ f  z, where z is the dimensionless z-potential at the wall. Results for the potential and velocity and the concentrations scaled on the ionic strength are presented on Fig. 6 [1]. Note that for E << 1 the fluid velocity is constant and away from the walls of the channel, unlike the Poiseuille flow of pressure-driven flow. experiments. Note that this similarity analysis does not apply for unsteady flow since there is no time derivative in the potential equation. Dimensional Analysis The equivalence of the dimensionless velocity and the dimensionless potential is important because measurements of the velocity are equivalent to measurements of the electric potential. Moreover, it is noted that a Debye length of l ¼ 1 nm in a h ¼ 100 nm channel gives the same value of e ¼ 0.01 as a l ¼ 100 nm Debye length in a h ¼ 10 mm channel so that miscoscale measurements can be validated by nanoscale computations and conversely, nanoscale computations can be validated by miscoscale Closure Physics of fluids at the nanoscale is dominated by the large surface-to-volume ratio inherent at this length scale, and thus the surfaces of a channel or membrane become extremely important. Indeed, the pressure drop across a rectangular channel in a nanopore membrane, Dp
h13 , is prohibitively large for the efficient operation of a nanofluidic device if h ¼ O(nm). Thus fluid, charged biomaterials such as proteins, and colloidal particles are most often transported electrokinetically at the nanoscale. The art of designing micro and nano fluidic devices therefore requires a significant amount of knowledge of fluid flow, mass transfer and often heat transfer, electrostatics, electrokinetics, electrochemistry, and molecular biology. The common thread is micro/nanofluidics, which plays the role of unifying and integrating these fields. In particular, nanofluidics opens the door to reveal the structure and behavior of flows around nanoparticles and the conformation of proteins and other biomolecules using molecular simulations. C 470 Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling Cross-References Definitions ▶ Applications of Nanofluidics ▶ Electrokinetic Fluid Flow in Nanostructures ▶ Micro/Nano Flow Characterization Techniques ▶ Nanochannels for Nanofluidics: Fabrication Aspects ▶ Rapid Electrokinetic Patterning Nanomaterials (or nanostructured materials, nanocomposites) are materials with characteristic size of structural elements on the order of less than several hundreds of nanometers at least in one dimension. Examples of nanomaterials include nanocrystalline materials, nanofiber, nanotube, and nanoparticlereinforced nanocomposites, and multilayered systems with submicron thickness of the layers. Atomistic modeling is based on atoms as elementary units in the models, thus providing the atomic-level resolution in the computational studies of materials structure and properties. The main atomistic methods in material research are (1) molecular dynamics technique that yields “atomic movies” of the dynamic material behavior through the integration of the equations of motion of atoms and molecules, (2) Metropolis Monte Carlo method that enables evaluation of the equilibrium properties through the ensemble averaging over a sequence of random atomic configurations generated according to the desired statistical-mechanics distribution, and (3) kinetic Monte Carlo method that provides a computationally efficient way to study systems where the structural evolution is defined by a finite number of thermally activated elementary processes. Mesoscopic modeling is a relatively new area of the computational materials science that considers material behavior at time- and length-scales intermediate between the atomistic and continuum levels. Mesoscopic models are system- /phenomenon-specific and adopt coarse-grained representations of the material structure, with elementary units in the models designed to provide a computationally efficient representation of individual crystal defects or other elements of micro/nanostructure. Examples of the mesoscopic models are coarse-grained models for molecular systems, discrete dislocation dynamics model for crystal plasticity, mesoscopic models for nanofibrous materials, cellular automata, and kinetic Monte Carlo Potts models for simulation of microstructural evolution in polycrystalline materials. References 1. Conlisk, A.T.: Essentials of Micro and Nanofluidics with Application to the Biological and Chemical Sciences. Cambridge University Press, Cambridge (2011) 2. Bockris, J.O.M, Reddy, A.K.N.: Modern Electrochemistry, vol. 1 Ionics, 2 edn. Plenum, New York/London (1998) 3. Gouy, G.: About the electric charge on the surface of an electrolyte. J. Phys. A 9, 457–468 (1910) 4. Chapman, D.L.: A contribution to the theory of electrocapillarity. Phil. Mag. 25, 475–481 (1913) 5. Debye, P., Huckel, E.: The interionic attraction theory of deviations from ideal behavior in solution. Z. Phys. 24, 185 (1923) 6. Alberts, B., Bray, D., Hopkin, K., Johnson, A., Lewis, J., Raff, M., Roberts, K., Walter, P.: Essential Cell Biology. Garland Publishing, New York (1998) Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling Leonid V. Zhigilei1, Alexey N. Volkov1 and Avinash M. Dongare2 1 Department of Materials Science and Engineering, University of Virginia, Charlottesville, VA, USA 2 Department of Materials Science and Engineering, North Carolina State University, Raleigh, NC, USA Synonyms Carbon nanotube materials; Carbon nanotubes; Computer modeling and simulation of materials; Dislocation dynamics; Kinetic Monte Carlo method; Mechanical properties of nanomaterials; Mesoscopic modeling; Metropolis Monte Carlo method; Molecular dynamics method; Multiscale modeling; Nanocrystalline materials; Nanofibrous materials and composites; Nanomaterials Computer Modeling of Nanomaterials Rapid advances in synthesis of nanostructured materials combined with reports of their enhanced or unique properties have created, over the last decades, a new Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling active area of materials research. Due to the nanoscopic size of the structural elements in nanomaterials, the interfacial regions, which represent an insignificant volume fraction in traditional materials with coarse microstructures, start to play the dominant role in defining the physical and mechanical properties of nanostructured materials. This implies that the behavior of nanomaterials cannot be understood and predicted by simply applying scaling arguments from the structure–property relationships developed for conventional polycrystalline, multiphase, and composite materials. New models and constitutive relations, therefore, are needed for an adequate description of the behavior and properties of nanomaterials. Computer modeling is playing a prominent role in the development of the theoretical understanding of the connections between the atomic-level structure and the effective (macroscopic) properties of nanomaterials. Atomistic modeling has been at the forefront of computational investigation of nanomaterials and has revealed a wealth of information on structure and properties of individual structural elements (various nanolayers, nanoparticles, nanofibers, nanowires, and nanotubes) as well as the characteristics of the interfacial regions and modification of the material properties at the nanoscale. Due to the limitations on the time- and length-scales, inherent to atomistic models, it is often difficult to perform simulations for systems that include a number of structural elements that is sufficiently large to provide a reliable description of the macroscopic properties of the nanostructured materials. An emerging key component of the computer modeling of nanomaterials is, therefore, the development of novel mesoscopic simulation techniques capable of describing the collective behavior of large groups of the elements of the nanostructures and providing the missing link between the atomistic and continuum (macroscopic) descriptions. The capabilities and limitations of the atomistic and mesoscopic computational models used in investigations of the behavior and properties of nanomaterials are briefly discussed and illustrated by examples of recent applications below. Atomistic Modeling In atomistic models [1, 2], the individual atoms are considered as elementary units, thus providing the 471 C atomic-level resolution in the description of the material behavior and properties. In classical atomistic models, the electrons are not present explicitly but are introduced through the interatomic potential, r 2 ; :::;~ r N Þ, that describes the dependence of the U ð~ r 1 ;~ potential energy of a system of N atoms on the positions ~ ri of the atoms. It is assumed that the electrons adjust to changes in atomic positions much faster than the atomic nuclei move (Born–Oppenheimer approximation), and the potential energy of a system of interacting atoms is uniquely defined by the atomic positions. The interatomic potentials are commonly described by analytic functions designed and parameterized by fitting to available experimental data (e.g., equilibrium geometry of stable phases, density, cohesive energy, elastic moduli, vibrational frequencies, characteristics of the phase transitions, etc.). The interatomic potentials can also be evaluated through direct quantum mechanics–based electronic structure calculations in socalled first principles (ab initio) simulation techniques. The ab initio simulations, however, are computationally expensive and are largely limited to relatively small systems consisting of tens to thousands of atoms. The availability of reliable and easy-to-compute interatomic potential functions is one of the main conditions for the expansion of the area of applicability of atomistic techniques to realistic quantitative analysis of the behavior and properties of nanostructured materials. The three atomistic computational techniques commonly used in materials research are: 1. Metropolis Monte Carlo method – the equilibrium properties of a system are obtained via ensemble averaging over a sequence of random atomic configurations, sampled with probability distribution characteristic for a given statistical mechanics ensemble. This is accomplished by setting up a random walk through the configurational space with specially designed choice of probabilities of going from one state to another. In the area of nanomaterials, the application of the method is largely limited to investigations of the equilibrium shapes of individual elements of nanostructures (e.g., nanoparticles) and surface structure/composition (e.g., surface reconstruction and compositional segregation [3]). 2. Kinetic Monte Carlo method – the evolution of a nanostructure can be obtained by performing atomic rearrangements governed by pre-defined C C 472 Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling Molecular Dynamics Technique Molecular dynamics (MD) is a computer simulation technique that allows one to follow the evolution of a system of N particles (atoms in the case of atomistic modeling) in time by solving classical equations of motion for all particles in the system, mi d2~ ri ~ ¼ Fi ; dt2 i ¼ 1; 2; . . . ; N (1) where mi and ~ ri are the mass and position of a particle i, ~i is the force acting on this particle due to the and F interaction with other particles in the system. The force acting on the ith particle at a given time is defined by the gradient of the inter-particle interaction potential U ð~ r 1 ;~ r 2 ; :::;~ r N Þ that, in general, is a function of the positions of all the particles: ~i ¼ ~ F Hi U ð~ r 1 ;~ r 2 ; :::;~ rN Þ (2) Once the initial conditions (initial positions and velocities of all particles in the system) and the interaction potential are defined, the equations of motion, (Eq. 1), can be solved numerically. The result of the solution is the trajectories (positions and velocities) of all the particles as a function of time, ~ ri ðtÞ;~ vi ðtÞ, which is the only direct output of an MD simulation. From the trajectories of all particles in the system, however, one can easily calculate the spatial and time evolution of structural and thermodynamic parameters of the system. For example, a detailed atomic-level analysis of continuum 10 μm 1 μm 3 2 c pi co os ls es de m mo transition rates between the states, with time increments formulated so that they relate to the microscopic kinetics of the system. Kinetic Monte Carlo is effective when the structural and/or compositional changes in a nanostructure are defined by a relatively small number of thermally activated elementary processes, for example, when surface diffusion is responsible for the evolution of shapes of small crystallites [4] or growth of twodimensional fractal-dendritic islands [5]. 3. Molecular dynamics method – provides the complete information on the time evolution of a system of interacting atoms through the numerical integration of the equations of motion for all atoms in the system. This method is widely used in computational investigations of nanomaterials and is discussed in more detail below. 100 nm atomistic MD 10 nm 1 nm ab initio fs 3 1 ps ns μs Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling, Fig. 1 Schematic representation of the time- and length-scale domains of first-principles (ab initio) electronic structure calculations, classical atomistic MD, and continuum modeling of materials. The domain of continuum modeling can be different for different materials and corresponds to the time- and lengthscales at which the effect of the micro/nanostructure can be averaged over to yield the effective material properties. The arrows show the connections between the computational methods used in multiscale modeling of materials: The red arrow #1 corresponds to the use of quantum mechanics–based electronic structure calculations to design interatomic potentials for classical MD simulations or to verify/correct the predictions of the classical atomistic simulations; the green arrow #2 corresponds to the direct use of the predictions of large-scale atomistic simulations of nanostructured materials for the design of continuum-level constitutive relations describing the material behavior and properties; and the two blue arrows #3 show a two-step path from atomistic to continuum material description through an intermediate mesoscopic modeling the development of the defect structures or phase transformations can be performed and related to changes in temperature and pressure in the system (see examples below). The main strength of the MD method is that only details of the interatomic interactions need to be specified, and no assumptions are made about the character of the processes under study. This is an important advantage that makes MD to be capable of discovering new physical phenomena or processes in the course of “computer experiments.” Moreover, unlike in real experiments, the analysis of fast non-equilibrium processes in MD simulations can be performed with unlimited atomic-level resolution, providing complete information of the phenomena of interest. The predictive power of the MD method, however, comes at a price of a high computational cost of the simulations, leading to severe limitations on time and length scales accessible for MD simulations, as shown schematically in Fig. 1. Although the record length- Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling scale MD simulations have been demonstrated for systems containing more than 1012 atoms (corresponds to cubic samples on the order of 10 mm in size) with the use of hundreds of thousands of processors on one of the world-largest supercomputers [6], most of the systems studied in large-scale MD simulations do not exceed hundreds of nanometers even in simulations performed with computationally efficient parallel algorithms (shown by a green area extending the scales accessible for MD simulations in Fig. 1). Similarly, although the record long time-scales of up to hundreds of microseconds have been reported for simulations of protein folding performed through distributed computing [7], the duration of most of the simulations in the area of materials research does not exceed tens of nanoseconds. Molecular Dynamics Simulations of Nanomaterials Both the advantages and limitations of the MD method, briefly discussed above, have important implications for simulations of nanomaterials. The transition to the nanoscale size of the structural features can drastically change the material response to the external thermal, mechanical, or electromagnetic stimuli, making it necessary to develop new structure–properties relationships based on new mechanisms operating at the nanoscale. The MD method is in a unique position to provide a complete microscopic description of the atomic dynamics under various conditions without making any a priori assumptions on the mechanisms and processes defining the material behavior and properties. On the other hand, the limitations on the time- and length-scales accessible to MD simulations make it difficult to directly predict the macroscopic material properties that are essentially the result of a homogenization of the processes occurring at the scale of the elements of the nanostructure. Most of the MD simulations have been aimed at investigation of the behavior of individual structural elements (nanofibers, nanoparticles, interfacial regions in multiphase systems, grain boundaries, etc.). The results of these simulations, while important for the mechanistic understanding of the elementary processes at the nanoscale, are often insufficient for making a direct connection to the macroscopic behavior and properties of nanomaterials. With the fast growth of the available computing resources, however, there have been an increasing 473 C number of reports on MD simulations of systems that include multiple elements of nanostructures. A notable class of nanomaterials actively investigated in MD simulations is nanocrystalline materials – a new generation of advanced polycrystalline materials with submicron size of the grains. With a number of atoms on the order of several hundred thousands and more, it is possible to simulate a system consisting of tens of nanograins and to investigate the effective properties of the material (i.e., to make a direct link between the atomistic and continuum descriptions, as shown schematically by the green arrow #2 in Fig. 1). MD simulations of nanocrystalline materials addressing the mechanical [8, 9] and thermal transport [10] properties as well as the kinetics and mechanisms of phase transformations [11, 12] have been reported, with several examples illustrated in Fig. 2. In the first example, Fig. 2a, atomic-level analysis of the dislocation activity and grain-boundary processes occurring during mechanical deformation of an aluminum nanocrystalline system consisting of columnar grains is performed and the important role of mechanical twinning in the deformation behavior of the nanocrystalline material is revealed [9]. In the second example, Fig. 2b, the processes of void nucleation, growth and coalescence in the ductile failure of nanocrystalline copper subjected to an impact loading are investigated, providing important pieces of information necessary for the development of a predictive analytical model of the dynamic failure of nanocrystalline materials [8]. The third example, Fig. 2c, illustrates the effect of nanocrystalline structure on the mechanisms and kinetics of short pulse laser melting of thin gold films. It is shown that the initiation of melting at grain boundaries can steer the melting process along the path where the melting continues below the equilibrium melting temperature, and the crystalline regions shrink and disappear under conditions of substantial undercooling [11]. The brute force approach to the atomistic modeling of nanocrystalline materials (increase in the number of atoms in the system) has its limits in addressing the complex collective processes that involve many grains and may occur at a micrometer length scale and above. Further progress in this area may come through the development of concurrent multiscale approaches based on the use of different resolutions in the description of the intra-granular and grain boundary regions in a well-integrated computational model. An example of a multiscale approach is provided in Ref. [13], where C C 474 Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling a 7.9 ps 98.6 ps 173.5 ps b 0 ps 40 ps c 0 ps 20 ps 50 ps 150 ps Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling, Fig. 2 Snapshots from atomistic MD simulations of nanocrystalline materials: (a) mechanical deformation of nanocrystalline Al (only atoms in the twin boundaries left behind by partial dislocations and atoms in disordered regions are shown by red and blue colors, respectively) [9]; (b) spallation of nanocrystalline Cu due to the reflection of a shock wave from a surface of the sample (atoms that have local fcc, hcp, and disordered structure are shown by yellow, red, and green/blue colors, respectively) [8]; and (c) laser melting of a nanocrystalline Au film irradiated with a 200 fs laser pulse at a fluence close to the melting threshold (atoms that have local fcc surroundings are colored blue, atoms in the liquid regions are red and green, and in the snapshots for 50 and 150 ps the liquid regions are blanked to expose the remaining crystalline regions) [11] scale-dependent constitutive equations are designed for a generalized finite element method (FEM) so that the atomistic MD equations of motion are reproduced in the regions where the FEM mesh is refined down to atomic level. This and other multiscale approaches can help to focus computational efforts on the important regions of the system where the critical atomic-scale processes take place. The practical applications of the multiscale methodology so far, however, have been largely limited to investigations of individual elements of material microstructure (crack tips, interfaces, and dislocation reactions), with the regions represented with coarse-grained resolution serving the purpose of adoptive boundary conditions. The perspective of the concurrent multiscale modeling of nanocrystalline materials remains unclear due to the close coupling between the intra-granular and grain boundary processes. To enable the multiscale modeling of dynamic processes in nanocrystalline materials, the design of advanced computational descriptions of the coarsegrained parts of the model is needed so that the plastic deformation and thermal dissipation could be adequately described without switching to fully atomistic modeling. Mesoscopic Modeling A principal challenge in computer modeling of nanomaterials is presented by the gap between the atomistic description of individual structural elements and the macroscopic properties defined by the Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling collective behavior of large groups of the structural elements. Apart from a small number of exceptions (e.g., simulations of nanocrystalline materials briefly discussed above), the direct analysis of the effective properties of nanostructured materials is still out of reach for atomistic simulations. Moreover, it is often difficult to translate the large amounts of data typically generated in atomistic simulations into key physical parameters that define the macroscopic material behavior. This difficulty can be approached through the development of mesoscopic computational models capable of representing the material behavior at timeand length-scales intermediate between the atomistic and continuum levels (prefix meso comes from the Greek word mEsoB, which means middle or intermediate). The mesoscopic models provide a “stepping stone” for bridging the gap between the atomistic and continuum descriptions of the material structure, as schematically shown by the blue arrows #3 in Fig. 1. Mesoscopic models are typically designed and parameterized based on the results of atomistic simulations or experimental measurements that provide information on the internal properties and interactions between the characteristic structural elements in the material of interest. The mesoscopic simulations can be performed for systems that include multiple elements of micro/ nanostructure, thus enabling a reliable homogenization of the structural features to yield the effective macroscopic material properties. The general strategy in the development of a coarse-grained mesoscopic description of the material dynamics and properties includes the following steps: 1. Identifying the collective degrees of freedom relevant for the phenomenon under study (the focus on different properties of the same material may affect the choice of the structural elements of the model) 2. Designing, based on the results of atomic-level simulations and/or experimental data, a set of rules (or a mesoscopic force field) that governs the dynamics of the collective degrees of freedom 3. Adding a set of rules describing the changes in the properties of the dynamic elements in response to the local mechanical stresses and thermodynamic conditions While the atomistic and continuum simulation techniques are well established and extensively used, the mesoscopic modeling is still in the early development stage. There is no universal mesoscopic technique or 475 C methodology, and the current state of the art in mesoscopic simulations is characterized by the development of system- /phenomenon-specific mesoscopic models. The mesoscopic models used in materials modeling can be roughly divided into two general categories: (1) the models based on lumping together groups of atoms into larger dynamic units or particles and (2) the models that represent the material microstructure and its evolution due to thermodynamic driving forces or mechanical loading at the level of individual crystal defects. The basic ideas underlying these two general classes of mesoscopic models are briefly discussed below. The models where groups of atoms are combined into coarse-grained computational particles are practical for materials with well-defined structural hierarchy (that allows for a natural choice of the coarse-grained particles) and a relatively weak coupling between the internal atomic motions inside the coarse-grained particles and the collective motions of the particles. In contrast to atomic-level models, the atomic structure of the structural elements represented by the coarsegrained particles is not explicitly represented in this type of mesoscopic models. On the other hand, in contrast to continuum models, the coarse-grained particles allow one to explicitly reproduce the nanostructure of the material. Notable examples of mesoscopic models of this type are coarse-grained models for molecular systems [14–16] and mesoscopic models for carbon nanotubes and nanofibrous materials [17–19]. The individual molecules (or mers in polymer molecules) and nanotube/nanofiber segments are chosen as the dynamic units in these models. The collective dynamic degrees of freedom that correspond to the motion of the “mesoparticles” are explicitly accounted for in mesoscopic models, while the internal degrees of freedom are either neglected or described by a small number of internal state variables. The description of the internal states of the mesoparticles and the energy exchange between the dynamic degrees of freedom and the internal state variables becomes important for simulations of non-equilibrium phenomena that involve fast energy deposition from an external source, heat transfer, or dissipation of mechanical energy. Another group of mesoscopic models is aimed at a computationally efficient description of the evolution of the defect structures in crystalline materials. The mesoscopic models from this group include the C C 476 Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling discrete dislocation dynamics model for simulation of crystal plasticity [20–22] and a broad class of methods designed for simulation of grain growth, recrystallization, and associated microstructural evolution (e.g., phase field models, cellular automata, and kinetic Monte Carlo Potts models) [21–23]. Despite the apparent diversity of the physical principles and computational algorithms adopted in different models listed above, the common characteristic of these models is the focus on a realistic description of the behavior and properties of individual crystal defects (grain boundaries and dislocations), their interactions with each other, and the collective evolution of the totality of crystal defects responsible for the changes in the microstructure. Two examples of mesoscopic models (one for each of the two types of the models discussed above) and their relevance to the investigation of nanomaterials are considered in more detail next. Discrete Dislocation Dynamics The purpose of the discrete dislocation dynamics (DD) is to describe the plastic deformation in crystalline materials, which is largely defined by the motions, interactions, and multiplication of dislocations. Dislocations are linear crystal defects that generate longrange elastic strain fields in the surrounding elastic solid. The elastic strain field is accounting for
90% of the dislocation energy and is responsible for the interactions of dislocations among themselves and with other crystal defects. The collective behavior of dislocations in the course of plastic deformation is defined by these long-range interactions as well as by a large number of local reactions (annihilation, formation of glissile junctions or sessile dislocation segments such as Lomer or Hirth locks) occurring when the anelastic core regions of the dislocation lines come into contact with each other. The basic idea of the DD model is to solve the dynamics of the dislocation lines in elastic continuum and to include information about the local reactions. The elementary unit in the discrete dislocation dynamics method is, therefore, a segment of a dislocation. The continuous dislocation lines are discretized into segments, and the total force acting on each segment in the dislocation slip plane is calculated. The total force includes the contributions from the external force, the internal force due to the interaction with other dislocations and crystal defects that generate elastic fields, the “self-force” that can be represented by a “line tension” force for small curvature of the dislocation, the Peierls force that acts like a friction resisting the dislocation motion, and the “image” force related to the stress relaxation in the vicinity of external or internal surfaces. Once the total forces and the associated resolved shear stresses, t*, acting on the dislocation segments are calculated, the segments can be displaced in a finite difference time integration algorithm applied to the equations connecting the dislocation velocity, v, and the resolved shear stress, for example, [21]     m t DU v¼A exp  kT t0 (3) when the displacement of a dislocation segment is controlled by thermally activated events (DU is the activation energy for dislocation motion, m is the stress exponent, and t0 is the stress normalization constant) or v ¼ t b=B (4) that corresponds to the Newtonian motion equation accounting for the atomic and electron drag force during the dislocation “free flight” between the obstacles (B is the effective drag coefficient and b is the Burgers vector). Most of the applications of the DD model have been aimed at the investigation of the plastic deformation and hardening of single crystals (increase in dislocation density as a result of multiplication of dislocations present in the initial system). The extension of the DD modeling to nanomaterials is a challenging task as it requires an enhancement of the technique with a realistic description of the interactions between the dislocations and grain boundaries and/or interfaces as well as an incorporation of other mechanisms of plasticity (e.g., grain boundary sliding and twinning in nanocrystalline materials). There have only been several initial studies reporting the results of DD simulations of nanoscale metallic multilayered composites [24]. Due to the complexity of the plastic deformation mechanisms and the importance of anelastic short-range interactions among the crystal defects in nanomaterials, the development of novel hybrid computational methods combining the DD technique with other mesoscopic methods is likely to be required for realistic modeling of plastic deformation in this class of materials. Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling a 477 C b C Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling, Fig. 3 Schematic representation of the basic components of the dynamic mesoscopic model of a CNT-based nanocomposite material (a) and a corresponding molecular-level view of a part of the system where a network of CNT bundles (blue color) is embedded into an organic matrix (green and red color) (b) Mesoscopic Model for Nanofibrous Materials The design of new nanofibrous materials and composites is an area of materials research that currently experiences a rapid growth. The interest in this class of materials is fueled by a broad range of potential applications, ranging from fabrication of flexible/ stretchable electronic and acoustic devices to the design of advanced nanocomposite materials with improved mechanical properties and thermal stability. The behavior and properties of nanofibrous materials are defined by the collective dynamics of the nanofibers and, in the case of nanocomposites, their interactions with the matrix. Depending on the structure of the material and the phenomenon of interest, the number of nanofibers that has to be included in the simulation in order to ensure a reliable prediction of the effective macroscopic properties can range from several hundreds to millions. The direct atomic-level simulation of systems consisting of large groups of nanofibers (the path shown by the green arrow #2 in Fig. 1) is beyond the capabilities of modern computing facilities. Thus, an alternative two-step path from atomistic investigation of individual structural elements and interfacial properties to the continuum material description through an intermediate mesoscopic modeling (blue arrows #3 in Fig. 1) appears to be the most viable approach to modeling of nanofibrous materials. An example of a mesoscopic computational model recently designed and parameterized for carbon nanotube (CNT)-based materials is briefly discussed below. The mesoscopic model for fibrous materials and organic matrix nanocomposites adopts a coarse- grained description of the nanocomposite constituents (nano-fibers and matrix molecules), as schematically illustrated in Fig. 3. The individual CNTs are represented as chains of stretchable cylindrical segments [19], and the organic matrix is modeled by a combination of the conventional “bead-and-spring” model commonly used in polymer modeling [14, 15] and the “breathing sphere” model developed for simulation of simple molecular solids [16] and polymer solutions [25]. The degrees of freedom, for which equations of motion are solved in dynamic simulations or Metropolis Monte Carlo moves are performed in simulations aimed at finding the equilibrium structures, are the nodes defining the segments, the positions of the molecular units, and the radii of the spherical particles in the breathing sphere molecules. The potential energy of the system can be written as U ¼ UTðintÞ þ UTT þ UMM þ UMðintÞ þ UMT (5) where UT(int) is the potential that describes the internal strain energy associated with stretching and bending of individual CNTs, UT-T is the energy of intertube interactions, UM-M is the energy of chemical and nonbonding interactions in the molecular matrix, UM(int) is the internal breathing potential for the matrix units, and UM-T is the energy of matrix – CNT interaction that can include both non-bonding van der Waals interactions and chemical bonding. The internal CNT potential UT(int) is parameterized based on the results of atomistic simulations [19] and accounts for the transition to the anharmonic regime of stretching (nonlinear C 478 Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling a 0 ns b 100 nm 0.5 ns c 1 ns 3 ns d Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling, Fig. 4 Snapshots from mesoscopic simulations of systems consisting of (10,10) single-walled carbon nanotubes: (a) spontaneous self-organization of CNTs into a continuous network of CNT bundles (CNT segments are colored according to the local intertube interaction energy) [18]; (b) an enlarged view of a structural element of the CNT network (CNT segments colored according to the local radii of curvature and the red color marks the segments adjacent to buckling kinks) [26]; (c) a cross-section of a typical bundle showing a hexagonal arrangement of CNTs in the bundle [18]; (d) snapshot from a simulation of a highvelocity impact of a spherical projectile on a free-standing thin CNT film stress–strain dependence), fracture of nanotubes under tension, and bending buckling [26]. The intertube interaction term UT-T is calculated based on the tubular potential method that allows for a computationally efficient and accurate representation of van der Waals interactions between CNT segments of arbitrary lengths and orientation [18]. The general procedure used in the formulation of the tubular potential is not limited to CNTs or graphitic structures. The tubular potential (and the mesoscopic model in general) can be parameterized for a diverse range of systems consisting of various types of nano- and micro-tubular elements, such as nanotubes, nanorodes, and microfibers. First simulations performed with the mesoscopic model demonstrate that the model is capable of simulating the dynamic behavior of systems consisting of thousands of CNTs on a timescale extending up to tens of nanoseconds. In particular, simulations performed for systems composed of randomly distributed and oriented CNTs predict spontaneous self-assembly of CNTs into continuous networks of bundles with partial hexagonal ordering of CNTs in the bundles, Fig. 4a–c [18, 26]. The bending buckling of CNTs (e.g., see Fig. 4b) is found to be an important factor responsible for the stability of the network structures formed by defect-free CNTs [26]. The structures produced in the simulations are similar to the structures of CNT films and buckypaper observed in experiments. Note that an atomic-level simulation of a system similar to the one shown in the left panel of Fig. 4 would require
2.5  109 atoms, making such simulation unfeasible. Beyond the structural analysis of CNT materials, the development of the mesoscopic model opens up opportunities for investigation of a broad range of important phenomena. In particular, the dynamic nature of the model makes it possible to perform simulations of the processes occurring under conditions of fast mechanical loading (blast/impact resistance, response to the shock loading, etc.), as illustrated by a snapshot from a simulation of a high-velocity impact of a spherical projectile on a free-standing thin CNT film shown in Fig. 4d. With a proper parameterization, the mesoscopic model can also be adopted for calculation of electrical and thermal transport properties of complex nanofibrous materials [27]. Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling Future Research Directions The examples of application of the atomistic and mesoscopic computational techniques, briefly discussed above, demonstrate the ability of computer modeling to provide insights into the complex processes that define the behavior and properties of nanostructured materials. The fast advancement of experimental methods capable of probing nanostructured materials with high spatial and temporal resolution is an important factor that allows for verification of computational predictions and stimulates the improvement of the computational models. With further innovative development of computational methodology and the steady growth of the available computing resources, one can expect that both atomistic and mesoscopic modeling will continue to play an increasingly important role in nanomaterials research. In the area of atomistic simulations, the development of new improved interatomic potentials (often with the help of ab initio electronic structure calculations, red arrow #1 in Fig. 1) makes material-specific computational predictions more accurate and enables simulations of complex multi-component and multiphase systems. Further progress can be expected in two directions that are already actively pursued: (1) largescale MD simulations of the fast dynamic phenomena in nanocrystalline materials (high strain rate mechanical deformation, shock loading, impact resistance, response to fast heating, etc.) and (2) detailed investigation of the atomic structure and properties of individual structural elements in various nanomaterials (grain boundaries and interfaces, nanotubes, nanowires, and nanoparticles of various shapes). The information obtained in largescale atomistic simulations of nanocrystalline materials can be used to formulate theoretical models translating the atomic-level picture of material behavior to the constitutive relations describing the dependence of the mechanical and thermal properties of these materials on the grain size distribution and characteristics of nanotexture (green arrow #2 in Fig. 1). The results of the detailed analysis of the structural elements of the nanocomposite materials can be used in the design and parameterization of mesoscopic models, where the elementary units treated in the models correspond to building blocks of the nanostructure (elements of grain boundaries, segments of dislocations, etc.) or groups of atoms that have some distinct properties (belong to a molecule, a mer unit of 479 C a polymer chain, a nanotube, a nanoparticle in nanocomposite material, etc.). The design of novel system-specific mesoscopic models capable of bridging the gap between the atomistic modeling of structural elements of nanostructured materials and the continuum models (blue arrows #3 in Fig. 1) is likely to become an important trend in the computational investigation of nanomaterials. To achieve a realistic description of complex processes occurring in nanomaterials, the description of the elementary units of the mesoscopic models should become more flexible and sophisticated. In particular, an adequate description of the energy dissipation in nanomaterials can only be achieved if the energy exchange between the atomic degrees of freedom, excluded in the mesoscopic models, and the coarse-grained dynamic degrees of freedom is accounted for. A realistic representation of the dependence of the properties of the mesoscopic units of the models on local thermodynamic conditions can also be critical in modeling of a broad range of phenomena. In general, the optimum strategy in investigation of nanomaterials is to use a well-integrated multiscale computational approach combining the ab initio and atomistic analysis of the constituents of nanostructure with mesoscopic modeling of the collective dynamics and kinetics of the structural evolution and properties, and leading to the improved theoretical understanding of the factors controlling the effective material properties. It is the improved understanding of the connections between the processes occurring at different time- and length-scales that is likely to be the key factor defining the pace of progress in the area of computational design of new nanocomposite materials. Acknowledgment The authors acknowledge financial support provided by NSF through Grants No. CBET-1033919 and DMR0907247, and AFOSR through Grant No. FA9550-10-1-0545. Computational support was provided by NCCS at ORNL (project No. MAT009). Cross-References ▶ Ab initio DFT Simulations of Nanostructures ▶ Active Carbon Nanotube-Polymer Composites ▶ Carbon-Nanotubes ▶ Finite Element Methods for Computational Nanooptics C C 480 ▶ Mechanical Properties of Nanocrystalline Metals ▶ Modeling Thermal Properties of Carbon Nanostructure Composites ▶ Molecular Modeling on Artificial Molecular Motors ▶ Nanomechanical Properties of Nanostructures ▶ Plasticity Theory at Small Scales ▶ Reactive Empirical Bond-Order Potentials ▶ Self-Assembly of Nanostructures ▶ Vertically Aligned Carbon Nanotubes, Collective Mechanical Behavior References 1. Allen, M.P., Tildesley, D.J.: Computer Simulation of Liquids. Clarendon, Oxford (1987) 2. Frenkel, D., Smit, B.: Understanding Molecular Simulation: From Algorithms to Applications. Academic, San Diego (1996) 3. Kelires, P.C., Tersoff, J.: Equilibrium alloy properties by direct simulation: Oscillatory segregation at the Si-Ge(100) 2  1 surface. Phys. Rev. Lett. 63, 1164–1167 (1989) 4. Combe, N., Jensen, P., Pimpinelli, A.: Changing shapes in the nanoworld. Phys. Rev. Lett. 85, 110–113 (2000) 5. Liu, H., Lin, Z., Zhigilei, L.V., Reinke, P.: Fractal structures in fullerene layers: simulation of the growth process. J. Phys. Chem. C 112, 4687–4695 (2008) 6. Germann, T.C., Kadau, K.: Trillion-atom molecular dynamics becomes a reality. Int. J. Mod. Phys. C 19, 1315–1319 (2008) 7. http://folding.stanford.edu/ 8. Dongare, A.M., Rajendran, A.M., LaMattina, B., Zikry, M.A., Brenner, D.W.: Atomic scale studies of spall behavior in nanocrystalline Cu. J. Appl. Phys. 108, 113518 (2010) 9. Yamakov, V., Wolf, D., Phillpot, S.R., Mukherjee, A.K., Gleiter, H.: Dislocation processes in the deformation of nanocrystalline aluminium by molecular-dynamics simulation. Nat. Mater. 1, 45–49 (2002) 10. Ju, S., Liang, X.: Investigation of argon nanocrystalline thermal conductivity by molecular dynamics simulation. J. Appl. Phys. 108, 104307 (2010) 11. Lin, Z., Bringa, E.M., Leveugle, E., Zhigilei, L.V.: Molecular dynamics simulation of laser melting of nanocrystalline Au. J. Phys. Chem. C 114, 5686–5699 (2010) 12. Xiao, S., Hu, W., Yang, J.: Melting behaviors of nanocrystalline Ag. J. Phys. Chem. B 109, 20339–20342 (2005) 13. Rudd, R.E., Broughton, J.Q.: Coarse-grained molecular dynamics and the atomic limit of finite elements. Phys. Rev. B 58, R5893–R5896 (1998) 14. Colbourn, E.A. (ed.): Computer Simulation of Polymers. Longman Scientific and Technical, Harlow (1994) 15. Peter, C., Kremer, K.: Multiscale simulation of soft matter systems. Faraday Discuss. 144, 9–24 (2010) 16. Zhigilei, L.V., Leveugle, E., Garrison, B.J., Yingling, Y.G., Zeifman, M.I.: Computer simulations of laser ablation of molecular substrates. Chem. Rev. 103, 321–348 (2003) 17. Buehler, M.J.: Mesoscale modeling of mechanics of carbon nanotubes: self-assembly, self-folding, and fracture. J. Mater. Res. 21, 2855–2869 (2006) Computational Systems Bioinformatics for RNAi 18. Volkov, A.N., Zhigilei, L.V.: Mesoscopic interaction potential for carbon nanotubes of arbitrary length and orientation. J. Phys. Chem. C 114, 5513–5531 (2010) 19. Zhigilei, L.V., Wei, C., Srivastava, D.: Mesoscopic model for dynamic simulations of carbon nanotubes. Phys. Rev. B 71, 165417 (2005) 20. Groh, S., Zbib, H.M.: Advances in discrete dislocations dynamics and multiscale modeling. J. Eng. Mater. Technol. 131, 041209 (2009) 21. Kirchner, H.O., Kubin, L.P., Pontikis, V. (eds.): Computer simulation in materials science. Nano/meso/macroscopic space and time scales. Kluwer, Dordrecht (1996) 22. Raabe, D.: Computational materials science: the simulation of materials microstructures and properties. Wiley-VCH, Weinheim, New York (1998) 23. Holm, E.A., Battaile, C.C.: The computer simulation of microstructural evolution. JOM-J. Min. Met. Mat. S. 53, 20–23 (2001) 24. Akasheh, F., Zbib, H.M., Hirth, J.P., Hoagland, R.G., Misra, A.: Dislocation dynamics analysis of dislocation intersections in nanoscale metallic multilayered composites. J. Appl. Phys. 101, 084314 (2007) 25. Leveugle, E., Zhigilei, L.V.: Molecular dynamics simulation study of the ejection and transport of polymer molecules in matrix-assisted pulsed laser evaporation. J. Appl. Phys. 102, 074914 (2007) 26. Volkov, A.N., Zhigilei, L.V.: Structural stability of carbon nanotube films: the role of bending buckling. ACS Nano 4, 6187–6195 (2010) 27. Volkov, A.N., Zhigilei, L.V.: Scaling laws and mesoscopic modeling of thermal conductivity in carbon nanotube materials. Phys. Rev. Lett. 104, 215902 (2010) Computational Systems Bioinformatics for RNAi Zheng Yin, Yubo Fan and Stephen TC Wong Center for Bioengineering and Informatics, Department of Systems Medicine and Bioengineering, The Methodist Hospital Research Institute, Weill Cornell Medical College, Houston, TX, USA Synonyms Automatic data analysis workflow for RNAi; Systems level data mining for RNAi Definition Computational systems bioinformatics for RNAi screening and therapeutics is defined as complete Computational Systems Bioinformatics for RNAi computational workflow applicable to the hypothesis generation from large-scale data from image-based RNAi screenings as well as the improvement of RNAi-based therapeutics; the workflow includes automatic image analysis compatible to large-scale cell image data, together with unbiased statistical analysis and gene function annotation. Introduction RNA interference (RNAi) defines the phenomenon of small RNA molecules binding to its complementary sequence in certain messenger RNA, recruiting a specific protein complex to dissect the whole mRNA and thus silencing the expression of the corresponding gene. It is a highly conserved system within living cells to quantitatively control the activity of genes. In 1998, Fire et al. first clarified the causality of this phenomenon and named it as RNAi [1], and the following decade saw RNAi evolving into a powerful tool for gene function study. In 2006, Fire and Mello were awarded Nobel Prize in Physiology or Medicine; and by 2007, scientific papers using high-throughput screening based on RNAi kept piling up while clinical trials of RNAi-based therapeutics on various diseases raised the expectation on a trend of soon-to-come “super drugs.” Unfortunately, by late 2009 nearly all the first trend clinical trials have been terminated, meanwhile, researchers are struggling to effectively quantify the high-content information obtained from RNAi-based screening experiments. Although facing obstacles, it is still believed that the combination of nanotechnology and systems biology would restore and amplify the glory of RNAi on both research and therapeutic areas. This entry will summarize the challenges facing RNAi-based therapeutics – especially the difficulty of delivery and some possible solution through chemoinformatics in nanometer scale; also difficulties facing RNAi-based high-content screening will be reviewed with suggestions on possible solutions. RNAi-Based Therapeutics: How-to and What-to Deliver Currently, various RNAi-based therapeutics are being tested in clinical trials but before the application 481 C becomes clinical, several problems have first to be solved. One of the critical challenges is how to specifically and effectively deliver the objects into the targeted cells. Serious side-effects in patients could be caused by off-target effects or immune response as reported in literatures [2, 3]. Apparently, the therapeutic goal is to achieve RNAi therapy, that is, systemically administered nucleic acids must survive in circulation long enough to reach their target tissue, enter the desired cells, “escape” their endosome or delivery packaging, and finally become incorporated into the RNA-induced silencing complex (RISC) – a towering task, and surprisingly, researchers have advanced a number of plausible solutions in recent years, including the use of specialized nanoparticle filled with an RNAi-based cancer therapy to target human cancer cells and silence the target gene. However, what making the task even more difficult for therapeutics is: everything now happens in vivo. Chemoinformatics Solutions Libraries have been built based on natural products and combinatorial chemistry with millions of compounds to date. The compound library is used to screen and locate small molecules to bind a particular protein, RNA or DNA. Virtual screening utilizing molecule fragments even can design unknown compounds with reasonable binding affinity to desired targets. However, these large-scale screening compounds tend to bind multiple targets (off-target effects) even after optimizations [4]. Also, it is virtually impossible to derive a solid algorithm or theory to screen and locate all bindings and inhibitions and their combinations every protein currently know. The lack of genomescale coverage is a clear disadvantage of a compound based assay. Along with the lack of specificity comes the challenge of on-target potency. Even after a target has been identified the drug-like compounds have to be further engineered to increase efficiency and decrease off-target activities. To a large extent, chemoinformatic approaches, where the target of a compound can be predicted by in silico alignment, modeling algorithms, and virtual screening, provide a more efficient way to screen compounds to the desired targets. A successful application has been reported for the inhibition of SARS protease C C 482 [5]. The chemoinformatics approach excels at optimization around a small number of well defined targets, while RNAi approaches can more readily identify unknown pathways and phenotypes with no prior knowledge of the target. Data Analysis for RNAi HCS: Challenge from Millions of Cells Large volumes of datasets generated from RNAi HCS of RNAi prohibits manual or even semi-manual analysis; thus, automated data analysis is desperately needed [6, 7]. In the context of genome-wide RNAi HCS using cultured cells from Drosophila, a series of automated methods on cell image processing [8], online phenotype discovery [9, 10], cell classification, and gene function annotation [11] have been developed. All these methods are integrated into an automated data analysis pipeline, G-CELLIQ (Genomic CELLular Image Quantitator), to support genome-wide RNAi HCS. A lot of decisions need to be made en route to a genome-scale RNAi screens, including the selections of appropriate animal models, reagents for igniting RNAi, screening formats, and type of readouts (see [12] for a review). The focus here is arrayed highcontent RNAi screening, a systematic screen with reagents spotted in 384- or 96-well plates and where each gene or gene group is knocked down individually. The readout of HCS is obtained through microscopy, which captures multiple phenotypic features simultaneously [13]. Compared with other screening approaches, RNAi HCS can offer broader insight into cellular physiology and provide informative and continuous phenotypic data generated by RNAi. However, the automated processing and analysis of the magnitude of image data presents great challenges to applying such findings at the genome level. Image Processing: Cell Segmentation and Quantification The scale of datasets has always been a huge obstacle. For example, in [6], approximately 17,000 overlapping cells were segmented semi-manually across 10 months from an HCS targeting around 200 genes or gene combinations. However, such low-dimensional screens are not genome-wide, as they can only cover 1–2% of the genome. For automatic data analysis, the following work needs to be accomplished. Computational Systems Bioinformatics for RNAi 1. Preprocessing: where empty images, incomplete cells, and other artifacts are identified and discarded 2. Cell segmentation: where dense or overlapping cells are segmented accurately 3. Feature extraction: where informative features are extracted to quantify cell morphology Cell segmentation is the cornerstone for the whole data analysis workflow. While existing image analysis methods can handle the processing of standard images, they are limited in their scope and capability to handle genome-wide RNAi HCS analysis. Thresholding methods basically set a cutoff on the intensity of pixels and classify them into background and foreground (cell), and they may fail due to uneven background and illumination levels. Rule-based correction on over- and under-segmentation starts from relatively simple segmentation methods (like watershed), and use a series of heuristic rules, like distance between neighborhood nuclei and properties of putative cell boundaries, such methods suffer from difficulties in devising rules to merge the cell cytoplasm. Phenotype Identification, Validation, and Classification Given the quantified features describing cell morphology, the following work is essential to address the biological function of morphological profiles. 1. Define biologically meaningful phenotypes to compose cell populations based on single cell morphology. 2. Model existing phenotypes and identify novel phenotypes online to continuously generate new data. 3. Assign cellular phenotypes into different subphenotypes to address morphological changes caused by RNAi treatment. Statistical tests, artificial neural networks [6], Support Vector Machine-Recursive Feature Elimination (SVM-RFE) [14], genetic algorithms, and various other methods have been used to select or extract informative subsets of features and model certain phenotypes. However, phenotypes are usually defined a priori from pilot datasets. Human intervention is currently necessary for image-based datasets of genetic or chemical perturbations where the dynamic range of cellular phenotypes cannot be predicted before data collection. Failing to accurately measure phenotypic variations will cause concomitant classification errors and mislead functional analysis; and it is impossible to perform manual analysis during the Computational Systems Bioinformatics for RNAi screening process where millions of images are acquired. Thus, the ability of these screens to identify new phenotypes is greatly limited [9]. Statistical Analysis and Gene Function Annotation RNAi HCS inherits various statistical questions from traditional high-throughput screening, and the properties of image readouts raise specific challenges. 1. Summarizing gene function scores from quantified morphological change 2. Data triage and normalization based on readout from positive and negative control wells 3. Repeatability test and consolidation of scores from biological replicates 4. Cluster analysis, visualization, and biological interpretation of results A comprehensive review [15] summarizes different dataset generated by RNAi screens and traditional small molecule screens. It also reviews statistical analysis methods applicable to most of problems outlined above. Quantitative morphological signatures (QMS) [6] represent the efforts on interpreting RNAi HCS datasets: use the similarity score to a panel of existing phenotypes to explore the broad phenotypic space. However, problems remain open when the discriminative ability of such scores is confounded by multiple phenotypes following a single RNAi treatment, so repeatability tests can become more complicated [6]. The use of publicly available databases on drugs and disease-related biological processes to interpret RNAi HCS datasets also remains unresolved. G-CELLIQ: An Integrated Automated Data Analysis Tool for RNAi High-Content Screening Computational Architecture of G-CELLIQ G-CELLIQ (Genomic CELLular Imaging Quantitator) is developed to process large volumes of digital images generated from large-scale HCS studies. The workflow for G-CELLIQ can be simplified as “three modules handling three databases.” Images generated from HCS are stored in a Raw image database; the Image processing/cell morphology quantification module segments each image into single cells and creates a quantified cell database; the Phenotype modeling and cell classification module compares each cell’s 483 C morphology to a panel of “reference” phenotypes (which can be defined both manually and automatically) and generates a morphology score; the annotation of gene function module then summarize single cell scores into scores for cell populations, images and wells, and the consolidated scores for involved genes form a single gene function profile database. Image Processing and Cell Morphology Quantification A segmentation method consisting of nuclear segmentation, cell body segmentation, and over-segmentation correction [8] is used in G-CELLIQ. Each cell body is described by 211 morphological features; automatic image quality control is applied to filter images where the signal from certain channels is extremely dark or bright or cells are located at the edge of an image. Nuclear segmentation cell nuclei are first separated from background using a binarization method which implements an adaptive thresholding [8]. To segment clustered nuclei, the nuclei shape and intensity information are integrated into a combined image and processed with Gaussian filter and the nuclei centers are detected as local maxima in the gradient vector field (GVF); after that, nuclei are segmented using the marker-controlled seeded-watershed method. Cell body segmentation Preliminary cell body segmentation is done using an adaptive thresholding algorithm. Due to the large size of HCS dataset, seeded-watershed method is used to segment the touching cell bodies. Results from nuclei segmentation are used as the seed information [16]. Over-segmentation correction is necessary when there are multiple nuclei existing within cells. For cell segments smaller than a given size threshold, its neighboring cell segments sharing the longest common boundary are determined and the intensity variation in a rectangular region across the common boundary of touching cell segments are calculated. If the intensity variation was smaller than a given threshold, the corresponding cell segments are merged. Feature extraction The detailed shape and boundary information of nuclei and cell bodies is obtained through the proposed segmentation method. To capture the geometric and appearance properties, 211 morphology features belonging to five categories were extracted following [11]. C C 484 Online Phenotype Discovery, Phenotype Modeling, and Cell Classification Expert opinion is implemented to identify a panel of reference phenotypes while candidate groups of informative features are selected to describe typical phenotypes. SVM-RFE and GA-SVM method are used to select the feature sets with the best performance for cross validation. A series of SVM classifiers are trained to differentiate the reference phenotypes from all others, and a continuous value rather than binary class label is used as the output of SVM to indicate each cell’s morphological similarity to typical phenotypes. A novel method is designed to do online phenotype discovery [9, 10]; it is based on online phenotype modeling and iterative phenotype merging. For the modeling part, each existing phenotype is modeled through a Gaussian Mixture Model (GMM), and each model is continuously updated according to information of newly incorporated cells with a minimum classification error (MCE) method. For the merging part, the newly generated cell population is iteratively combined with each existing phenotype, one at a time. Then an improved gap statistics method is used to identify the number of possible phenotypes in the combination. Through cluster analysis, some of the cells in the new populations are assigned into the same cluster as samples from existing phenotypes, and those cells are merged by existing phenotypes to help update the phenotype models. On the other hand, some cells are never merged and remain as candidate novel phenotype for validation by statistical tests [9, 10]. Annotation of Gene Function After assigning each cell a score vector corresponding to its similarity to reference phenotypes, all cells in control condition are pooled to model a baseline for cell morphology. All morphology scores are normalized to the Z-score relative to this control baseline. The scores for the qualified cells are then averaged to form scores for each well. In order to select repeatable wells from those undergoing the same dsRNA treatment, a series of repeatability tests are applied to scores for different well. The weighted average of the scores is then calculated for the repeatable wells and generated scores for each treatment condition (TC). Similar procedure consolidates the score from biological replicate TCs to form a score vector for each gene. Hierarchical Computational Systems Bioinformatics for RNAi clustering is implemented group genes with similar function scores into the same pheno-cluster. RNAi HCS Applying G-CELLIQ Since more than 75% of human disease genes have Drosophila orthologs, the pursuit of genes involved in normal fly morphogenesis and migration is expected to reveal mechanisms conserved in humans [17]. An example of using G-CELLIQ in the context of RNAi HCS using cultured Drosophila cell lines is presented next. Regulatory of cell shape change A genome-scale RNAi screening is carried out for regulators of Drosophila cell shape. Drosophila Kc187 cell lines are utilized in the screen and wild-type cells have hemocyte-like properties. Using dsRNA to target and inhibit the activity of specific genes/proteins, the role of individual genes in regulating morphology can be systematically determined. Work in [9, 10] relies on part of the genome dataset to target the group of kinase-phosphatases in order to develop and validate online phenotype discovery methods. A panel of five existing phenotypes is set by expert labeling and online phenotype discovery. In order to address the level of penetrance in this dataset, the phenotypic scores for single cells are sorted according to similarity to wild-type cells, and if a certain well has significantly less (at least one standard-deviation) wild-type cells than control wells, wild-type cells from this well are removed to reveal the phenotypic change relating to RNAi treatment. Roles of Rho family small GTPases in development and cancer Three automated and quantitative genomewide screens for dsRNAs that induce the loss of the Rho-induced cytoskeletal structures (lamellipodia, filopodia, and stress fibers) are performed to identify putative Rho protein effectors. Candidate effectors identified in such image-based screens are readily validated in the context of the whole organism using the large number of mutant fly lines coupled with the vast arrays of in vivo techniques available to fly biologists [18]. Following image segmentation, feature extraction, classification of single cells, and scoring of individual wells, an image descriptor is assigned to each gene. Hierarchical clustering can then be used to cluster the Concentration Polarization image descriptor and identify groups of genes when targeted by RNAi result in quantitatively similar morphologies. Previous studies reported in [6] demonstrated that clustering results identifies groups of functionally related genes that operate in similar signaling pathways. These groups of genes are termed as “Pheno-clusters” [19]. To validate the hypothesis generation ability of G-CELLIQ, 32 dsRNAs/wells are randomly selected from a dataset where individual kinases and phosphatases were inhibited by RNAi. Each dsRNA/well was assigned an image descriptor, and hierarchical clustering was used to group genes/ wells. dsRNAs in this analysis clustered into two broad groups. One group of 19 conditions included 10/10 control conditions, as well as dsRNAs targeting the Insulin receptor (InR). Strikingly, the other large cluster of 13 conditions included 3/3 dsRNAs previously identified in a genome-wide screen for regulators of MAPK/ERK activation downstream of the EGF/EGFR activity [20]. These results demonstrate that automated high-throughput imaging can discriminate distinct morphologies and be used to model functional relationships between signaling molecules. Cross-References ▶ RNAi in Biomedicine and Drug Delivery References 1. Fire, A., et al.: Potent and specific genetic interference by double-stranded RNA in Caenorhabditis elegans. Nature 391(6669), 806–811 (1998) 2. Hornung, V., et al.: Sequence-specific potent induction of IFN-[alpha] by short interfering RNA in plasmacytoid dendritic cells through TLR7. Nat. Med. 11(3), 263–270 (2005) 3. Grimm, D., et al.: Fatality in mice due to oversaturation of cellular microRNA/short hairpin RNA pathways. Nature 441(7092), 537–541 (2006) 4. Copeland, R.A., Pompliano, D.L., Meek, T.D.: Drug-target residence time and its implications for lead optimization. Nat. Rev. Drug Discov. 5(9), 730–739 (2006) 5. Plewczynski, D., et al.: In silico prediction of SARS protease inhibitors by virtual high throughput screening. Chem. Biol. Drug Des. 69(4), 269–279 (2007) 6. Bakal, C., et al.: Quantitative morphological signatures define local signaling networks regulating cell morphology. Science 316, 1753–1756 (2007) 485 C 7. Zhou, X., Wong, S.T.C.: Computational systems bioinformatics and bioimaging for pathway analysis and drug screening. Proc. IEEE 96(8), 1310–1331 (2008) 8. Li, F.H., et al.: High content image analysis for human H4 neuroglioma cells exposed to CuO nanoparticles. BMC Biotechnol. 7, 66 (2007) 9. Yin, Z., et al.: Using iterative cluster merging with improved gap statistics to perform online phenotype discovery in the context of high-throughput RNAi screens. BMC Bioinformatics 9(1), 264 (2008) 10. Yin, Z., et al.: Online phenotype discovery based on minimum classification error model. Pattern Recogn. 42(4), 509–522 (2009) 11. Wang, J., et al.: Cellular phenotype recognition for high-content RNA interference genome-wide screening. J. Mol. Screen. 13(1), 29–39 (2008) 12. Perrimon, N., Mathey-Prevot, B.: Applications of high-throughput RNAi screens to problems in cell and developmental biology. Genetics 175, 7–16 (2007) 13. Carpenter, A.E., Sabatini, D.M.: Systematic genome-wide screens of gene function. Nat. Rev. Genet. 5(1), 11–22 (2004) 14. Loo, L., Wu, L., Altshuler, S.: Image based multivariate profiling of drug responses from single cells. Nat. Methods 4(5), 445–453 (2007) 15. Birmingham, A., et al.: Statistical methods for analysis of high-throughput RNA interference screens. Nat. Methods 6(8), 569–575 (2009) 16. Yan, P., et al.: Automatic segmentation of RNAi fluorescent cellular images with interaction model. IEEE Trans. Inf. Technol. Biomed. 12(1), 109–117 (2008) 17. Reiter, L.T., et al.: A systematic analysis of human diseaseassociated gene sequences in Drosophila melanogaster. Genome Res. 11, 1114–1125 (2001) 18. Bier, E.: Drosophila, the golden bug, emerges as a tool for human genetics. Nat. Rev. Genet. 6(1), 9–23 (2005) 19. Piano, F., et al.: Gene clustering based on RNAi phenotypes of ovary-enriched genes in C. elegans. Curr. Biol. 12(22), 1959–1964 (2002) 20. Friedman, A., Perrimon, N.: Functional genomic RNAi screen for novel regulators of RTK/ERK signaling. Nature 444, 230–234 (2006) Computer Modeling and Simulation of Materials ▶ Computational Study of Nanomaterials: From Large-Scale Atomistic Simulations to Mesoscopic Modeling Concentration Polarization ▶ Concentration Polarization at Micro/Nanofluidic Interfaces C C 486 Concentration Polarization at Micro/ Nanofluidic Interfaces Vishal V. R. Nandigana and N. R. Aluru Department of Mechanical Science and Engineering, Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana – Champaign, Urbana, IL, USA Concentration Polarization at Micro/Nanofluidic Interfaces ions get strongly attracted toward the channel surface due to the electrostatic force is called the inner layer with a typical thickness of one ion diameter. The outer Helmholtz plane separating the liquid and the diffusive layer constitute the liquid side part of the EDL. The ionic species in the diffusive layer are influenced by the local electrostatic potential, and the species distribution at equilibrium can be described by the Boltzmann equation. The thickness of the diffusive layer spans between 1 and 100 nm. The EDL thickness (lD) is given by [1]: Synonyms Concentration polarization; Micro/nanofluidic devices; Nonlinear electrokinetic transport Definition Concentration polarization (CP) is a complex phenomenon observed at the interfaces of micro/nanofluidic devices due to the formation of significant concentration gradients in the electrolyte solution resulting in accumulation and depletion of ions near the interfaces. This chapter provides an overview of the underlying theory and physics that is predominantly observed on the integration of microfluidic channels with nanofluidic devices. Nonlinear electrokinetic transport and concentration polarization phenomenon are discussed in detail along with recent advancements in utilizing this phenomenon for designing novel devices. Electrical Double Layer (EDL) and Electroosmotic Flow A solid in contact with an aqueous solution acquires a surface charge ðss Þ due to the dissociation of ionizable groups on the solid walls. The fixed surface charge on the solid surface in contact with the liquid develops a region of counterions (ions with charges opposite to the solid surface) in the liquid to maintain the electroneutrality at the solid–liquid interface. This screening region is denoted as the electrical double layer (EDL) or Debye length (DL). For instance, in the case of silica channel with KCl electrolyte solution, the dissociation of silanol groups would make the channel negatively charged and affect the distribution of K þ counterions in the solution. The layer where the lD ¼  e0 er RT P 2 F2 m i¼1 zi c0 1=2 (1) where F is Faraday’s constant, zi is the valence of ionic species i, c0 is the bulk concentration of the electrolyte solution, e0 is the permittivity of free space, er is the relative permittivity of the medium, m is the total number of ionic species, R is the universal gas constant, and T is the absolute temperature. The thickness of the EDL plays a significant role in the transport of miniaturized devices. In microfluidic channels, the fluid transport is often controlled by electric fields, as it eliminates the use of external mechanical devices [1]. The electric field aids better control when compared to using pressure-driven techniques. Furthermore, the electric fields overcome the high pressures needed to transport the fluid at such length scales as the pressure follows a power–law relation with respect to the height (h) of the channel. The electric field acts on the charged counterions present at the interface of solution and stationary charged wall (i.e., at the EDL regions), resulting in the motion of the fluid which is referred as electroosmotic flow (EOF) [1]. As the EDL thickness in these  channels is much smaller compared to their height lhD 1 , the fluid flow has a plug-like flow characteristic. However, recent advancements in the fabrication technology [2] have motivated researchers around the globe to investigate the transport phenomenon in channel sizes of the order of few hundreds of nanometers. Transport in these devices is referred to as “nanofluidics.” The electrical double layer in these devices spans much of the diameter or channel height leading to many interesting transport phenomena compared to its microscopic counterpart. The electroosmotic velocity no longer follows a plug-like flow characteristic but follows a Poiseuille-like (parabolic) characteristic as the Concentration Polarization at Micro/Nanofluidic Interfaces electrokinetic body force is not just confined to a thin layer adjacent to the channel surface. Along with the aforementioned difference, the micro and nanofluidic systems also exhibit a different ion transport characteristic which is discussed below. In nanofluidic systems, as the EDL  thickness becomes comparable to the channel height lhD  1 , there is a predominant transport of the counterions inside the channel, thus enabling the channel to be ion-selective [1]. These features are not observed in the microfluidic channels, as the counter-ionic space charge is confined to a very thin layer adjacent to the surface and the region away from the surface is essentially quasi-electroneutral (i.e., both co-ions and counterions are present away from the surface). Along with the EDL, the surface charge also plays a prominent role in controlling the transport inside the nanofluidic systems [2]. Similar ion-selective phenomenon was also observed in the intraparticle and intraskeleton mesopores of particulate and in membrane science [3]. Owing to the differences in the electrokinetic transport phenomena between the micro- and nanofluidic devices, the integration of these two devices paves way to complex physics. The models and the underlying theory developed to understand the electrokinetic transport in such systems are elaborated in section “Electrokinetic Theory for Micro/Nanochannels.” A detailed discussion on the concentration polarization phenomenon and its applications are presented in section “Concentration Polarization.” Finally, a brief summary is presented in section “Summary.” Electrokinetic Theory for Micro/ Nanochannels A complete set of equations for modeling the electrokinetic transport and to account for the EDL effects in micro/nanofluidic channels are presented. To understand the electrokinetic transport, space charge model developed by Gross et al. [4] is used extensively in the literature. The model solves the classical Poisson– Nernst–Planck (PNP) equations, which describe the electrochemical transport and the incompressible Navier–Stokes along with the continuity equations are solved to describe the movement of the fluid flow. These coupled systems of equations are more intensive, mathematically complicated, and computationally expensive. Though many linearized approximations 487 C were proposed to this model to study the electrokinetic transport [2], the governing equations of the complete nonlinear space charge model is discussed in this chapter. In electrokinetic flows, the total flux is contributed by three terms: a diffusive component resulting from the concentration gradient, an electrophoretic component arising due to the potential gradient, and a convective component originating from the fluid flow. The total flux of each species in the solution is given by G i ¼ Di Hci  Oi zi Fci Hf þ ci u (2) where G i is the flux vector, Di is the diffusion coefficient, Oi is the ionic mobility, ci is the concentration of the ith species, u is the velocity vector of the fluid flow, and f is the electrical potential. Note that the ionic mobility is related to the diffusion coefficient by Di [5]. The electrical potenEinstein’s relation, Oi ¼ RT tial distribution is calculated by solving the Poisson equation, H ð2r HfÞ ¼  re 20 (3) where re is the net space charge density of the ions defined as re ¼ F m X zi ci i¼1 ! (4) The mass transfer of each buffer species is given by the Nernst–Planck equation, @ci ¼ H G i @t (5) Equations 3, 5, and 2 are the classical Poisson– Nernst–Planck (PNP) equations, which describe the electrochemical transport. The incompressible Navier–Stokes and the continuity equations are considered to describe the movement of the fluid flow through the channel, i.e.,   @u r þ u Hu ¼ Hp þ mH2 u þ re E @t (6) H u¼0 (7) C C 488 Concentration Polarization at Micro/Nanofluidic Interfaces Enrichment region Receiver Depletion region Nanochannel Cathode (−) Source Anode (+) Concentration Polarization at Micro/Nanofluidic Interfaces, Fig. 1 Schematic illustration of ion-enrichment and ion-depletion effect in cation-selective micro/nanofluidic channel. The solid arrows indicate the flux of cations and the dotted arrows indicate the flux of anions. At the nanochannel– anodic junction, both the anions and cations are depleted, while there is an enhancement of both the ions at the cathode– nanochannel junction where u is the velocity vector, p is the pressure, r and m are the density and the viscosity of the fluid, respectively, and E ¼ Hf is the electric field. re E is the electrostatic body force acting on the fluid due to the space charge density and the applied electric field. Elaborate details on other simplified models are discussed in the review article of Schoch et al. [2]. Enrichment/Depletion Effects In micro/nanofluidic devices, Pu et al. [8] first experimentally observed the CP effects near the interfaces and provided a simple model to explain the accumulation and depletion physics which is summarized below. For a negatively charged nanochannel, the EDL would be positively charged. For an overlapped EDL, as discussed in section “Electrical Double Layer (EDL) and Electroosmotic Flow,” the nanochannel becomes ion-selective, resulting in higher cation concentration than anions. Thus, the flux of cations is higher compared to the anions in the nanochannel. With the application of positive potential at the source microchannel or reservoir (see Fig. 1), the cations move from the source (anode) reservoir to the receiving (cathode) reservoir end, while the anions move in the opposite direction through the nanochannel. At the cathodic side, the anion flux from the ends of reservoir to the nanochannel junction is higher compared to the anion flux from the junctions to the nanochannel as the anions are repelled by the negatively charged nanochannel. This difference in fluxes causes an accumulation of anions at the cathode–nanochannel junction. The cation flux from the nanochannel to the cathode junction is greater than from the cathode junction to the reservoir as the cations have to balance the anions present at this junction. This results in an accumulation of cations as well at the nanochannel–cathode junction. At the anodic side, the anion flux from the nanochannel to the anode junction cannot balance the anion flux from the anode–nanochannel junction to the reservoirs due to the limited anions passing through the nanochannel. This results in a depletion of anions at this junction. The cation flux from the Concentration Polarization Concentration polarization (CP) is a complex phenomenon observed at the interface regions of micro/ nanofluidic devices due to the formation of significant concentration gradients in the electrolyte solution near the interfaces causing accumulation of ions on the cathodic side and depletion of ions on the anodic side for a negatively charged nanochannel surface. This phenomenon was also observed in the field of colloid science and in membrane science which was extensively studied for over 40 years, and the early works of CP phenomenon is comprehensively reviewed by Rubinstein et al. [6]. The pioneering works from Rubinstein and his coworkers had revealed electrokinetic instabilities [7] in the concentration polarization regions leading to the breakdown of limiting current and resulting in the overlimiting conductance regimes in the ion exchange membranes. Such complex phenomenon could not be postulated using the classical equilibrium model of EDL [6]. All the underlying CP physics observed near the interfaces of micro/ nanochannels are summarized in the following subsections. Concentration Polarization at Micro/Nanofluidic Interfaces reservoir to the anode–nanochannel junction is less than the cation flux entering the nanochannel as the cations are attracted by the positively charged nanochannel. This in turn leads to the depletion of cations at this junction. To summarize, for a negatively charged nanochannel, both the cations and anions accumulate at the cathodic interface and are depleted at the anodic interface. The phenomenon is reversed for a positively charged nanochannel surface. A schematic diagram highlighting the accumulation and the depletion physics for a negatively charged nanochannel is displayed in Fig. 1. Nonlinear Electroosmosis As discussed in the previous section, the integration of micro/nanofluidic devices leads to the accumulation and depletion of ions near the interfaces. Several numerical and experimental studies were carried out to gain a better understanding of the physics at these interfaces. The studies revealed complex and interesting physics at the depletion interface compared to the enrichment side. Rubinstein et al. [6] theoretically predicted the presence of space charges at the depletion region under large electric fields. The presence of the induced space charges near the depletion interface results in a nonequilibrium electrical double layer outside the nanochannel. The induced space charges under the action of the external electric field lead to nonlinear electroosmosis or otherwise known as electroosmosis of the second kind. The electroosmotic flow of the second kind was found to be directly proportional to the square of the applied electric field. Furthermore, the induced space charges also result in the generation of vortices at the depletion interface along with inducing large pressure and voltage gradients at this junction. Jin et al. [9] also reported similar physics from their extensive numerical study. In the case of a flat ion exchange membrane, Rubinstein et al. [7] derived a 2D nonequilibrium electroosmotic slip ðus Þ for an applied voltage ðVÞ using the linear stability analysis to impose strong vortex field near the membrane: @2 c 1 @x@y us ¼  V 2 @c 8 @y (8) where x and y are the axes parallel and perpendicular to the ion-exchange membrane, respectively. Experiments performed by Kim et al. [10] also reveal the 489 C nonequilibrium EOF near the micro/nanofluidic junctions. The application of electric field on the surface of particles also results in such induced space charges which spread over a larger region than the primary EDL resulting in highly chaotic flow patterns. A recent review by Höltzel and Tallarek [3] provide a detailed discussion on the polarization effects around membranes, packed beds, and glass monoliths. Recent advancements by Rubinstein and Zaltzman [11], however, revealed that the extended space charge region was not a part of the EDL, but develop from the counterion concentration minimum zone with the coions expelled under the action of the electric field. They further claim that the space charges would be present in the system even without equilibrium EDL. Their analysis included the study of the space charge dynamics in concentration polarization regions using 1D and three-layer models of EDL. From the understanding of these extended space charge layers, another important phenomenon, namely, the nonlinear current characteristics in micro/nanochannels is addressed below. Nonlinear Current–Voltage Characteristics There has been growing interest in the development of micro/nanofluidic devices as ionic filters and nanofluidic batteries to control both ionic and molecular transport in aqueous solutions [2]. As discussed before, when the diameter/height of a charged channel scales comparable to the EDL, there is a predominant transport of the counterions inside the channel. Thus, the transport of electrical current inside the nanochannel is primarily due to the counterions. This feature enables the micro/nanochannel to be used as an ion exchange membrane. However, understanding the passage of ionic currents through such ion-selective solids is the most fundamental physical problem that has stimulated extensive research in this field for over a decade. Furthermore, the concentration polarization physics near the interfaces play a pivotal role in understanding the current–voltage characteristics. Figure 2a shows that the concentration gradients (CP regions) near the interfaces become steeper with the decrease in the ionic strength ðcs Þ and at higher electric fields ðEx Þ. At low electric fields, the current increases linearly with the applied voltage following the Ohm’s law (region I in Fig. 2b). However, at higher electric fields, the ion concentration in the depleted CP zone (i.e., near the anodic C 490 Concentration Polarization at Micro/Nanofluidic Interfaces Cation-selective nanochannel a Ex b Ex Co Cs Current density C Co Cs Overlimiting region III Limiting region II I Ohmic region Ex x Cathodic DBL Anodic DBL Voltage Cation-selective nanochannel c Ex B A Co Co A B x Anodic DBL SCR Cathodic DBL Concentration Polarization at Micro/Nanofluidic Interfaces, Fig. 2 (a) Schematic distribution of ionic concentration in equilibrium concentration polarization under axial electric field (Ex). The local electroneutrality is maintained at both enrichment (cathodic interface) and depletion (anodic interface) diffusion boundary layers (DBL). The concentration gradients become steeper with the decrease in the ionic strengths (cs) and at higher electric fields, (b) displays the nonlinear current–voltage characteristics for an ion-selective micro/nanochannel, and (c) shows the nonequilibrium concentration distribution due to the induced space charge region (SCR) (shown as dotted lines (B)) in the depleted region under very large electric fields interface region) reaches toward zero and the classical Levich analysis [5] predicts a diffusion-limited current saturation according to which a saturation of current density occurs at a constant level described as the “limiting-current density” (region II in Fig. 2b). The ionic current can be calculated considering the Fick’s first law: where c0 is the bulk electrolyte solution and d is the diffusion boundary layer (DBL) thickness at the solid–liquid interface and cðx ¼ 0Þ represents the concentration at the anodic (depletion) solid–liquid interface. d typically ranges between 10 and 400 mm in ion exchange membranes [6], while it depends on the microchannel length in the case of micro/ nanochannels [10]. From Eq. 10 it is clear that the current I reaches a maximum value when the concentration at cðx ¼ 0Þ ¼ 0, resulting in a limiting/saturation current as predicted by the classical Levich theory. I ¼ nFAD dc dx (9) where n is the number of electrons transferred per molecule, D is the diffusion coefficient, and A is the electrode surface area. The concentration gradient is generally approximated by a linear variation (5), I ¼ nFAD c0  cðx ¼ 0Þ d (10) Ilim ¼ nFADc0 d (11) However, experimental studies in micro/ nanofluidic devices (also in membrane science) revealed ionic currents larger than the limiting value and this regime was termed as the overlimiting current Concentration Polarization at Micro/Nanofluidic Interfaces regime (region III in Fig. 2b). Further, the limiting current region is termed as the limiting resistance region (in micro/nanofluidic devices) due to the large but finite limiting differential resistance as the current does not saturate to a limiting value but has a slope which is smaller than the ohmic region. The nonlinear current characteristics in micro/nanochannels are a subject of intensive discussions in the literature [3]. Earlier studies indicated that water dissociation effects leading to the generation of Hþ and OH  ions were responsible for such overlimiting currents. Later, Rubinstein et al. [7] postulated that some mechanism of mixing should be present that destroys the DBL as lower d leads to higher currents (from Eq. 11). Maletzk et al. [12] coated the surface of cation-exchange membranes with a gel which does not allow mixing. From these experiments, they observed a saturation of current with no enhancements in the current, thereby confirming the earlier postulation of Rubinstein. Further, the fluid flow in the overlimiting regime revealed strong fluctuations indicating convection close to the surface. This convection was first attributed to the gravitational buoyant forces due to the concentration and temperature gradients. However, later theories have argued that the convection was not due to the gravitational instability in CP zones. Dukhin et al. [13] suggested the mechanism for mixing to be electroconvection. The type of electroconvection present in the overlimiting regime was revealed as the electroosmotic flow of the second kind. As discussed in the previous section, the induced space charges in the depletion zone (see Fig. 2c) under the action of the electric field results in the EOF of second kind and this convective instability tends to destroy the DBL leading to overlimiting currents as shown in Fig. 2b. Experiments by Kim et al. [10] also revealed the nonlinear currents due to the nonequilibrium EOF in the ionselective micro/nanochannels. Similar physics was also observed in the perm-selective membranes and ion-selective particles [3]. In spite of all these postulations, the physics behind the extended space charge layer still remains largely unclear and there are a lot of potential research opportunities to fully understand this complex physics. Propagation of Concentration Polarization In this section, the conditions and the scenarios which can lead to the propagation of CP in micro/nanofluidic devices are highlighted. Zangle et al. [14] highlighted 491 C the phenomena of concentration polarization propagation using a simplified model of charged species transport and validated the same by conducting experiments and by comparing with other experimental results. The CP phenomenon was found to be governed by a type of Dukhin number, relating the bulk and the surface conductance. The inverse Dukhin number, for a symmetric electrolyte was specified as: Gbulk Fhzc0 ¼ Gs s (12) where Gbulk is the bulk conductance, Gs is the surface conductance, c0 is the concentration outside the EDL, and s is the wall surface charge density. Zangle et al. postulated that CP depends on Dukhin number and not  on the ratio of channel height to the Debye length lhD . Using their simplified model, they showed that both the enhancement and the depletion regimes at the interfaces of micro/nanofluidic channels propagate as shock waves under the following condition: co;r hn < maxðu2 ; 2u2  1Þ ðu z u z ÞFh c (13) 2 2 n o;r where co;r hn ¼ 1 1 2u is an inverse Dukhin 1s number describing the ratio of bulk to surface conductance as mentioned before. u2 ¼ u2zzn2EF is the mobility of the co-ion nondimensionalized by the electroosmotic mobility. co;r is the reservoir electrolyte concentration, hn is the nanochannel height, u1 and u2 are the mobilities, and z1 and z2 are the valences of the positive and negative ionic species, respectively. zn is the nanochannel zeta potential, E is the permittivity and  is the viscosity. Elaborate details of the model can be referred in [14]. From this model, they proposed a thumb rule to avoid propagation of CP and it was found that co;r hn 1. This condition for CP propagation was compared with 56 sets of experimental literature values and was found to give a sufficient first-hand prediction with regard to the concentration polarization propagation. Though the model considers the effects of surface charge and the electrolyte concentration, the finite Pe effects which also play a critical role in the concentration polarization were not considered. Further, the experiments of Kim et al. [10] and the numerical studies performed by Jin et al. [9] also revealed that the applied potential also plays a pivotal role in the C C 492 Concentration Polarization at Micro/Nanofluidic Interfaces concentration polarization generation and propagation apart from the inverse Dukhin number. The shortcoming of the model was also highlighted by Zangle et al. in their work. Thus, still a clear and complete understanding of the CP regimes is yet to be reached and continuous efforts are being made to understand the physics at the micro/nanofluidic junctions to design advanced and novel devices. rate at a power consumption of less than 3.5Wh/L [17]. The CP depletion layer acts as a barrier for any charged species and these species were diverted away from the desalted water using suitable pressure and voltage fields. Their design also ensured salt ions and other debris to be driven away from the membrane thereby preventing any membrane fouling which is often observed in other desalination techniques. Applications In this section, various applications that have been developed utilizing the concentration polarization phenomenon are addressed. The applications range from preconcentrating biomolecules to fluid pumping and mixing and also in water desalination. A brief discussion of the aforementioned applications is presented below. Mixing, Pumping, and Other Applications Preconcentration Wang et al. [15] used the depletion region observed at the micro/nanojunction to preconcentrate proteins. The energy barrier created at the depletion region (due to the large voltage drop induced at this junction) prevents the entry of charged molecules into the nanochannel. This results in an increase in the concentration of the molecules near the depletion region. In their experiments, Wang et al. used two anodic microchannels which were independently controlled so that the direction of EOF can be aligned perpendicular to the axis of the nanopore. An increase of about 106–108 fold in the concentration of the protein was reported in their study. Over the past couple of years, similar preconcentration devices utilizing the CP effects were experimentally fabricated [3, 14]. Wang et al. [16] also presented an experimental approach to improve the binding kinetics and the immunoassay detection sensitivity using concentration polarization in micro/nanofluidic devices. The antigens were preconcentrated at the depletion region due to the strong electric field gradients resulting in the enhancement in the binding rates with the antibody beads. Seawater Desalination The phenomenon of concentration polarization witnessed in ion-selective membranes was successfully implemented to address the freshwater shortage issue by providing energy-efficient solution to water desalination. A microfluidic device was fabricated which provides 99% salt rejection at 50% recovery As discussed earlier, the induced space charges observed at the depletion region of micro/nanochannel along with the large electric field gradients at this region result in strong vortices. Kim et al. [18] enhanced the mixing efficiency of microfluidic devices using the vortices created at this interface. Further, as the electroosmotic flow of the second kind observed at the depletion region is directly proportional to the square of the applied electric field, Kim et al. [19] was able to pump fluids using the nonequilibrium EOF and observed a fivefold increase in the volumetric flow rates compared to similar devices utilizing equilibrium EOF. Yossifon et al. [20] used an asymmetric microchannel in conjunction with the nanochannels. The application of forward and reverse bias (at the overlimiting regime) voltage led to an asymmetric space charge polarization which resulted in the rectification of the current. Such membranes have potential applications in selective species separation. Summary The origin and the underlying physics that is present at the interfaces of micro/nanofluidic devices were discussed. The complex phenomenon of concentration polarization (CP) leading to the enrichment and depletion of ions near the micro–nano junctions and the concepts of induced space charges and nonlinear electrokinetic transport were briefly discussed in this chapter. Further, the various controversies surrounding the physical mechanism for nonlinear current characteristics have been highlighted. The criteria for concentration polarization propagation and the various applications that have been developed utilizing the concentration polarization phenomenon were discussed. Though, a lot of extensive work has been carried out to understand the CP physics, a clear and complete understanding of the CP regimes and the Conduction Mechanisms in Organic Semiconductors induced space charge dynamics is yet to be reached and efforts need to be directed in this area to understand the physics at the micro/nanofluidic junctions to design novel devices. Cross-References ▶ Computational Micro/Nanofluidics: Unifier of Physical and Natural Sciences and Engineering ▶ Electrokinetic Fluid Flow in Nanostructures ▶ Integration of Nanostructures within Microfluidic Devices ▶ Surface-Modified Microfluidics and Nanofluidics References 1. Karniadakis, G.E., Beskok, A., Aluru, N.R.: Microflows and Nanoflows: Fundamentals and Simulation. Springer, New York (2005) 2. Schoch, R.B., Han, J., Renaud, P.: Transport phenomena in nanofluidics. Rev. Mod. Phys. 80, 839–883 (2008) 3. Höltzel, A., Tallarek, U.: Ionic conductance of nanopores in microscale analysis systems: where microfluidics meets nanofluidics. J. Sep. Sci. 30, 1398–1419 (2007) 4. Gross, R.J., Osterle, J.F.: Membrane transport characteristics of ultrafine capillaries. J. Chem. Phys. 49, 228–234 (1968) 5. Probstein, R.F.: Physiochemical Hydrodynamics: An Introduction. Wiley, New York (1994) 6. Rubinstein, I.: Electrodiffusion of Ions. SIAM, Philadelphia (1990) 7. Rubinstein, I., Zaltzman, B.: Electro–osmotic slip of the second kind and instability in concentration polarization at electrodialysis membranes. Math. Models Meth. Appl. Sci. 11, 263–300 (2001) 8. Pu, Q., Yun, J., Temkin, H., Liu, S.: Ion–enrichment and ion–depletion effect of nanochannel structures. Nano Lett. 4, 1099–1103 (2004) 9. Jin, X., Joseph, S., Gatimu, E.N., Bohn, P.W., Aluru, N.R.: Induced electrokinetic transport in micro – nanofluidic interconnect devices. Langmuir 23, 13209–13222 (2007) 10. Kim, S.J., Wang, Y.-C., Lee, J.H., Jang, H., Han, J.: Concentration polarization and nonlinear electrokinetic flow near a nanofluidic channel. Phys. Rev. Lett. 99, 044501 (2007) 11. Rubinstein, I., Zaltzman, B.: Dynamics of extended space charge in concentration polarization. Phys. Rev. E. 81, 061502 (2010) 12. Maletzki, F., Rösler, H.-W., Staude, E.: Ion transfer across electrodialysis membranes in the overlimiting current range: stationary voltage current characteristics and current noise power spectra under different conditions of free convection. J. Membr. Sci. 71, 105–116 (1992) 13. Dukhin, S.S.: Electrokinetic phenomena of the second kind and their applications. Adv. Colloid Interface Sci. 35, 173–196 (1991) 493 C 14. Zangle, T.A., Mani, A., Santiago, J.G.: Theory and experiments of concentration polarization and ion focusing at microchannel and nanochannel interfaces. Chem. Soc. Rev. 39, 1014–1035 (2010) 15. Wang, Y.-C., Stevens, A.L., Han, J.: Million–fold preconcentration of proteins and peptides by nanofluidic filter. Anal. Chem. 77, 4293–4299 (2005) 16. Wang, Y.C., Han, J.: Pre-binding dynamic range and sensitivity enhancement for immuno-sensors using nanofluidic preconcentrator. Lab Chip 8, 392–394 (2008) 17. Kim, S.J., Ko, S.H., Kang, K.H., Han, J.: Direct seawater desalination by ion concentration polarization. Nat. Nanotechnol. 5, 297–301 (2010) 18. Kim, D., Raj, A., Zhu, L., Masel, R.I., Shannon, M.A.: Nonequilibrium electrokinetic micro/nano fluidic mixer. Lab Chip 8, 625–628 (2008) 19. Kim, S.J., Li, L.D., Han, J.: Amplified electrokinetic response by concentration polarization near nanofluidic channel. Langmuir 25, 7759–7765 (2009) 20. Yossifon, G., Chang, Y.-C., Chang, H.-C.: Rectification, gating voltage and interchannel communication of nanoslot arrays due to asymmetric entrance space charge polarization. Phys. Rev. Lett. 103, 154502 (2009) Conductance Injection ▶ Dynamic Clamp Conduction Mechanisms in Organic Semiconductors Weicong Li and Harry Kwok Department of Electrical and Computer Engineering, University of Victoria, Victoria, Canada Definition Conduction mechanisms in organic semiconductors refer to the means by which electronic charges move through organic semiconductors under external stress particularly under the influence of an electrical field. Overview In order to understand the conduction mechanisms in organic semiconductors, it is necessary to first introduce the concept of band theory, which is well established in solid-state physics. Solids in general C 494 Conduction Mechanisms in Organic Semiconductors Conduction Mechanisms in Organic Semiconductors, Fig. 1 Band structures of metal, semiconductor, and insulator Conduction band Electron energy C 1 1 þ exp½ðE  EF Þ=kT Conduction band Band gap Valence band Valence band Metal Semiconductor are made of atoms, each of which is composed of a positively charged nucleus surrounded by negatively charged electrons. In quantum mechanical terms, these electrons effectively reside in discrete energy states in orbits. When a large number of atoms (of order 1020 or more) are brought together to form a solid, the discrete energy states are so close together that energy bands begin to form. At the same time, there will be gaps between the energy bands which are known as the band gaps. Because of the presence of these energy gaps, there will be some energy bands that are almost fully occupied (known as the valence bands) and energy bands that are almost unoccupied (known as the conduction bands). Based on the band theory, solids are typically divided into the following three categories: metals, semiconductors, and insulators. In metals, there is an overlap between the energy bands so that the energy bands are partly filled by electrons at any temperature (even T ¼ 0 K), while both the semiconductors and the insulators have fully filled valence bands and empty conduction bands at T ¼ 0 K. To further study the distribution of electrons occupying the energy states in solids, it is also necessary to introduce the concept of Fermi level into the band theory, which represents the maximum energy of states that electrons can occupy at T ¼ 0 K. Accordingly, all the allowed energy states below the Fermi level are occupied by electrons, and all the energy states above it are empty. When temperature is above 0 K, the probability that electrons occupy the state with energy E under thermodynamic equilibrium condition is given by Fermi-Dirac distribution function: f ðEÞ ¼ Conduction band (1) where EF is the Fermi level, k is the Boltzmann constant, and T is the temperature in Kelvin. As mentioned Fermi level Valence band Insulator earlier, in semiconductor and insulator, the valence band is fully occupied by electrons, and the conduction band is empty at T ¼ 0 K. Therefore, one can infer that the Fermi level lies in the bandgap, between the valance and conduction bands. On the other hand, in metal, due to the fact that the energy bands are partly filled by electrons at any temperature, the Fermi level lies within the energy bands. The band structures of metal, semiconductor, and insulator, and the position of Fermi level in them are shown in Fig. 1. The distinction between the semiconductors and the insulators appears when temperature rises above 0 K. Because the band gap between conduction and valence bands in semiconductors is much narrower than that found in insulators, a fair amount of electrons can be thermally excited from the valence band to the conduction band in semiconductors at finite temperature, leading to measurable conductivity. This is not found in the insulators due to the larger band gaps even at room temperature, which lead to negligible probability of electrons occupying energy states in the conduction band, according to Eq. 1. The characteristic semiconductor band structure has allowed it to play an important role as the materials of choice in the prosperous electronic industry in the last few decades. While inorganic semiconductors such as silicon dominated the electronic industry in the twentieth century, tremendous effort has been spent in the research and development of organic electronics in last decade due to the fact that organic semiconductors are usually easier and cheaper to form. Soluble organic materials, such as conjugated polymers, can be deposited in liquid phase (e.g., by printing and spin coating) onto large substrate areas at low processing temperature (below 100 C). Due to this advantage, organic electronics are particularly attractive in the making of displays, sensors, light sources, photovoltaic panels, radiofrequency identification detectors (RFID), and in Conduction Mechanisms in Organic Semiconductors 495 devices used in optical communications. As a consequence, research on the charge conduction mechanisms in organic semiconductors and devices is of significant importance. In general, conduction mechanisms primarily describe how electronic charges (referred to as carriers) move inside the solids under the influence of an external electrical field. The process produces a current. At the macroscopic level, the current density J in solids produced by external electrical field is given by J ¼ env ¼ enmF Energy states Energy LUMO DOS(Energy) Fermi level HOMO (2) where e is elementary charge of a single carrier, n the charge density, and v is the drift velocity. Furthermore, v can be expressed as the product of the charge mobility m and the electrical field F. As can be seen in Eq. 2, a large current requires the presence of a substantial number of mobile charge carriers (electrons or holes). In organic semiconductors, mobile carriers are known to be produced from the distributed p-bonds, which are covalent chemical bonds resulting from the overlap of atomic orbitals. Thus, the limited current flow in many organic semiconductors are related to their irregular molecular structures which can result in low charge mobility in comparison to values found in silicon and other inorganic semiconductors. In addition, the more established conduction mechanisms based on band theory normally found in inorganic crystalline semiconductors are absent in the organic semiconductors. As mentioned earlier, band theory states that carriers could only exist and move in either the conduction bands or the valence bands because there are permitted energy states where carriers can reside and their movement between the energy states will produce a current. The use of “energy band” diagrams (see Fig. 1) to explain charge transport in organic semiconductors however is not possible. This is because of the presence of high densities of defects and trap states. Instead, charge transport in organic semiconductors is directly explained in terms of the energy (orbital) states which are termed either the lowest unoccupied molecular orbital (LUMO) or the highest occupied molecular orbital (HOMO). As such, LUMO and HOMO levels are not genuine energy bands and they are used merely to serve as references to demarcate ground state energy and the next activated state energy [1]. The distribution of these energy states known as the density of states (DOS) is usually considered to be C DOS(Energy) Electron Hole Conduction Mechanisms in Organic Semiconductors, Fig. 2 Density of (energy) states in an organic semiconductor Gaussian centered at the LUMO and HOMO (see Fig. 2). The width of the Gaussian energy states depends on both the regularity of the molecular structure and the impurities present in the organic semiconductor. As expected, in most organic semiconductors both the carrier density n and the charge mobility m are low and the value of the latter often depends on the strength of the electrical field F in contrast to what is observed in inorganic semiconductors. In some organic semiconductors, the molecular structures can be highly disordered and different conduction mechanisms are found to predominate depending on the associated manufacturing process. Basic Methodology Many useful techniques have been proposed to study the conduction mechanisms in organic semiconductors including time-of-flight (TOF) experiment, space charge limited current (SCLC) measurement, and field-effect measurements using organic field-effect transistors (OFETs). These techniques, combined with the dependence on temperature, provide important information on mobility, trap concentration which are useful to assess the conduction mechanisms. Brief introductions to several different techniques commonly used are given here. C C 496 Conduction Mechanisms in Organic Semiconductors – + V – + V Carrier – Laser pulse Oscilloscope I Organic semiconductor Organic semiconductor – Transparent metal contact Metal contact Conduction Mechanisms in Organic Semiconductors, Fig. 3 A schematic of the time-of-flight (TOF) experiment setup Time-of-Flight (TOF) Experiment As implied by its name, time-of-flight (TOF) experiment is the method of measuring the time it takes for one or a few carrier to travel a distance through the solid. When TOF experiment is used to study the conduction mechanisms in organic semiconductors (see Fig. 3), two metal electrodes (forming the contact) are deposited on the two ends of the organic semiconductor (one of the two is usually transparent). Initially, a few carriers are generated at one end near the transparent metal electrode using a short laser pulse with energy greater than the energy difference between HOMO and LUMO levels of organic semiconductor. The photo-generated carriers are then drifted toward the opposite end by an external electrical field generating a current pulse. By measuring the time delay of the current pulse, the velocity and the mobility of the charge carriers through the organic semiconductors can be computed. Space Charge Limited Current (SCLC) Measurement Due to the low mobility of carriers in organic semiconductors, the measured I–V characteristics usually deviate from Ohm’s law (i.e., the linear relationship between current and voltage). This is illustrated in Fig. 4. In this case, if efficient charge injection from the metal electrode is achieved by choosing a suitable metal, the I–V characteristics will follow the space charge limited current equation as given by 9 V2 I ¼ ye0 er mA 3 8 L Metal Metal (3) L Conduction Mechanisms in Organic Semiconductors, Fig. 4 A simplified schematic of the setup used for space charge limited current (SCLC) measurement where y is a parameter dependent on the traps present in the semiconductors, e0 the free space permittivity, er the relative dielectric constant, m drift mobility of injected charge carrier, A the cross section area of semiconductor, and L the distance between metal contacts. By measuring the I–V characteristics at different temperatures, one can determine the mobility and the trap density of traps. Thus, SCLC measurement is a very useful technique reflecting the conduction mechanisms in organic semiconductors. Measurement Based on the Organic Field-Effect Transistor (OFET) Field-effect transistor (FET) is an electronic device widely used in active circuits. It consists of a semiconductor with a conducting channel, an isolated gate controlling charge flow in the channel, a gate dielectric between the semiconductor and the gate, as well as source and drain regions forming the output terminals. Organic field-effect transistor (OFET) is a field-effect transistor formed on an organic semiconductor. The device configuration can have a top-gate or a bottom-gate as shown in Fig. 5. The basic operation principle of the OFETs is very simple. When a bias voltage is applied between gate and source electrodes, carriers are injected from the source into the organic semiconductor forming an extremely thin accumulation layer (2 ~ 3 nm) at the interface between organic semiconductor and the dielectric. The carriers conduct a current across the source and the drain regions, and the current depends on the gate voltage as well as the charge mobility Conduction Mechanisms in Organic Semiconductors Conduction Mechanisms in Organic Semiconductors, Fig. 5 Two device configurations for the OFETs: (a) top-gate, (b) bottom-gate a 497 Gate Gate dielectric Organic semiconductor Source Drain Substrate b C Drain Source Organic semiconductor Gate dielectric Gate C which is also dependent on the gate voltage and the drain-to-source voltage. The operation of the OFETs therefore relies on carrier accumulation in the fieldeffect structure in contrast to the case of the inorganic FETs which rely on either charge depletion or inversion. Therefore, OFETs are efficient tools to investigate the interfacial conduction mechanisms, while TOF and SCLC measurements are mainly used to study the bulk conduction mechanisms in the organic semiconductors. For example, I–V characteristics of OFETs are usually analyzed to determine parameters such as the charge mobility and the threshold voltage both of which are closed related to the density of traps at the interface. In addition, spectroscopic techniques are sometimes used to probe the morphology of the organic semiconductor interface, to look for the potential relationship between regularity of molecular structure and conduction performance. Key Research Findings Band-Like Transport For highly purified and ordered organic molecular crystals, it is possible that band-like charge transport similar to that of the inorganic semiconductors may occur. The main feature found in band-like charge transport is the fact that the temperature dependence of the charge mobility has the following form: mðTÞ / T n ; with n ¼ 1; 2; 3 . . . (4) In practice, n is usually positive which leads to increasing charge mobility when temperature decreases. In general, because the electrons are usually weakly delocalized even in the highly ordered organic crystals, the band widths of the HOMO and the LUMO are small compared to energy bands found in the inorganic semiconductors. As a result, room temperature charge mobilities observed in organic semiconductor crystals can only reach values in the range 1–20 cm2/ Vs [2]. Band-like charge transport has been observed in small-molecules and in single-crystal organic semiconductors (such as rubrene) formed by vapor deposition process. As a matter of fact, in the majority of organic semiconductors, traps and defects are formed during deposition which tends to destroy band-like properties. Polaron Transport A polaron is a quasiparticle composed of a charge carrier and its induced polarization field. In many organic materials, due to the low charge mobilities, carriers tend to polarize their surrounding lattice. As a result, polarization fields are formed around the carriers, which can no longer be considered as “naked.” Instead, the carriers will be localized in potential minima created by the so-called molecular deformations [1]. In other words, a charge is trapped by the deformation it induces. Such an entity is known as a “polaron.” Polarons can move between molecules similar to the carriers except that they also carry the deformations along. In many disordered molecular organic semiconductors, deformations associated with the trapped charges can be considerable and conduction mechanisms characterizing polaron transport are under intensive research by many research groups across the world. Variable Range Hopping (VRH) Transport For most organic semiconductors such as polymers and oligomers, their molecular structures are highly disordered and have considerable densities of defects and traps. The energy band diagrams are no longer suitable to describe the densities of states as these energy states are now localized. Furthermore, bandlike charge transport can no longer explain the observed low charge mobilities and the fact that their values increase with temperature (as opposed to what is observed in band-like charge transport). One of the widely accepted theories, known as the variable range hopping (VRH) transport, is proved to give a reasonable explanation by describing charge transport in terms of hopping of the charge carriers between localized states as shown in Fig. 6. C 498 Energy Conduction Mechanisms in Organic Semiconductors Transport band Localized states Charge carriers Localized energy states Conduction Mechanisms in Organic Semiconductors, Fig. 7 Multiple trap and release transport (MTR) model x Conduction Mechanisms in Organic Semiconductors, Fig. 6 Hopping transport in organic semiconductors Cathode Hopping can be used to explain the lower mobility found in disordered organic semiconductors and instead of the power law dependence on temperature as in band-like charge transport, the temperature dependence in VRH charge transport exhibits temperature-dependent activation as well as dependence on the applied electric field as given by [3] Electron transport layer pffiffiffi mðF; TÞ / expðDE=kTÞ  expðb F=kTÞ Electroluminescent layer Hole transport layer (5) where m is mobility, F is the electrical field, T is the temperature, DE is the activation energy, and b is a parameter related to disorder. Multiple Trap and Release (MTR) Transport Charge transport in OFETs is affected by defects and impurities which exist in the intrinsic part of the organic semiconductors per se and can also be linked to an inferior semiconductor/dielectric interface. As a result, the performance of OFETs is sample-dependent, which is one of the major difficulties in characterizing the properties of OFETs. As mentioned earlier, VRH transport is more suitable to account for charge transport in highly disordered organic semiconductors and, in contrast, another well-established and widely accepted charge transport model known as the “multiple trap and release” (MTR) model is frequently applied to the relatively well-ordered organic semiconductors, such as small molecules and molecular crystals. The basic principle of MTR model includes two important components: (1) a transport band containing delocalized energy states whereby carriers can move freely and (2) the presence of a high density of localized energy states located in the vicinity of the edge of transport band acting as traps. During charge transport, carriers move freely in the transport band Anode Conduction Mechanisms in Organic Semiconductors, Fig. 8 Schematic of a typical organic light-emitting diode (OLED) with a high probability of being trapped at the localized energy states and then subsequently thermally released into transport band again. The basic illustration of MTR transport process is shown in Fig. 7. The effective mobility (meff) in the MTR model is actually smaller than the “real” mobility (m0) in the transport band in the absence of localized energy states and is given by [4] meff ¼ m0 a expðEt =kTÞ (6) where a is the ratio of the effective density of energy states at the edge of the transport band to the density of traps in the localized energy states, and Et is the energetic distance between the edge of the transport band and the localized energy states. If the localized energy states are energetically dispersive, a and Et must be recalculated according to the trap distribution. In the study of OFETs, the MTR model is widely used to account for charge transport due to the fact that it offers a reasonable explanation on the gate voltage Conduction Mechanisms in Organic Semiconductors a 499 Electron Electron injecting electrode Organic semiconductor Light emission b injecting Hole injecting electrode electrode C Light emission Hole injecting electrode hole-type organic semiconductor Gate dielectric Gate dielectric Gate Gate electron-type organic semiconductor Conduction Mechanisms in Organic Semiconductors, Fig. 9 Schematic illustrations of organic light-emitting field-effect transistors (OLEFETs) with light emission in: (a) the single-layer configuration and (b) the multilayer configuration dependent mobility usually observed in OFETs. As mentioned above, unlike inorganic semiconductors, organic semiconductors usually have Gaussian density of states (DOS). When a bias is applied to the gate of an OFET, the Fermi level at the dielectric-semiconductor interface will be shifted toward the transport band so that a fair amount of localized energy states near the edge of the transport band will be filled when the Fermi level is located closer to the transport band. As a result, the mobility of the carriers in the MTR model is actually improved because of the reduced density of traps leading to a reduced value of Et. Therefore, the gatevoltage dependent effective mobility of the carriers in the MTR transport model is given by [5] meff Nc Ci ðVG  VT Þ ¼ m0 qNt0 Nt0 T0 T 1 (7) where m0 is the mobility of the carriers in the transport band, Nc the effective density of states at the edge of the transport band, Nt0 the total density of traps, Ci the capacitance of the insulator per unit area, and T0 is a characteristic temperature related to the distribution of the DOS. Devices Organic Light-Emitting Diode (OLED) Organic light-emitting diode (OLED) is an electroluminescent diode composed of organic materials serving as the electroluminescent layer and charge transport layer. The typical OLED structure is shown in Fig. 8. During operation of an OLED, a bias voltage is applied between anode and cathode. Holes (electrons) are injected from the anode (cathode) into the electroluminescent layer through the hole (electron) transport layer. Because electrons and holes exist simultaneously in the same layer, there is a high probability that they recombine with each other due to electrostatic forces, leading to radiative emission. Therefore, efficient charge injection from both cathode and anode is requisite for the efficient operation of the OLED and current conduction is dominated by space charge limited current as introduced earlier. Organic Light-Emitting Field-Effect Transistor (OLEFET) Organic light-emitting field-effect transistor (OLEFET) is a novel organic device combining the function of current conduction of an OFET with electroluminescence in an OLED. The operation of the OLEFET is actually the same as that of the OFET. However, if proper materials are chosen as the source and the drain to give efficient charge injection and under favorable voltage bias condition, electrons and holes can be injected and transported separately in the OFET channel(s). This type of charge transport is known as ambipolar charge transport, which is unique and only found in an organic field-effect transistor. Furthermore, in ambipolar charge transport if the electrons and holes are allowed to recombine radiatively in an emitter layer to give out light, this type of OFET with electroluminescence functionality is usually called OLEFET. Various device structures have been proposed to realize ambipolar charge transport and light emission in OLEFETs, and, in most cases, the proposed structures fall into two main categories as far as the charge layers are concerned. These are the single-layer OLEFET and multilayer OLEFET as shown in Fig. 9. C C 500 Cross-References ▶ Electrode–Organic Interface Physics ▶ Flexible Electronics ▶ Optical and Electronic Properties ▶ Surface Electronic Structure References 1. Kwok, H.L., Wu, Y.L., Sun, T.P.: Charge transport and optical effects in disordered organic semiconductors. In: Noginov, M.A., Dewar, G., McCall, M.W., Zheludev, N.I. (eds.) Tutorials in Complex Photonic Media, pp. 576–577. SPIE Press, Bellingham (2009) 2. Podzorov, V., Menard, E., Borissov, A., Kiryukhin, V., Rogers, J.A., Gershenson, M.E.: Intrinsic charge transport on the surface of organic semiconductors. Physical Review Letters 93, 086602 (2004) 3. Br€utting, W.: Physics of Organic Semiconductors. Wiley, Weinheim (2005) 4. Horowitz, G.: Organic field-effect transistors. Adv. Mat. 10, 365–377 (1998) 5. Bao, Z., Locklin, J.: Organic Field-Effect Transistors. CRC Press, Boca Raton (2007) Confocal Laser Scanning Microscopy Reinhold Wannemacher Madrid Institute for Advanced Studies, IMDEA Nanociencia, Madrid, Spain Synonyms Confocal scanning optical microscopy (CSOM); Laser scanning confocal microscopy Definition A Confocal Laser Scanning Microscope (CLSM) images a point light source used for excitation onto the sample via the objective lens and images the excited focal volume onto a point detector using reflected, transmitted, emitted, or scattered light. In contrast to conventional microscopes, this scheme permits strong rejection of out-of-focus light and optical sectioning of the sample. In order to obtain an image, the focal volume must be scanned relative to the Confocal Laser Scanning Microscopy sample. Scanning can be performed in the lateral as well as in the axial directions and three-dimensional images of the sample can be generated in this way. Operating Principle Figure 1 illustrates the working principle of a fluorescence confocal microscope. A point light source, here the end of an optical fiber carrying the excitation light, is imaged onto the sample by the objective lens via a beam splitter. Emitted light from the focal spot is imaged through the beam splitter onto a pinhole via an auxiliary lens. The light is registered by the detector, because it passes through the pinhole (left-hand side of Fig. 1). Fluorescence from a fluorophore at an out-offocus position in the excitation cone, on the other hand, arrives at the screen defocused. Therefore, only a small fraction of it passes through the pinhole and reaches the detector. The effect is understated by the schematic figure and actually much stronger in a real microscope because of the short focal length of the objective lens. The optical sectioning capability is the core of confocal microscopy and allows to render the object threedimensionally under different angles by appropriate software once a stack of images at different depths has been acquired. In addition, the lateral resolution is slightly improved in confocal microscopy, compared to conventional microscopy, when the pinhole is small. On the other hand, an image can be acquired in this way only by serial scanning of the sample or of the excitation beam (for technical improvements in this respect see section Confocal Microscopy Involving Modified Illumination). Because a CLSM operates with light, it may be used to image many different physical quantities. These may be simply reflected or transmitted intensity (brightfield confocal microscopy) or the intensity of fluorescence excited in the sample (fluorescence confocal microscopy). Other options include polarization and phase of reflected or transmitted light, as well as the intensity, wavelength, lifetime, time correlation, or recovery after photobleaching of fluorescence from the sample, or intensity, wavelength, and polarization of inelastically (Raman) scattered light. Moreover, a nonlinear response of the sample to the optical excitation near the laser focus, based, for example, on multiphoton excitation, second harmonic generation, or stimulated scattering processes may be used for Confocal Laser Scanning Microscopy 501 C Confocal Laser Scanning Microscopy, Fig. 1 Principle of confocal laser scanning microscopy, demonstrating the strong rejection of out-of-focus light C confocal microscopy. Some of these options will be discussed in section Variants of Confocal Laser Scanning Microscopy Optical sectioning and three-dimensional image acquisition being the essential feature of confocal optical microscopy, it is worth mentioning here that an alternative (diffraction-limited) brightfield optical microscopy technique with similar capability is digital holographic microscopy, although this technique is far less widely known and used. Here, the object is reconstructed from an intensity camera image and no scanning is necessary. Moreover, a phase image is obtained in addition to an amplitude image. Significant improvements in object reconstruction algorithms have been made in recent years and lens-based versions with external reference beam as well as lensless versions have been demonstrated. Lens-based instruments with external reference beam are commercially available. Basic Theory of the Confocal Microscope The spatial resolution of modern high-quality conventional, as well as confocal microscopes is limited by diffraction. This means that within the design spectral range of the objective lens aberrations, such as spherical aberration, astigmatism, coma, field curvature, distortion, and chromatic aberration have a significantly smaller impact on the resolution of the microscope than diffraction. An important exception to this statement arises from aberrations due to refraction, when sample regions well inside refracting samples have to be imaged. Most modern microscope objectives are now infinity-corrected, that means they are corrected for forming an image at infinity. A tube lens is in this case required to form a real image at finite distance. The limitations by diffraction are, however, in any case dominated by the objective lens and not by the tube lens. This is due to the dependence of the diffraction limit on the opening angle of the rays contributing to the image, which is much smaller for the tube lens than for the objective lens. Point Spread Function of the Confocal Microscope The three-dimensional intensity distribution in the image space corresponding to a single-point object, demagnified by the magnification of the optical system, is called the (intensity) point spread function (PSF) of the lens. The PSF of a confocal microscope with an infinitesimally small pinhole is given by [1, 2]: PSCF ðx; y; zÞ ¼ PSFill ðx; y; zÞ PSFdet ðx; y; zÞ (1) Here, PSCF, PSFill, and PSFdet represent the point spread functions for the confocal imaging and the illumination and detection paths respectively. Neglecting the contribution from the tube lens, as well as aberrations of the objective lens, the latter two functions are simply the point spread functions of a simple lens, which, in the paraxial and scalar approximation, can be calculated by means of the Huygens-Fresnel principle as [3] PSFðx; y; zÞ ¼ jhðx; y; zÞj2 (2) with * hðrÞ ¼ C0 l Z Z A * eiks C dA  l s Z Z * eik~q r dO (3) O Here, hðr Þ represents the scalar complex amplitude * of the field in the image space at a position r ¼ ðx; y; zÞ C 502 Confocal Laser Scanning Microscopy Confocal Laser Scanning Microscopy, Fig. 2 Intensity point spread function of a conventional (a) and a confocal (b) microscope in the scalar and paraxial approximations. The numerical aperture of the objective lens is N.A.¼0.5. z is the coordinate along the axis of the lens and r the lateral dimension, both measured in wavelengths 1 1 0.5 0.5 0 −20 0 −20 1 −15 0.8 0.7 −15 −10 0.6 −10 0.5 −5 −5 0.4 z 0 z 0 0.3 5 5 0.2 10 10 0.1 15 15 20 −5 relative to the location of the geometric focus, l is the wavelength, k ¼ 2p=l, s the distance between the * point P at position r in the image space and a point * Q at position f q in the pupil A of the lens of focal * length f, q is a unit vector in the direction of Q, O is the solid angle subtended by the aperture of the lens as seen from the origin at the geometric focus, and C and C0 are constants. Equation 3 assumes homogeneous illumination of the lens. Inhomogeneous illumination can be taken into account by multiplying the integrand with a corresponding pupil function (compare section Apodization). For a confocal microscope operating in reflection or transmission modes PSFill ðx; y; zÞ  PSFdet ðx; y; zÞ and both functions are then identical to the one given in Eq. 2. This results in PSCF ðx; y; zÞ ¼ jhðx; y; zÞj4 0.9 (4) Experimentally, this function would be observed when a point object is scanned through the focus of the instrument in both cases. It should be kept in mind here that the paraxial approximation is contrary to the actual typical situation in optical microscopy. The paraxial approximation, nevertheless, works surprisingly well even for N.A.  0.5 (y0 ¼ 30 in the case of a dry lens, see below) and gives reasonable estimates of the diffraction limit even for higher numerical aperture objective lenses. Deficiencies of the scalar approximation will be discussed in section 0 r 5 20 −5 0 r 5 Deficiencies of the Scalar and Paraxial Approximation. Effects of the Vector Character of Light Figure 2 displays the three-dimensional intensity point spread function of a conventional (a) and a confocal (b) microscope calculated according to Eqs. 1 and 3. The numerical aperture of the objective lens is N.A. ¼ 0.5. z is the coordinate along the axis of the lens and r the coordinate perpendicular to the optical axis, measured in wavelengths. The drastic reduction in side lobes for a confocal microscope is immediately evident. This also leads to drastic reduction in laser speckle in the case of coherent illumination. Single-Point Resolution in the Focal Plane In the focal plane (z ¼ 0) the PSF of the conventional microscope, as calculated from Eq. 3, coincides with the well-known Airy pattern   2J1 ðvÞ 2 PSF ¼ v (5) v ¼ kr n sin y0 ¼ kr N:A: (5a) with Here, r is the distance from the axis, y0 is the angle of a marginal ray passing through the aperture toward the geometric focus, relative to the optical axis, and n is the refractive index in the image space. From this Confocal Laser Scanning Microscopy 503 expression, the lateral full width at half maximum (FWHM) of the PSF of the conventional microscope FWHM ¼ 0:51l N:A: (5b) is easily calculated. Equation 4 yields PSFCF ¼   2J1 ðvÞ 4 v (5c) for the PSF of the confocal microscope in reflection mode and, therefore FWHMCF ¼ 0:37l N:A: (5d) Equations 5b and 5d demonstrate the enhancement in lateral (single-point) resolution for a confocal microscope relative to a conventional one, in the case when the pinhole is closed completely. As an example, for l ¼ 488 nm, N.A. ¼ 0.9, FWHM ¼ 277 nm and FWHMCF ¼ 201 nm. Single-Point Resolution on the Axis: Depth Response Similarly, the PSF of the conventional microscope on the optical axis is calculated from Eq. 3 as   sinðu=4Þ 2 PSFðuÞ ¼ u=4 (6) u ¼ nkz sin2 y0 (6a) with It turns out that the definition u ¼ 4kz sin2 ðy0 =2Þ ¼ 2nkzð1  cos y0 Þ 0:89l nð1  cos y0 Þ PSFCF ðuÞ ¼  4 (6d) FWHMCF ¼ 0:64l nð1  cos y0 Þ (6e) sinðu=4Þ u=4 and For the parameters used in the example above, l ¼ 488 nm, N.A. ¼ 0.9, FWHM ¼ 770 nm and FWHMCF ¼ 554 nm. The FWHM of the point spread functions on the optical axis of the microscope is therefore about three times larger than the lateral FWHM for the conventional as well the confocal microscope, although the precise value depends on the numerical aperture and although the scalar theory used here is not applicable at large numerical aperture (see section Deficiencies of the Scalar and Paraxial Approximation. Effects of the Vector Character of Light. V(z) A clearer demonstration of the different optical sectioning capabilities of the conventional and confocal microscopes is obtained with planar objects instead of point objects. The depth response of a confocal microscope operating in reflection is often characterized by axially scanning a mirror through the focus position and registering the light intensity behind the pinhole during the scan. The corresponding amplitude function is called V(z), an expression coined originally for the acoustic microscope, which is also a confocal instrument and to which the same scalar theory is applicable. Because the image of the illuminating infinitesimal pinhole is moving by a distance of 2z when the mirror moves by a distance z, the intensity detected behind the pinhole is derived from Eqs. 6, b, by replacing u by 2u: (6b) which is equivalent to Eq. 6a in the paraxial approximation, is more appropriate at higher numerical apertures. The corresponding FWHM of the depth response is therefore FWHM ¼ C (6c) for the conventional microscope. Correspondingly, for the confocal microscope   sinðu=2Þ2   IðzÞ ¼ jVðzÞj ¼  u=2    sinðnkzð1  cos y0 ÞÞ2  ¼  nkzð1  cos y Þ  2 (7) 0 This equation predicts a central maximum of width FWHMCF ¼ 0:44l nð1  cos y0 Þ (8) C C 504 Confocal Laser Scanning Microscopy which, in paraxial approximation y0 << 1 for a dry lens becomes FWHM CF 0:89l  N:A:2 (9) Equation 7 also implies symmetric side lobes for negative and positive defocus. As an example, for a dry lens of N.A. ¼ 0.8 and an operating wavelength of 488 nm Eq. 8 yields FWHM ¼ 537 nm. In a conventional microscope, on the other hand, the signal received by a large area detector would be independent of the position of the mirror. The simple theory presented so far predicts a symmetric V(z) function. Whereas the width of the main peak of the V(z) is typically very close to measurements performed with real lenses, the aberrations present in any real objective lens typically lead to deviations as far as the side maxima are concerned and in particular to asymmetry in V(z) for positive and negative defocus. This may be used for quantitative characterization of, for example, the amount of spherical aberration present in the optical system. Interferometric versions of confocal microscopy, however, have been more traditionally used for this purpose. for the conventional microscope and dR;CF ¼ 0:56l N:A: (11) for the confocal microscope, just 8% less than for the conventional microscope. In the case of coherent sources, dR depends on the phase difference of the sources: whereas two out-of-phase coherent sources can be clearly resolved, because there will be a zero of intensity halfway between the images of the two sources, the sources cannot be resolved, if they are in phase with each other, because the maximum will lie in the middle between the images of the two individual sources. There is another criterion for the two-point resolution, which is more generally applicable, because it does not refer to a zero of the response function. The Sparrow criterion states that two sources are considered to be resolved, when the intensity halfway between the two images is the same as the one at the individual image locations. For incoherent illumination, this results in dS ¼ 0:51l N:A: (12) for both the conventional and confocal microscopes. Two-Point Resolution: Rayleigh and Sparrow Criteria The single-point resolution of the confocal microscope, as given by the PSF discussed above, is in most cases not the relevant quantity to judge the resolution of the instrument, because what is really desired is the capability to resolve certain details of a microscopic object consisting of various parts. It is therefore important to quantify the two-point resolution of the instrument, which means the capability to resolve two point objects close to each other. There is some arbitrariness in this definition, because it depends on the subjective judgment, under which conditions two point objects are resolved in an image. Only the lateral two-point resolution will be discussed here. The Rayleigh criterion defines two-point sources as resolved, if the image of the second point lies at the first zero of the image of the first one or at a larger distance. This leads to a lateral resolution dR ¼ 0:61l N:A: (10) Coherence in Brightfield and Fluorescence Microscopy The imaging of extended objects differs significantly for the conventional and confocal microscopes, respectively. For a CLSM operating in brightfield (reflection or transmission) mode the imaging is spatially coherent, because the illumination generates a spatially coherent field distribution in the focus of the objective lens, as given by the complex amplitude point spread function of the objective lens. For a CLSM in reflection mode this field distribution has to be multiplied by the reflectance R of the sample and this weighted field distribution is then imaged onto the pinhole implying convolution with the combined amplitude point spread function of the objective and pinhole relay lenses. Assuming imaging of the object onto the pinhole by the same objective lens that is used for excitation and neglecting contributions to the point spread function from the pinhole relay lens in the second imaging step the intensity behind the infinitesimally small pinhole can therefore be written as the convolution Confocal Laser Scanning Microscopy 505 2 Z Z Z   I ¼  hðx; y; zÞRðx; y; zÞhðx; y; zÞdxdydz (13) Here, h(x,y,z) is the amplitude PSF of the objective lens (compare Eq. 3). For the case of an even PSF Eq. 13 is equivalent to  2 I ¼  h 2  R (14) that means the signal is given by the absolute square of the convolution of the amplitude PSF of the confocal * microscope, h2 ðr Þ, with the local amplitude reflectivity of the sample. In the case of a conventional microscope, on the other hand, the illumination is approximately incoherent and therefore I ¼ jhj2 j Rj2 (15) In reality, for the conventional microscope, imaging is partially coherent, because the emission from each emitting point on the illumination source is imaged, due to diffraction at the condenser aperture, into a finite spatial region, which is occupied by a coherent field due to that point emitter, and the regions in the image space corresponding to neighboring, incoherently emitting points on the source partially overlap on the sample. This is true for critical illumination, where the spatial region would be given by the PSF of the condenser, as well as for Köhler illumination, where the spatial region is the whole illuminated region of the sample. Fluorescence Confocal Microscopy Fluorescence imaging is incoherent. Assuming that the fluorescence intensity is proportional to the excitation intensity the signal obtained in a confocal microscope with an infinitesimally small pinhole is 2 2 I ¼ ðjhðlÞj jhðblÞj Þ  f (16) where it has again been assumed that, as in standard commercial confocal microscopes, the same objective lens is used for excitation and fluorescence imaging, respectively. Here, f represents the distribution of fluorescent centers in the sample, l the excitation wavelength, bl the fluorescence wavelength, and b the ratio of both wavelengths, the Stokes ratio. C In the case of several different types of emitters, f would have to be weighted according to the spectral contribution of each emitter to the detected signal, which depends on the filters employed in detection for rejection of the excitation and also on the wavelength-dependent sensitivity of the detector. For a point emitter placed at the focus, the convolution with a d function just yields the first two terms on the right-hand side of Eq. 16. For a conventional microscope, on the other hand, because the whole sample is illuminated, the single-point resolution is only determined by the intensity PSF of the objective lens at the fluorescence wavelength, and the excitation wavelength is irrelevant. Effects of Finite Pinhole Size A finite size of the pinhole is obviously required in order to obtain a measurable signal. This will reduce the lateral resolution as well as the optical sectioning capability. It can be shown [2] that for a single-point object the lateral resolution is almost unaffected by the size of the pinhole, if vP ¼ 2p r p sin a  0:5 l M (17) where rP is the radius of the pinhole and M the magnification of the lens. The maximum value allowed by Eq. 17 therefore sets a reasonable value for the pinhole size, if optimum lateral resolution and reasonable signal are desired. As an example, in the case of a 100x/0.8N.A. objective lens and l = 514 nm a critical diameter of the pinhole of 10.2 mm is calculated from Eq. 17. The depth discrimination, as measured by moving a mirror through focus, on the other hand, is less affected and is essentially unaltered if vp  2:5 (18) It is obvious from these equations that the pinhole size must be adapted when the objective lens is changed. Apodization Equation 3 assumed rectangular apodization, that is, homogeneous illumination of the lens pupil and neglects reflection losses at the lens. For a given objective lens rectangular apodization yields the smallest FWHM of the focus in the focal plane, at the expense C C 506 of larger side maxima, compared to pupil functions falling off toward the edge of the objective lens. It can be achieved only approximately with Gaussian laser beams and is then equivalent to loss of a large fraction of the power of the excitation beam. Gaussian apodization, on the other hand, increases the FWHM, but reduces the side maxima. Deconvolution As described by Eq. 16 the image acquisition process in confocal fluorescence microscopy can be modeled as a convolution of the spatially dependent fluorescence of the sample with a point spread function (PSF) of the imaging system. This is true also for conventional nonconfocal fluorescence microscopy. In addition, random noise is superimposed on the image. Deconvolution with the PSF would naturally seem the appropriate way to determine the true fluorescence distribution in both cases. Applied to confocal images the resolution may be improved. In the case of conventional microscopy optical sectioning and removal of out-of-focus blur may be achieved by post-processing instead of employing hardware in the optical setup. In general, the 3-D PSF, necessary for this procedure, can be obtained experimentally or analytically. In the experimental methods, images of one or more point-like objects are collected. The problem with this technique lies in the poor signal-to-noise ratio obtainable with very small objects and the fact that the PSF may vary depending on the sample. In analytical calculations of the PSF aberrations of the optical system are often partially taken into account, whereas, on the other hand, the scalar approximation is most often used and the effects of the vector character of light (compare section Deficiencies of the Scalar and Paraxial Approximation: Effects of the Vector Character of Light) are neglected. Many 3D deconvolution methods are currently employed and some are available in commercial and non-commercial software packages [4]. The simplest class are neighboring methods, in which out-of-focus blur is removed by subtraction of neighboring (filtered) images within a stack. This method does not sufficiently remove noise. In contrast to that linear methods apply deconvolution to the whole stack of images at once. Examples are inverse filtering, Wiener filtering, the linear least squares, and the Tikhonov filtering techniques. The last three methods do not restore high frequency object components beyond the Confocal Laser Scanning Microscopy bandwidth of the PSF and inverse filtering suffers from noise amplification. All methods are very sensitive to error in the PSF. Therefore constrained iterative nonlinear algorithms are often employed, in which, starting from a guess for the true object, an error is minimized under certain constraints (positiveness of the image, finite support of the sample, etc.). In cases of strong noise in the image statistical iterative methods (like the maximum likelihood method) are favored. A computational alternative are blind deconvolution methods, in which the PSF of the optical system and the “true object” are simultaneously determined in a converging iteration. These methods are, however, computationally demanding, sensitive to noise, and solutions may be non-unique. Deficiencies of the Scalar and Paraxial Approximation: Effects of the Vector Character of Light In view of many more recent developments in confocal microscopy, it appears useful to shortly discuss deviations from the simple scalar theory. These deviations become increasingly important with increasing numerical aperture of the objective lens. In the case of linear polarization of the excitation beam, cylindrical symmetry is lost. The field in the focal plane and exactly on axis is then polarized in the direction of the excitation, but, away from the axis, it is elliptically polarized with a longitudinal component pointing in the direction of the optical axis. The intensity PSF becomes elongated and approximately elliptical, with the major axis of the ellipse in the direction of the excitation. Both effects increase with increasing numerical aperture. In the case of linearly polarized excitation, the two-point resolution of an ordinary optical microscope equipped with a well-corrected high numerical aperture objective lens (as well as that of a similar confocal microscope) therefore depends on the orientation of the line connecting the two points relative to the incoming polarization (as well as on the orientation of the dipolar point reflectors or absorbers/emitters). Figure 3 shows the PSF for numerical aperture N.A. ¼ 0.95 and the absolute squares of all electric field components in the focal plane. In the scalar approximation, the lines of constant intensity would, of course, be circles, which is clearly not the case in the figure appearing in the lower right corner of Fig. 3. Moreover, the maxima of the longitudinal field occur on both wings of the main maximum, along the direction of the incoming Confocal Laser Scanning Microscopy 2 C 2 |Ey|2 |Ez|2 0 y y 1 0 0.9 0.8 0.7 −2 −2 0 x 2 −2 −2 2 0 x 2 0.6 0.5 2 |Ex|2 |Etot|2 0.4 0.3 0 y y Confocal Laser Scanning Microscopy, Fig. 3 PSF of a well-corrected microscope objective of a high-numerical aperture lens (N.A. ¼ 0.95) for the case of linear polarization of the incoming beam, calculated using the DebyeWolf integral. A constant pupil function has been assumed here and, correspondingly, reflection losses in the lens have been neglected. z is the coordinate along the axis of the lens and x, y are the lateral coordinates, all measured in wavelengths. The incoming polarization is along the x axis 507 0.2 0 0.1 0 −2 −2 polarization (x direction) and the maximum absolute square of Ez is approximately 20% of that of Ex. The vector diffraction problem was first solved by Richards and Wolf [5]. A somewhat more physical treatment employs expansion of the field in the image space into vector multipoles centered at the focus [6]. This latter approach is also of interest for matching the focal field distribution to the fields of a dipole via the amplitude distribution and the polarization in the pupil plane. In this way, the coupling of the field to single atoms or molecules can be significantly enhanced, which is of interest, for example, for quantum optical applications. The vectorial approach is in general also required to treat focusing of other distributions of intensity, phase, and polarization in the pupil plane of the lens. Examples relevant for applications include radially polarized excitation, producing a longitudinally polarized focus, azimuthally polarized excitation, which leads to a doughnutshaped intensity distribution in the focal plane, or combinations of these distributions with scalar vortices, that means helical phase fronts. A longitudinal focus is essential for tip-enhanced Raman microscopy (TERS, compare section Confocal Raman Microscopy, ▶ Scanning Near-Field Optical Microscopy), 0 x 2 −2 −2 0 x 2 which combines a near field technique with confocal imaging. Other distributions are relevant for optical tweezers. Instrumental Details Scanning Techniques Whereas scanning the sample is an option, particularly in laboratory setups, many commercial confocal microscopes employ lateral scanning of the laser beam and sample scanning in the vertical direction. Beam scanning is typically achieved using mirrors mounted on galvanometer motors, which allow to vary the angle at which the beam passes the rear focal plane of the objective lens. An example for a telecentric 4f system that allows to vary this angle without displacing the beam in the rear focal plane is shown in Fig. 4. Recently, resonant galvanometer-based beam scanning systems have become commercially available, which employ torsion-spring based sinusoidal oscillations of the scan mirror with frequencies in the kilohertz range with open-loop operation for the fast scan axis. This permits frame rates on the order of C C 508 Confocal Laser Scanning Microscopy G f L1 f f L2 f RFP OL Confocal Laser Scanning Microscopy, Fig. 4 Telecentric lens system minimizing beam walk off. f: focal length of lenses L1 and L2, G: scan mirror mounted on galvanometer scanner, RFP: rear focal plane of objective lens OL 30 frames per second and in this way allows to study fast processes, such as diffusion in biological cells or to avoid blurring of the image due to movement of organs in in vivo studies. Alternative fast scanning confocal microscopes are discussed in section Variants of Confocal Laser Scanning Microscopy. Excitation Sources and Beam Delivery Ion lasers, HeNe lasers, diode-pumped solid state lasers, and diode lasers have all been used as continuous wave light sources in confocal laser scanning microscopy. Argon ion lasers provide a choice of several excitation wavelengths and are therefore still popular in spite of their low efficiency. In order to combine several laser beams dichroic beam splitters are employed. In many cases a software-controlled acousto-optic tunable filter (AOTF) selects the excitation wavelength(s) of choice from this combined beam. The AOTF is based on the diffraction of light from ultrasonic waves generated in a birefringent crystal by an ultrasonic transducer. Incident and diffracted waves propagate as ordinary and extraordinary wave in the crystal, respectively, or vice versa, and are therefore polarized perpendicular to each other. This allows convenient rejection of the incident light by a polarizer. Because of momentum conservation the difference in the optical wave vectors of both beams must be equal to the acoustic wave vector. This means that the wavelength of the diffracted beam is controlled by the ultrasonic frequency. In a collinear AOTF both optical beams and the acoustic wave propagate in the same direction, independent of the optical wavelength. The optical output of the AOTF is typically fed into an optical fiber, which delivers the beam to the input optics of the confocal scan head of the microscope. In brightfield reflection confocal microscopy, the beam splitter which directs the excitation light toward the objective lens (compare Fig. 1) induces a considerable loss for the excitation as well as for the detected light. This loss is minimized by a 50/50 beam splitter. In fiber-based confocal microscopes, instead of beam splitters 2  2 fiber couplers are typically employed. An improved version would make use of optical circulators, but these are presently not widely available for wavelengths in the visible range. In fluorescence confocal microscopy, the detected wavelength differs from the excitation wavelength and therefore dichroic dielectric beam splitters are used which are highly reflecting (transmitting) at the excitation (detection) wavelength and minimize losses in this way. Detection Standard detectors in confocal laser scanning microscopes are photomultipliers, which are in many cases operated in the analogue mode, which means by measuring the anode current and integrating over the pixel dwell time. Photon counting, on the other hand, is typically employed in fluorescence correlation and fluorescence lifetime microscopy (compare section Variants of Confocal Laser Scanning Microscopy), where usually avalanche photodiodes with high quantum efficiency and fast response times replace photomultipliers as detectors. Confocal microscopes which allow to spectrally disperse the light passing the confocal aperture often employ a charge-coupled camera (CCD) attached to a spectrograph to register the spectra. Back-illuminated Peltier or liquid nitrogen cooled CCD’s provide high quantum efficiency (above 90% over a wide spectral range) and low background noise, which is important for single molecule detection or when working with less photostable fluorescent probes. Confocal Laser Scanning Microscopy 509 C species with correspondingly longer excitation wavelengths are of particular interest because of reduced autofluorescence, deeper penetration, and better resistance toward high excitation density of biological tissue in this spectral region. Research is ongoing to improve brightness and photostability, reduce oligomerization and pH sensitivity, improve the appropriateness for fusion tagging, and reduce the time required for maturation of the protein in living organisms. Variants of Confocal Laser Scanning Microscopy Confocal Laser Scanning Microscopy, Fig. 5 Tertiary structure of the Green Fluorescent Protein [8]. The fluorescent chromophore, composed of three amino acids is located in the center of the beta barrel protein cage, length about 4 nm, which prevents quenching of the fluorescence by water Fluorescent Probes Many samples are autofluorescent and therefore allow fluorescent imaging without having to introduce additional fluorescent probes. In biological samples, however, autofluorescence is typically weak and therefore the sample frequently had to be stained with appropriate dyes. The latter, however, are often highly phototoxic in living cells. An important development in light microscopy, including confocal laser scanning microscopy, started in the year 1994 when the green fluorescent protein (GFP, see Fig. 5) from the jellyfish Aequorea victoria was genetically expressed in bacteria making them fluorescent at room temperature. In the same way, it is now generally possible to label proteins of interest in biological cells with fluorescent proteins by genetic manipulation, which can be achieved, for example, by injection of a virus vector. The number of fluorescent proteins used in the field has exploded by now and they are widely used in optical microscopy because of their considerably reduced phototoxicity, brightness, and photostability [7]. Genetically modified fluorescent proteins from Aequorea victoria now cover the spectral range from the deep blue to yellow and others derived from Anthozoa species (corals and anemones), as well as other sources, span the entire visible spectrum. The tertiary structure and size of these fluorescent proteins is very similar to those derived from Aequorea victoria, although the amino acid sequences are quite different. Red emitting Confocal Microscopy Involving Modified Illumination Slit-Scanning Confocal Microscopes Scanning a line focus, generated, for example, by a cylindrical lens, over the sample and imaging this line focus onto a slit aperture parallel to the image of the line focus still provides the optical sectioning capability of the confocal microscope, because the slit rejects out-of-focus light. At the same time, the frame rate is significantly increased, because scanning is necessary only in one direction. A spatially sensitive detector must be used to resolve light passing through different positions along the exit slit. This may be achieved by imaging the exit slit onto a onedimensional detector array, read out synchronously with the scan, or by scanning an image of the exit slit, synchronously with the scan, across a twodimensional detector, such as a CCD camera, forming a confocal image in this way. Disadvantages over the single-point scanning technique include reduced lateral resolution in the direction of the line focus, enhanced out-of-focus background, and, in the case of coherent illumination, increased laser speckle. Spinning Disk Confocal Microscopes Instead of scanning a single-point focus across the sample, multiple focal spots may be simultaneously generated and imaged each onto a confocal aperture. A white light version of such a confocal microscope based on a spinning Nipkow disk was introduced by Petran and Hadravsky already in the 1960s and later improved by Xiao, Corle, and Kino. The disk contains pinholes arranged in a spiral pattern, which are slightly displaced such that the whole sample is illuminated after the disk has rotated by a certain angle. In more C C 510 recent versions, light from each focal spot on the sample passes the same pinhole in the disk that was used for excitation. A two-dimensional detector, such as the eye of the observer or a CCD camera, is used to register a confocal image while the disk is spinning. Nipkow disk based confocal microscopes are now commercially available from several manufacturers and provide fast confocal imaging, but at the cost of reduced flexibility, because the pinhole size cannot be varied and because the beams cannot be steered at will, as it is necessary, for example in some experiments involving photobleaching. Moreover, cross-talk between the different focal spots may occur. Another version of confocal microscopy with multiple focal spots, swept field confocal microscopy, leaves the pinhole array stationary and sweeps the image of this array over the sample. By switching between different pinhole arrays, the pinhole size can be varied. The light efficiency of Nipkow disk based confocal microscopes may be significantly improved by adding a microlens array, mounted on a second disk, which is spinning on the same axis as the Nipkow disk and placed on top of the latter. Each microlens focuses incoming light onto one of the pinholes of the Nipkow disk. A dichroic beam splitter between both disks may be used to direct the detected fluorescence onto a camera. Another option is to use slit-shaped apertures on the Nipkow disk, which results in the same advantages and disadvantages as already described in section Slit-Scanning Confocal Microscopes. For a review of applications of spinning disk microscopes in life science, see reference [9]. Chromatic Brightfield Confocal Microscopy A chromatic confocal microscope operating in reflection deliberately introduces chromatic aberrations into the imaging system. Scanning in the vertical direction is then replaced by simultaneous detection of different spectral components which encode the depth information, because the depth of the focus depends on the wavelength. A complete stack of images can be acquired in this way in a single two-dimensional mechanical scan of the sample or of the excitation beam. Broadband excitation may be provided by a white light lamp or by a femtosecond laser generated supercontinuum. Structured Illumination Microscopy (SIM) The SIM technique does not employ any pinhole and can be used with white light, but is related to Confocal Laser Scanning Microscopy confocal microscopy in its optical sectioning capability [10]. Optical sectioning is achieved by acquiring a sequence of images of the sample with structured illumination. The simplest case of structured illumination is thereby produced by placing a grid of fully transparent and fully opaque stripes of equal width (one half the period L of the grid) into the illumination path and projecting this grid onto the sample. Only sample structures that are in focus will lead to significant variations of the image when the grid is displaced along the direction of periodicity, because only in focus the image of the grid within the sample is sharp. After acquiring three images with the grid displaced by 0, L/3, and 2L/3 the optical section can be calculated (in the simplest version of the SIM algorithm) as the root mean square of the three differences between the three images. More sophisticated deconvolution algorithms are available and many other structures for illumination can be used. Movement of a grid illumination pattern across the sample may be replaced by the generation of arbitrary patterns by digital mirror devices (DMD) based on microelectromechanical systems (MEMS), or on spatial light intensity modulators (SLM), based on liquid crystals. Whereas optical sectioning can be achieved more easily in this way than in a standard confocal system with a single-point focus, there are also some problems related to this approach. SIM works badly in strongly scattering samples, because small differences on a large background have to be determined. This is related to a significant loss in bit resolution and, hence, dynamic range in the final image. Moreover, the optical sectioning capability of SIM is slightly worse than for the standard single focus confocal microscope. Another version of SIM employs either a grid pattern or random aperture arrays on a spinning disk, which both allow a large throughput of the light from the illumination source of the order of 50%. Because of cross-talk between the light transmitted through neighboring apertures, the image acquired through the disk will contain a part that is not in focus. This part has to be subtracted from the image. A corresponding conventional image, to be subtracted from the partly confocal image, may be acquired by tilting the disk slightly, reflecting a second light source from the rear side of the disk toward the microscope objective and registering the corresponding image using a second Confocal Laser Scanning Microscopy camera. This procedure allows rapid optical sectioning and, hence, in vivo imaging of biological samples with very good signal-to-noise ratio with a comparatively simple instrument. A similar approach of subtracting the conventional image is based on a DMD, instead of a spinning disk and was termed programmable array microscope. Versions of structured illumination microscopy providing a moderate degree of super-resolution are based on Moiré patterns produced by projecting a high spatial frequency grid onto the sample (high resolution SIM, HR-SIM). The Moiré pattern arises, because the observed signal is the product of the spatial distribution of the excitation with the concentration of the fluorophore and therefore contains spatial frequencies equal to differences between sample spatial frequencies and the one of the grid. The grid pattern must not only be shifted, but also be rotated, in order to be able to calculate the image. The method is able to increase the resolution by a factor of two beyond the diffraction limit. Confocal Microscopy Beyond Brightfield and Standard Fluorescence Confocal Raman Microscopy In the Raman spectroscopy mode the inelastically scattered light from the sample is detected, where the frequency shift toward lower (Stokes signal) or higher photon energy (anti-Stokes signal) coincides with an internal vibration of the sample. Raman scattering is typically very weak and strong rejection of elastically scattered laser light is necessary. Historically, triple monochromators were often used for this purpose and still yield the highest spectral resolution, but in recent years dielectric long-pass filters with ultra-sharp transmission edges and high rejection factors have become available. This simplifies the instruments significantly, increases light efficiency, and in combination with slitscanning techniques (see section Slit-Scanning Confocal Microscopes) allows rapid multispectral confocal Raman imaging with acquisition times per pixel in the millisecond range. Coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS) microscopies are nonlinear variants of Raman microscopy based on stimulated Raman scattering. Both techniques require two short pulse lasers operating at different frequencies n1 and n2. The overlapping beams are focused by the microscope objective into the sample. When the difference in the optical frequencies is tuned to the 511 C frequency of a characteristic vibrational frequency nv of the sample (n1  n2 ¼ nv), it will excite this vibration and at the same time generate anti-Stokes Raman scattered light at a frequency 2n1  n2 ¼ n1 + nv (CARS) or weakly deplete the pump and enhance the Stokes beams (SRS). Stimulated Raman scattering can be several orders of magnitude stronger than spontaneous Raman scattering and therefore allows rapid label-free imaging of a particular molecule in the sample. At least one of the two lasers necessary for CARS or SRS has to be tunable. This requirement maybe fulfilled, for example, by a Ti:sapphire laser or an optical parametric oscillator. Video rate in vivo SRS microscopy has been recently demonstrated and offers a number of advantages over CARS microscopy. Another variant of confocal Raman microscopy, tip-enhanced Raman scattering (TERS) employs optical near fields of a sharp tip in order to increase the spatial resolution in Raman scattering over the diffraction limit. Confocal excitation and detection thereby reduces elastically scattered background in this setup. Multiphoton Microscopy Fluorescence excitation may in general be based on a linear process, in which a single photon excites the emitter into the excited state at energy E1 from which it fluoresces, or on nonlinear processes, in which the emitter simultaneous absorbs n photons of energy E1/n. Nonlinear processes are usually much less likely to occur than linear ones, with the probability strongly decreasing with the number of photons. Therefore, pulsed lasers are used for excitation, in which the optical power is concentrated in short pulses of widths in the femtosecond to picosecond range and the intensity during the pulse is very high. In addition, focusing the beam to a submicron spot leads, of course, to strong additional enhancement of the excitation probability. Multiphoton microscopy has several important advantages over optical microscopy employing linear excitation. First, it is inherently confocal, in the sense that out-of-focus contributions to the detected signal are very small and therefore a confocal pinhole is not required (or the existing pinhole can be opened fully without loosing the optical sectioning capability). This is advantageous, in particular for strongly scattering samples, because it yields a higher detection efficiency. Second, the photon energy used for excitation is only one half (for two-photon excitation) or one third (for three-photon excitation) of that used in the linear C C 512 case and the excitation wavelength therefore typically lies in the near infrared or even further in the infrared. Because the scattering in biological tissue and other inhomogeneous materials with inhomogeneities on a scale of the wavelength or below decreases strongly with the wavelength (proportional to l–4 for very small inhomogeneities) the penetration depth is considerably larger for multiphoton excitation compared to single photon excitation of the same fluorophore. This means that three-dimensional imaging deep into tissue becomes possible. Third, photo-induced damage to the sample and the fluorophore is reduced, also because of the lower photon energy. Instead of making use of multiple photon excitation of a fluorescent chromophore, it is also possible in some samples to detect light due to second harmonic generation (SHG) and to use that for label-free imaging. Other label-free and, in addition chemically specific, variants of multiphoton microscopy are CARS microscopy and SRS microscopy, as described in section Confocal Raman Microscopy. Confocal Laser Scanning Microscopy pulse of the exciting laser measures arrival times of fluorescence photons, typically detected by an avalanche photodiode with a short response time, and generates a histogram of delays. Fitting an exponential function to the histogram yields the fluorescence lifetime as the decay time of the exponential. Complications may arise when the decays are actually nonexponential. In the case of relatively long lifetimes in the nanosecond range, instead of TCSPC a gated image intensifier may be used to measure the number of photons falling into a time window defined by the gate. Both techniques operate in the time domain. When the lifetime is comparatively long, it is also possible to modulate the laser pulses in the MHz range and detect the phase shift of the corresponding modulation in detected fluorescence intensity, which depends in a simple way on the fluorescence lifetime. TCSPC, however, is the most flexible way of measurement, because it is not restricted to long lifetimes and allows to analyze non-exponential decays as well. Fluorescence Resonance Energy Transfer (FRET) Fluorescence Lifetime Imaging (FLIM) The lifetime of a fluorophore may vary depending on local pH, oxygen or ion concentrations, or on intermolecular interactions, for example, fluorescence resonance energy transfer (FRET, see section Fluorescence Resonance Energy Transfer (FRET)). On the other hand, within some limits, it does not respond to the intensity of the excitation light, the fluorophore concentration, or photobleaching. The fluorescence lifetime is therefore a useful physical quantity that can be used for imaging and quantitative analysis of local pH, ion concentrations, or intermolecular interactions. Fluorescence lifetimes are typically in the range of a few picoseconds, when dominated by nonradiative processes, to several tens of nanoseconds, when limited by the radiative transition rate. In some cases it is useful to employ fluorophores with very long lifetimes, in particular when strong autofluorescence is present. Performing a lifetime measurement at each position of the excitation laser focus within the sample and representing the corresponding values as a gray scale or color value from a look-up table yields a confocal lifetime image. At the same time, a standard fluorescence intensity image can be obtained. A common method of measuring fluorescence lifetime is time-correlated single photon counting (TCSPC). Here, a correlator, triggered by the short FRET [11] is a resonant non-radiative energy transfer from a donor to an acceptor fluorophore due to the dipolar interaction. As the dipolar interaction energy is proportional to the third power of the donor acceptor distance R and because the probability for FRET to occur involves the square of an off-diagonal matrix element of the dipolar interaction, it falls off as R6 and depends on the relative orientation between the molecules and the spectral overlap between the emission spectrum of the donor and the absorption spectrum of the acceptor. The probability is highest for parallel orientation of the donor and acceptor transition dipoles. Because of the steep fall off FRET can only occur if R is sufficiently small, that means, if the donor and the acceptor are sufficiently close to each other. Because the critical distance is only 1–10 nm, typically 4–6 nm, in all practical cases, the FRET mechanism provides a molecular ruler for measuring the donoracceptor distance and in this way provides an indirect mechanism for studying structure on the nanometer scale, which cannot directly be resolved by diffraction-limited optical microscopy (super-resolution microscopy, compare section Super-Resolution, might, however, in the future partly supersede FRET studies). This is of particular interest for protein– protein and intra-protein interactions. For this purpose, the proteins of interest have to be labeled by Confocal Laser Scanning Microscopy fluorophores with overlapping emission and absorption spectra. In many cases fluorescent proteins (compare section Fluorescent Probes) are used for this purpose. A sensitive measure for FRET, which can be used for imaging, is the ratio of intensities of the donor and acceptor fluorescence peaks. A problem with FRET confocal microscopy is that the signal-to-noise ratio is often very poor. Therefore, in many cases only the occurrence or absence of FRET is detected. The signal-to-noise ratio may be improved by measuring the donor lifetime, instead of the acceptor/donor fluorescence intensity ratio. This imaging option is usually known as FLIM-FRET (fluorescence lifetime imaging – fluorescence resonance energy transfer). Fluorescence Recovery After Photobleaching (FRAP) This microscopy technique, most often performed in a confocal set-up, bleaches the fluorescence of a certain sample region, often an intracellular organelle, and registers the recovery of fluorescence due to diffusion of the fluorescently labeled molecules. It may therefore be used to investigate the mobility of the target molecule within the surrounding structures. Fluorescence Correlation Spectroscopy (FCS) In contrast to the previous techniques, FCS is usually used not as an imaging technique, but with a fixed position of the confocal volume within the sample. The laser (usually a continuous wave laser) thereby excites fluorescent particles within this volume and particle movement in and out of the volume produces fluorescence intensity fluctuations. The autocorrelation function of these fluctuations provides information about the concentration, diffusion coefficient, and the mass of the particles. The diffusion coefficient depends on the viscosity of the medium via the Einstein–Smoluchowski relation. It may also depend on interactions of the particles with a micro-structured environment. A variant of FCS is Fluorescence Cross Correlation Spectroscopy, in which the fluorescence intensity fluctuations of two fluorophores, labeling two different molecules, are measured simultaneously in two different channels. If the two molecules are bound in a dimer, the fluctuations will be highly correlated. Otherwise, no cross correlation is expected. The degree of cross correlation then is a measure of how many of the two different species of molecules are bound to each other. 513 C Confocal Microscopy Sensitive to Phase Interferometric versions of confocal laser scanning microscopy have been reported relatively early in the literature and were based on Mach-Zehnder or Michelson interferometers, for measurements in transmission or reflection, respectively [1]. In this way, it is possible to measure object topography or refractive index variations with interferometric precision. Interferometric confocal microscopes have the advantage over conventional interferometric microscopes that the shape of the wave front of the reference beam is irrelevant because only phase and amplitude at the pinhole is important. This means that the requirement for matching optics, otherwise necessary in interferometric microscopy, is strongly relaxed. Often two beam splitters and two detectors, each with its own pinhole, are employed in the detection beam path of interferometric confocal microscopes based on Michelson interferometers. This allows to separate the conventional confocal signal and the pure interference signal as the sum and difference of the outputs of the two detectors. This is based on the fact that two beams at the inputs of a symmetric lossless beam splitter are combined with a +p/2, p/2 phase shift relative to each other at the two outputs of the beam splitter, respectively. A non-interferometric variant of phase-sensitive confocal microscopy, differential phase-contrast confocal microscopy, is obtained by omitting the mirror generating the reference beam in the Michelson interferometer and obscuring one half of each of the relay lenses in a complimentary manner [2]. The sensitivity of interferometric confocal microscopes may be enhanced by using a spectrally shifted reference beam (heterodyne interferometric confocal microscopy). Topographic resolution of about 0.01 nm using such a heterodyne interferometric confocal microscope has been reported. In addition, the optical sectioning capability of the confocal microscope and the corresponding dependence of the signal on defocus can be used to avoid phase unwrapping ambiguities. Another variant of interferometric confocal microscopy is called 4p microscopy. Here, two opposing objective lenses are used for excitation and/or detection. This increases the available numerical aperture and therefore enhances the resolution, in particular in the axial direction. In addition, because of the coherent excitation the technique is inherently interferometric due to the interference of the two counterpropagating C C 514 excitation beams in the focal region. This generates a standing wave interference pattern which modulates the main lobe of the focus in the axial direction and in this way allows to improve the axial resolution by a factor of about 4.5. The side lobes, due to interference, within the confocal main lobe may be effectively suppressed in the case of two-photon excitation. This suppression is due to the nonlinearity of the excitation process and can be additionally enhanced, if the sample fluorescence is also detected through both objective lenses in an interferometric setup. Because the position of maxima of the PSF for excitation and detection depends on the phase of both beams used for excitation and detection, precise control of the phase of both beams is required. It should be mentioned here that a non-interferometric technique for phase measurement in optical microscopy is based on the so-called transport-of-intensity equation which can be derived from the paraxial time-dependent wave equation and relates the intensity and phase of a paraxial monochromatic wave to its longitudinal intensity derivative. The technique requires, however, the measurement of very small changes of intensity as a function of small defocus and therefore requires high bit resolution and long integration times. Super-Resolution Historically, the Abbe diffraction limit had been an insurmountable barrier in optical microscopy for many years. Many structures of interest, on the other hand, are significantly smaller than this limit and, particularly in biology, there has been a strong desire to significantly improve the resolution. In recent years, the diffraction limit has been broken in numerous ways, and this is based on some of the most exciting modern advancements in optics, which is still, although very old, a rapidly developing area of physics. The conceptually simplest way to achieve subAbbe resolution is by way of deconvolution (compare section Deconvolution). This does not, however, allow to achieve resolution enhancements of an order of magnitude, as urgently desired in many cases. One route to satisfy this demand employs non-propagating near fields. These near fields carry all optical information on length scales below the Abbe limit. The fact that these near fields are lost in far-field optical imaging can be viewed as the reason for the Confocal Laser Scanning Microscopy limited resolution of standard (including confocal) optical microscopes. Near-field related techniques (compare ! ▶ Scanning Near-Field Optical Microscopy) will not be discussed here (as an exception, compare TERS, section Confocal Raman Microscopy), although some of them may be combined with confocal imaging, for example solid immersion lens microscopy (SIL) or total internal reflection microscopy (TIRF). Other routes, however, have opened the way to far-field optical nanoscopy in recent years. They are based on prior knowledge about the sample (PALM/ STORM) or on optical nonlinearity (STED, SSIM) [12]. STED uses a specially designed optical excitation and is a scanning microscopy technique. PALM/STORM is non-confocal, because parallel detection using a CCD camera is used. These techniques have reached lateral resolution in the range of 20 nm and ongoing research attempts to further push the resolution toward the molecular level (1–5 nm). Commercial instruments based on these techniques are increasingly becoming available [13]. The techniques will be shortly described, because of their potential as far field microscopy techniques to partly supersede standard confocal fluorescence laser scanning microscopy. Super-Resolution by Prior Knowledge (Profilometry, PALM/STORM) In cases where it is known that one and only one sharp interface between two homogeneous materials or one and only one point emitter or reflector in the focus is present, the position of the interface in the z direction or the three-dimensional position of the point emitter/ reflector can be determined with a precision that is far beyond the Abbe limit and is limited only by the total amount of photons detected. The simplest such technique consists in profiling a surface using a standard non-interferometric CLSM. By adjusting the z position to the wing of the V(z) function, very slight changes in the topography of the sample surface can be measured, if the material and, hence, the reflectivity do not vary. Sub-nanometer depth resolution has been achieved in this way, averaged over the diffraction-limited lateral size of the confocal spot. As discussed in section Confocal Microscopy Sensitive to Phase the technique may be combined with interferometry to increase the depth resolution to about 0.01 nm. In a principally similar way photoactivated localization microscopy (PALM) and stochastic optical Confocal Laser Scanning Microscopy reconstruction microscopy (STORM) determine the position of single-point emitters with nanometric precision. Because it must be avoided to have more than one molecule in the focal volume and such a sparse distribution of fluorescent emitters would not allow sub-diffraction limited resolution, it is necessary to separate the contributions from many individual molecules in some way. Temporal separation is key to current single molecule based super-resolution techniques. Other options, such as spectral separation, or more sophisticated schemes have also been discussed, however. Current techniques use photoactivable molecules, which can be statistically turned on by another light source. Irreversible or reversible processes may be employed to turn off an activated subset. Determination of the positions of the activated molecules and multiple repetition then results in a software-generated image with nanometric resolution. Typically, thousands of images must be acquired and, correspondingly, imaging is slow, with a trade-off between spatial and temporal resolution. Effective frame rates of reconstructed images of about 1/(3 min) at a Nyquist-limited resolution of about 50 nm have been demonstrated for frames of 50  50 pixels. The obtainable resolution depends on the brightness of the fluorophores in the “on” state, as well as on the achievable contrast between “on” and “off” states. Fluorescent proteins as well as fluorescent dyes are currently in use [14]. PALM has hitherto mostly been used in total internal reflection configuration and is then restricted to essentially two-dimensional imaging. Lateral resolution down to 20 nm has been achieved. A threedimensional STORM technique has achieved an image resolution of 20–30 nm in the lateral dimensions and 50– 60 nm in the axial dimension in an illuminated sample volume of a few micrometers in thickness. An interferometric variant of PALM employing self-interference of fluorescent photons has improved the axial resolution to <20 nm for optically thin samples. It may finally be worth noting that, whereas photoswitching is considered by some authors as a kind of optical nonlinearity, an optically nonlinear response is not essential to the breaking of the diffraction limit in the far field in the case of single molecule based techniques. For example, sub-diffraction imaging due to binding and detachment of diffusing probes has been demonstrated, which obviously does not involve any optical nonlinearity. Moreover, photoswitching and temporal 515 C selection of molecules in general represent conceptually only one, although currently the standard and most successful option to ensure that only one molecule is detected within the PSF of the objective lens. It may therefore be said that in this case the breaking of the diffraction barrier is based on the knowledge, however obtained, to have only one molecule within the PSF and on the facts that the center-of-mass position of a diffraction-limited spot can be determined to a much higher precision than given by the size of this spot and that this position corresponds to the position of the fluorescent molecule. Super-Resolution due to Nonlinearity (SSIM, STED) Some examples of confocal imaging methods employing optical nonlinearity have already been discussed in sections Confocal Raman Microscopy (CARS, SRS), and Multiphoton Microscopy. By reducing the size of the PSF for photons of the same wavelength optical nonlinearity is in principle able to enhance the resolution. In multiphoton microscopy, however, because long wavelength photons are used for excitation of fluorophores, the resolution is actually worse than that of its linear counterpart. Saturated structured illumination microscopy, SSIM, on the other hand, provides resolution enhancement without fundamental limit. This nonlinear variant of SIM (compare section Structured Illumination Microscopy (SIM)) employs a nonlinear dependence of the fluorescence intensity on the excitation density, which is possible, for example by saturation of the excited state. In this case, the effective illumination generally contains higher spatial frequency components leading to increased resolution. Optical resolution beyond 50 nm has been reported using this technique. Stimulated emission depletion microscopy (STED) selectively de-excites fluorescing molecules within the PSF of the confocal microscope by stimulated emission, using a second laser. This laser beam passes a spiral phase mask, which generates a zero of intensity in the center of the beam cross section. Stimulated emission and corresponding deexcitation of fluorophores then occur everywhere within the PSF, except close to the axis of the beam, and the size of the region, from which fluorescence can be collected depends on the intensity of the deexciting laser. Lateral resolution of about 20 nm has been reported using this setup. If STED is combined with a 4Pi setup, fluorescing spherical focal volumes of 40–45 nm diameter can C C 516 be generated. STED is currently able to operate considerably faster than PALM. Video-rate STED microscopy with about 60 nm lateral resolution has been reported with a frame rate of 28 frames/s. Confocal Scanning Optical Microscopy (CSOM) Confocal Scanning Optical Microscopy (CSOM) ▶ Confocal Laser Scanning Microscopy Cross-References ▶ Optical Techniques for Nanostructure Characterization ▶ Scanning Near-Field Optical Microscopy Conformal Electronics ▶ Flexible Electronics References Contour-Mode Resonators 1. Corle, R.C., Kino, G.S.: Confocal Scanning Optical Microscopy and Related Imaging Systems. Academic, San Diego (1996) 2. Wilson, T.: Confocal Microscopy. Academic, London (1990) 3. Born, M., Wolf, E.: Principles of Optics, 6th edn. Pergamon, Oxford (1980) 4. Sarder, P., Nehorai, A.: Deconvolution methods for 3-D fluorescence microscopy images. IEEE Signal Proc. Mag. 23, 32–45 (2006) 5. Richards, B., Wolf, E.: Electromagnetic diffraction in optical systems. II. Structure of the image field in an aplanatic system. Proc. Roy. Soc. A 253, 358–379 (1959) 6. Sheppard, C.J.R., Török, P.: Efficient calculation of electromagnetic diffraction in optical systems using a multipole expansion. J. Mod. Opt. 44, 803–818 (1997) 7. Chudakov, D.M., Matz, M.V., Lukyanov, S., Lukyanov, K.A.: Fluorescent proteins and their applications in imaging living cells and tissues. Physiol. Rev. 90, 1103–1163 (2010) 8. Ormo, M., Cubitt, A.B., Kallio, K., Gross, L.A., Tsien, R.Y., Remington, S.J.: Crystal structure of the Aequorea victoria green fluorescent protein. Sci. 273, 1392–1395 (1996). Image from the RCSB PDB (www.pdb.org) of PDB ID 1EMA (http://dx.doi.org/%2010.1021/ja1010652) 9. Gr€af, R., Rietdorf, J., Zimmermann, T.: Live cell spinning disk microscopy. Adv. Biochem. Engin./Biotechnol. 95, 57–75 (2005) 10. Langhorst, M.F., Schaffer, J., Goetze, B.: Structure brings clarity: structured illumination microscopy in cell biology. Biotechnol. J. 4, 858–865 (2009) 11. Piston, D.W., Kremers, G.J.: Fluorescent protein FRET: the good, the bad and the ugly. Trends Biochem. Sci. 32, 407–414 (2007) 12. Hell, S.W.: Far-field optical nanoscopy. Science 316, 1153–1158 (2007) 13. Chi, K.R.: Ever-increasing resolution. Nature 462, 675–678 (2009) 14. Heilemann, M., Dedecker, P., Hofkens, J., Sauer, M.: Photoswitches: key molecules for subdiffraction-resolution fluorescence imaging and molecular quantification. Laser Photon. Rev. 3, 180–202 (2009) ▶ Laterally Vibrating Piezoelectric Resonators Coupling Clamp ▶ Dynamic Clamp Creep ▶ Nanomechanical Properties of Nanostructures Cutaneous Delivery ▶ Dermal and Transdermal Delivery Cuticle ▶ Arthropod Strain Sensors Cylindrical Gold Nanoparticles ▶ Gold Nanorods