Review Article

https://doi.org/10.1038/s41566-018-0310-5

Integrated microwave photonics

David Marpaung 1
*, Jianping Yao 2
and José Capmany3

Recent advances in photonic integration have propelled microwave photonic technologies to new heights. The ability to interface
hybrid material platforms to enhance light–matter interactions has led to the development of ultra-small and high-bandwidth
electro-optic modulators, low-noise frequency synthesizers and chip signal processors with orders-of-magnitude enhanced
spectral resolution. On the other hand, the maturity of high-volume semiconductor processing has finally enabled the com-
plete integration of light sources, modulators and detectors in a single microwave photonic processor chip and has ushered the
creation of a complex signal processor with multifunctionality and reconfigurability similar to electronic devices. Here, we
review these recent advances and discuss the impact of these new frontiers for short- and long-term applications in communi-
cations and information processing. We also take a look at the future perspectives at the intersection of integrated microwave
photonics and other fields including quantum and neuromorphic photonics.

T
he use of optical devices and techniques to generate, manip- technological tools for integrated MWP, such as Kerr microresona-
ulate, transport and measure high-speed radio-frequency tor combs17, hybrid organic–plasmonic modulators18 and on-chip
(RF) signals, widely known as microwave photonics (MWP; stimulated Brillouin scattering (SBS)19. These tools married with
Box 1), has been the focus of intense research activities in recent entirely new concepts, for example making a universal reconfigu-
years1–5. The promise of abundant processing bandwidth obtained rable processor20, can significantly alter the capabilities of MWP
from upconverting radio-frequencies to optical frequencies, the systems to achieve higher performance, including modulation
availability of low-loss optical fibres as a transport medium, and bandwidth, spectral resolution, noise performance and reconfigu-
the flexibility in tailoring the RF response over decades of fre- rability. On the other hand, major advances in chip integration using
quency unlike anything achievable by traditional RF systems have a single material (monolithic) or multi-materials (hybrid or hetero-
been cited as key drivers in the early development of the technol- geneous) have allowed the integration of all the key components of
ogy. Landmark demonstrations include, among others, generation integrated MWP systems in a single chip21. The synergies between
of ultra-broadband signals6,7, distribution and transport of RF over integration, advanced functionalities and high performance have
fibre8, programmable MWP filters9,10 and a photonics-enhanced been the key highlights of the field in recent times.
radar system11. These progresses have subsequently positioned Here, we review the most recent advances in integrated MWP.
MWP as a prime technology solution to the impending challenges We focus on the new technological tools for integrated MWP, as
in communications, including the bandwidth bottleneck in com- illustrated in Fig. 1, which are derived from recent breakthroughs
munications systems12 and the Internet of Things, provided that and advances in integrated optics. These advances have consider-
the hurdles of size, reliability and cost effectiveness are overcome. ably expanded the performance and the scope of the field, allowing
While impressive the abovementioned demonstrations were in intersections with emerging technologies such as quantum photon-
bulky systems composed of relatively expensive discrete fibre-optic ics, optomechanics and neuromorphic photonics. We also provide a
components, which are sensitive to external perturbations such as perspective of how these fields can interact as well as the short- and
vibrations and temperature gradients. long-term applications of the technology.
It was thus serendipitous, but essential, that the rise of MWP
technology was paralleled by the surge of photonic integration Hybrid integration and emerging materials
technologies. The convergence of the two fields, aptly termed Material platforms are a key element in integrated MWP. The choice
integrated MWP, soon followed with profound impacts5,13. of material will determine the range of functionalities, performance
Leveraging photonic integration allowed a dramatic reduction and the size of the devices and systems. At its inception5, integrated
in the footprint of MWP systems with fairly high complexity, MWP was mainly demonstrated in a diverse range of materi-
making them more comparable to RF circuits. Optical loss in als that included GaAs, lithium niobate and doped silica. But for
MWP systems is important because an increase in this param- the past 10 years, most of the integrated MWP circuits have been
eter translates quadratically into RF loss in RF circuits5. For these based on three key platforms for monolithic integration: indium
reasons, efforts at the early stages of development of integrated phosphide22, silicon-on-insulator (SOI)23–25 and silicon nitride26,27.
MWP were focused on reducing on-chip losses14, on integrating Maturity in the fabrication processes of these materials and their
as many components as possible in a single chip15 and to demon- availability through cost-sharing initiatives that dramatically reduce
strate device reconfiguration7,16. the fabrication cost were the two key drivers of this polarization.
But integrated photonics offers much more than a reduction in The strengths and weaknesses of each of these materials are sum-
footprint and complexity. For example, confining light in the small marized in Box 2. Table 1 provides a comparison of these materials
mode volume enhances its interaction with matter, most of the in terms of the technology features and the quality of components
time through nonlinear optical processes, which resulted in new and functionality.

1
Laser Physics and Nonlinear Optics Group, Faculty of Science and Technology, MESA+​Institute for Nanotechnology, University of Twente, Enschede,
The Netherlands. 2School of Electrical Engineering and Computer Science, University of Ottawa, Ottawa, Ontario, Canada. 3ITEAM Research Institute,
Universitat Politècnica de València, Valencia, Spain. *e-mail: d.a.i.marpaung@utwente.nl

80 Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics

NaTuRe PHOTOnics Review Article
The most rapid advances occurred in the area of III–V integra-
Box 1 | Fundamentals of MWP tion with silicon or silicon nitride. These circuits were designed to
A canonical MWP system consists of a laser light source, an optical provide reliable light sources and modulators (in III–V semicon-
modulator, an optical signal processor and a photodetector. An ductor materials), low-loss circuits (in the case of silicon nitride),
input RF signal with frequencies fRF, acquired from an antenna or a versatile platform with electronic integration potential in the
or an RF source, modulates the output of an optical source to case of silicon. Hybrid devices have been exploited to show basic
upconvert its spectrum to optical frequencies. This typically and advanced functionalities, such as hybrid metal-oxide–semi-
forms a pair of sidebands at frequencies ν ±​ fRF, where ν is the conductor (MOS) Mach–Zehnder modulators32,33, high-gain, high-
central frequency of the optical source. The combined optical saturation semiconductor optical amplifiers, and integrated optical
signal is then processed by an optical system composed of sev- and RF sources29,34. Integration of multiple materials (silica, silicon
eral photonic devices forming an optical signal processor, which nitride, III–V-on-silicon) was recently attempted to create a precise
modifies the spectral characteristics of the sidebands. Finally, an optical frequency synthesizer35. Despite these advances consensus
optical detector is employed to downconvert the processed side- has not been reached on which is the most suitable approach for
bands by beating with the optical carrier so the processed RF MWP applications since, ideally, simultaneous photonic, RF and
signal is recovered. complementary metal-oxide–semiconductor (CMOS) electronic
The functionalities that can be carried out using an MWP compatibility36 should be targeted.
system include antenna remoting, RF photonic filtering, true- Apart from the main material platforms discussed above, other
time delay, phase shifting, optical beamforming, arbitrary material systems have been considered as well for integrated MWP
waveform generation, frequency up- and downconversion, implementation, albeit on a smaller scale and volume, and were
microwave signal generation and frequency measurement. The focused more on particular light–matter interactions of nonlin-
quality and range of these functionalities are typically dictated by ear optics, optomechanics and plasmonics. Chalcogenide glass37
the optical signal processor. (Fig. 1d), a highly nonlinear material with low two-photon absorp-
The performance of an MWP system is usually expressed by tion, has mainly been exploited for nonlinear opto-acoustic pro-
three key parameters, namely, the link gain, noise figure and the cessing based on Brillouin scattering. Emerging materials such as
SFDR. The link gain is a measure of the RF-to-RF loss occurring Hydex27, Ta2O5 (ref. 38) and aluminium nitride39 can also be of inter-
in the system while the noise figure quantifies the degradation of est for integrated MWP. Moreover, the availability of the lithium
the signal-to-noise ratio in the system. The SFDR considers the niobate-on-insulator40 platform has rekindled interest in lithium
RF nonlinearity occurring in the system and measures the range niobate circuits that have been shown to realize ultra-low-loss
of RF powers that can be accommodated in the system with waveguides and compact modulators (see section ‘Advanced opti-
sufficient signal-to-noise ratio and negligible intermodulation cal modulation’). Finally, 2D materials, especially graphene, on SOI
distortion. These parameters are largely influenced by the output have been proposed for the implementation of high-speed modula-
power and the intensity noise of the laser source, the insertion tors41, phase shifters, true time delay units and tunable filters42.
loss, half-wave voltage and linearity of the optical modulator, and
the power handling and responsivity of the photodetector. Key functionalities
Functionalities implemented in integrated MWP span optical mod-
ulation, generation, processing and measurement of microwave sig-
nals. Integration not only brings advantages to these functions, but
also enhanced performance. Here, we look at the technological tools
Receiver
that have become available in the past few years that have poten-
Antenna tially redefined the field of integrated MWP.
RF input RF output
Advanced optical modulation. Optical modulation is the first step
in all MWP systems, where the RF signal is encoded in the opti-
cal domain. This is a pivotal step that often determines the overall
system performance, including bandwidth, system loss, linearity
Laser source Optical modulator and dynamic range5. Typical modulators used in MWP43 are phase,
Optical signal Photodetector
processor intensity, or IQ modulators to generate single-sideband modulation
or complex modulation with tunable sideband phase and ampli-
tude19. In some works, polarization modulators are also utilized44.
Generally, optical modulation can be achieved through the varia-
tion of material absorption (for example in electro-absorption mod-
It has been clear for some time that none of these main material ulators) or through refractive index changes (for example in phase
platforms can provide all the required performance for integrated modulators and Mach–Zehnder-type intensity modulators).
MWP on their own. Attempts to create an all-integrated MWP chip For decades, the material of choice for optical modulation has
in a single platform (in this case indium phosphide) for functional- been lithium niobate45. In this material, a refractive index change
ities such as tunable filters21 and interference cancellation circuits28 is obtained through the Pockels electro-optic effect. Typical elec-
have been reported with encouraging results, but issues related to tro-optic lithium niobate modulators are formed either by proton
waveguide loss and elevated noise due to on-chip amplifiers limit exchange or by the diffusion of titanium, leading to low-index-
the performance of such circuits. Hence, researchers have turned to contrast waveguides with poor optical confinement, which directly
a different approach, namely hybrid or heterogeneous integration, leads to high drive voltages and large size. But recent availability
to combine different material platforms and to take advantage of of lithium niobate-on-insulator films and improvements in lithium
their individual strengths. Different approaches have been proposed niobate etching techniques have changed this landscape dramati-
for this, namely, chip-to-chip (hybrid) integration through vertical cally46–48. High-contrast, etched lithium niobate waveguides were
or edge coupling29, or wafer-scale (heterogeneous) integration tech- recently used to form compact miniaturized lithium niobate modu-
niques such as wafer bonding or direct epitaxial growth30,31 that are lators a few hundred micrometres in length46,47 (Fig. 2a). With such a
more suitable for mass production. technique, impressive modulator performance was achieved, includ-

Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics 81

Review Article NaTuRe PHOTOnics

Programmable processors

High-resolution filters
As2S3 ring

Silicon taper

Multipurpose signal processor

Low-noise sources

Silicon grating
tor coupler
tec On-chip Brillouin scattering
de
oto
Ph

Microresonator
frequency combs r
o
ss
ce
pro
n al
e
ur c Sig
l so
tica
Op
or
lat
o du
lm
tica
Op

Advanced modulators
Au Si SiO2
Quadrature
MZM
Multi-material integration
Input Output
2 × 2 MMI 2 × 1 MMI 4.4 µm
SiO2
III–V/Si laser
CW pump
18.5 µm Lactive In-phase 15 µm
PD 5 µm
MZM
CMOS
Si3N4
f–2f Plasmonic modulators

Fig. 1 | Overview of recent advances and technologies in integrated MWP. Kerr microresonator frequency combs can be used as high-quality
multiwavelength optical sources for microwave generation and tapped-delay-line microwave photonic filtering. Plasmonic modulators18 enable ultra-
high frequency modulation compatible with silicon platforms and direct RF-to-optical conversion. On-chip Brillouin scattering84 provides high-resolution
filtering unmatched by other on-chip optical filters. A new paradigm of universal programmable processors98 brings flexibility to MWP and intersects the
field with integrated quantum photonics and neuromorphic photonics. Many of these advances are enabled by progresses in hybrid and heterogeneous
multi-material integration35. CW, continuous wave; MMI, multimode interferometer; f–2f, frequency doubling. Figure adapted from: programmable
processors, ref. 98, under a Creative Commons licence (https://creativecommons.org/licenses/by/4.0/); high-resolution filters, ref. 84, OSA; advanced
modulators, ref. 18, IEEE; multi-material integration, ref. 35, Springer Nature Limited. Image of low-noise sources courtesy of the Laboratory of Photonics and
Quantum Measurements, EPFL, Swiss Federal Institute of Technology.

ing record-low drive voltage (Vπ) of 1.4 V for a single modulator Driven by the demand of a high-speed modulator with a small
and a bandwidth of 40 GHz in a 20-mm-long modulator47 (Fig. 2b). footprint compatible with silicon photonics technology, research-
The chip insertion loss of the modulator was less than 0.5 dB with 5 ers have proposed a hybrid integration route to combine materials
dB per facet coupling loss. Higher-frequency operation at 100 GHz that have a high electro-optic coefficient with strong mode confine-
was also demonstrated with the VπL figure of merit of 2.2 V cm, ment in silicon. In silicon–organic hybrid modulators52, the strong
where L is the length of the modulator. confinement comes from a silicon slot waveguide, whereas in plas-
Optical modulation has also been explored in silicon, indium monic–organic hybrid modulators (Fig. 2c) both optical and RF
phosphide and silicon nitride. The most straightforward way to signals are guided by thin metal sheets, that is, a metal slot wave-
achieve optical modulation in silicon is to exploit the free-carrier guide where the light propagates as a surface plasmon polariton
plasma dispersion effect, where changes in the electron and hole mode18,53–56. A plasmonic modulator has several advantages over
densities modify the refractive index and absorption of a silicon a silicon–organic hybrid modulator, including ultra-short length
waveguide49. Modulators based on this effect have shown excel- (tens of micrometres) and higher bandwidth due to ultra-small
lent performance including high-speed operation, but the concept capacitance. Bandwidth as high as 170 GHz has been achieved
does not allow high performance to be achieved simultaneously with plasmonic devices55 (Fig. 2d). These modulators have also
with low drive voltage and a small footprint. Optical modulation been seamlessly integrated with an antenna for direct conversion of
in the indium phosphide-based platform can be achieved via the millimetre waves to the optical domain56.
quantum-confined Stark effect that induces absorption and refrac- While improvements in bandwidth, size and energy consump-
tive index changes22. These indium phosphide modulators have tion are well within reach for a new class of modulator devices, an
shown wide bandwidth beyond 55 GHz (ref. 50). In silicon nitride, aspect that is often overlooked is linearity, which is indispensable for
on the other hand, modulation can only be achieved through hybrid analog and RF photonics applications. Linearization of the silicon
integration with indium phosphide29, 2D materials including gra- modulator for analog applications has been explored through het-
phene41, or ferroelectric materials such as lead titanate zirconate51. erogeneous integration with lithium niobate57 or III–V materials58

82 Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics

NaTuRe PHOTOnics Review Article
Another route to microwave generation is via an optoelectronic
Box 2 | The main material platforms for MWP oscillator (OEO)60 that is an MWP link consisting of a laser, a mod-
Indium phosphide is the only material that enables the mono- ulator, a high-Q optical cavity and a photodetector with a sustained
lithic integration of various active and passive photonic compo- oscillation through electronic feedback from the detector to the
nents, including lasers, modulators, optical amplifiers, tunable modulator. Numerous schemes to realize an OEO based on discrete
devices and photodetectors, and the platform allows the creation components have been reported, but recently, interest in integrated
of compact circuits with bending radius of the order of 100 µ​m. OEOs has peaked. For example, an integrated OEO, where the
But optical waveguides in this material have relatively high losses optical components, including a directly modulated laser source,
of the order of 1.5–3 dB cm–1. a spiral-shaped optical delay line and a high-speed photodetector,
Silicon photonics offers compatibility with microelectronic were fabricated on an indium phosphide substrate was reported61.
CMOS fabrication processes making electronic–photonic co- In another approach, integration on a silicon photonic chip using a
integration a real possibility. Silicon-on-insulator waveguides microdisk resonator as an optical filter was explored62. These initial
can exhibit a wide range of losses (0.1–3 dB cm–1) and minimum demonstrations, however, were plagued by high phase noise due to
bending radii (5–100 µ​m), depending on the thickness of the the lack of a high-Q resonator as the optical storage element. For
silicon layer (‘thin’ ~220 nm or ‘thick’ ~3 µ​m) and the geometry example, in ref. 61, the measured phase noise for an output frequency
of the waveguides (strip or rib). Silicon also shows strong third- of 7.3 GHz was –91 dBc Hz–1 at a 1-MHz offset frequency, which
order optical nonlinearities, which can be advantageous for was nearly three orders of magnitude higher compared with an RF
ultra-fast signal processing. But the material also suffers from electronic oscillator. A key challenge to address then is to integrate
high nonlinear loss through two-photon absorption (TPA) and an ultra-high-Q cavity63–65 with the rest of the OEO components.
free-carrier absorption (FCA). Thus silicon is a poor material for The most promising technology for pure microwave genera-
light sources, optical modulators, or photodetectors. But through tion is Kerr frequency combs63–68. The beat note generated from the
doping, high-speed modulators and photodetectors have been multiple phase-locked harmonics of the combs creates an extremely
demonstrated. high spectral purity at a frequency corresponding to the comb spac-
Silicon nitride waveguides are gaining popularity due to ing. These combs benefit from the division of the noise of the opti-
the potential of ultra-low-loss operation. Depending on the cal source dictated by the ratio of optical to beat note frequencies68.
deposition method of these layers, post-fabrication processing Harnessing these combs and other nonlinear effects in resonators,
and geometry, silicon nitride waveguides can be easily tailored such as Brillouin scattering, has resulted in integrated oscillators
to exhibit ultra-low propagation losses (0.01–0.2 dB cm–1), with impressive performance. For example, a Kerr optical fre-
be relatively compact (50–150 μ​m bending radius), or be able quency comb excited in a high-Q magnesium fluoride whispering-
to undergo dispersion engineering necessary for third-order gallery-mode resonator was used to generate a microwave signal68 at
nonlinear processes. The nonlinear coefficient of silicon nitride 9.9 GHz with a phase noise as low as –120 dBc Hz–1 at a 1-kHz offset
is about ten times lower compared with silicon. But the low loss frequency. In another key demonstration, cascaded Brillouin oscil-
and, crucially, the absence of TPA and FCA make it a material lation on an ultra-high-Q planar silica wedge resonator was used
of choice for microresonator-based Kerr frequency combs. to synthesize 21-GHz microwave signals with a record-low phase-
Lasers, modulators and detectors can be constructed in this noise floor of –160 dBc Hz–1 (ref. 69). These performances are com-
passive material only through hybrid integration with other parable to that of RF oscillators69.
material platforms including indium phosphide, or graphene, or
piezoelectric materials. Integrated MWP filters. Microwave filtering is one of the most
important and fundamental functions in signal processing used to
separate information signals from unwanted signals such as noise
and interferences. In general, there are two ways of implementing a
where the RF third-order nonlinearity term arising from the Mach– MWP filter: via a tapped-delay-line architecture, where a number of
Zehnder interferometer transfer function can be cancelled by the RF signal copies with well-tailored amplitude and delay profile are
nonlinearity of the quantum-confined Stark effect in the III–V summed to form a periodic frequency response, or via a downcon-
material59 (Fig. 2e). Dynamic range as high as 117 dB Hz2/3 has been verted response of an optical filter response5,9.
achieved using such modulators (Fig. 2f). Tapped-delay-line MWP filters typically consist of three key
components: a multiwavelength optical source to form the differ-
Low-noise sources and frequency combs. Traditionally, a micro- ent taps, a spectral shaper for shaping the amplitude and phase of
wave signal is generated using an electronic oscillator with many each of the taps, and a dispersive delay line for forming the basic
stages of frequency doubling to generate a microwave signal with unit delay between the taps, thereby dictating the filter free spectral
the desired frequency. But with frequency multiplication, the phase range70. Partial integration of these components has recently been
noise performance of a microwave signal would be degraded by explored (Fig. 3a). Microresonator-based frequency combs have
10log10M2, where M is the multiplication factor. Microwave photon- emerged as a powerful tool as a multiwavelength source71,72. For
ics, on the other hand, allows direct synthesis of radio-frequencies example, a Kerr frequency comb containing 45 lines based on a sili-
from the optical domain, opening the path to the generation of very con nitride chip was used as an optical source in a single band-pass
high frequencies with low phase noise. microwave photonic filter (MPF)71. In a similar work, a comb source
A straightforward way to optically generate microwaves is to in Hydex material was used in a tapped-delay-line MPF for Hilbert
beat the output of two lasers on a photodetector. For example, a transformation72. In addition, these combs have also been explored
microwave synthesizer in a heterogeneously integrated indium as sources for true time delay feeding of phased-array antennas72.
phosphide–silicon chip was recently reported34. The chip includes A pulse shaper has been integrated on an indium phosphide plat-
a high-speed photodetector, two tunable laser sources and a pair of form to modify the tap weight distribution using a 32-channel
phase modulators. By tuning the wavelength of one laser source, a arrayed waveguide grating and a dedicated semiconductor optical
microwave signal with a frequency tunable from 1 to 112 GHz was amplifier for line-by-line control of the channel gain73. By exploiting
generated. While simple and the frequency tuning is large, the phase the fast-programmable capability of the pulse shaper, a filter with
terms of the two optical waves are not correlated and the phase noise fast reconfigurability was realized. As for the dispersive delay lines,
of the generated microwave signal is high. photonic crystals in III–V materials have been considered as an

Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics 83

Review Article NaTuRe PHOTOnics

Table 1 | Parameters of the main material platforms relevant for MWP

Silicon-on-insulator Silicon nitride Indium phosphide
Refractive index 3.5 2.1 3.1
Waveguide refractive index contrast (%) >​100 >​25 10
Bending radius (µ​m) 5–100 50–150 100
Loss (dB cm ) –1
0.1–3 0.01–0.2 1.5–3
Nonlinear index (m2 W–1) 4.5 ×​ 10–18 2.6 ×​ 10–19 1.5 ×​ 10–17
Two-photon absorption (cm GW–1) 0.25 Negligible 60
Modulator technology (maximum speed) Free-carrier plasma dispersion With graphene (30 GHz) QCSE-EAM
(30 GHz) With PZT (33 GHz) (55 GHz)
Detector Ge (50 GHz) N/A 40 GHz
Laser output power N/A N/A >​20 mW
Fibre-to-chip coupling loss (dB) 2 0.5 3
CMOS compatibility Excellent Good N/A
Optical amplification N/A N/A >​20 dB
EAM, electro-absorption modulator; PZT, lead zirconate titanate; QCSE, quantum-confined Stark effect. N/A, not applicable.

integrated replacement for a long length of optical fibre70. Despite achieve high resolution with very high stopband rejection of the
these progresses, full-circuit integration of the tapped-delay-line fil- order of 60 dB while using only 0.8 dB of SBS gain19 (Fig. 3d).
ter has never been attempted. Achieving such an advanced filter using low powers will be impor-
A fully integrated frequency-tunable MPF based on the optical tant for applications including radar and satellite communications.
filter response downconversion concept, on the other hand, was Another advantage of SBS is all-optical reconfigurability. By pump-
recently demonstrated on an indium phosphide platform21. All the ing the SBS medium not only with a single laser, but with multiple
components, including a laser source, a modulator, an optical filter lines with shaped amplitude and frequency spacing, a broadened
based on a ring-assisted Mach–Zehnder interferometer and a pho- tailorable response can be achieved and one can generate unique
todetector, were integrated on a single chip (Fig. 3b). A similar filter filters with flat-top, sharp-edge frequency response and tunable
demonstration was reported shortly after where a phase modulator, bandwidth75. These devices are promising for channel selections in
an optical filter based on a microdisk resonator and a photodetector spectrally crowded RF environments (Fig. 3e).
were all integrated in a silicon chip74.
Programmable signal processing. A major paradigm shift recently
High-resolution filtering with stimulated Brillouin scattering. occurred in the area of MWP signal processing. To date, most of
Although optical-based RF photonic filters can be tuned over a wide the reported integrated photonic microwave signal processors have
frequency range, the spectral resolution of most optical filters, typi- been implemented as application-specific photonic integrated cir-
cally in the gigahertz, is too coarse for processing RF signals where cuits (ASPICs), which are designed to optimally perform a par-
the separation between adjacent information channels can be down ticular MWP functionality5. This results in a lack of universality
to only a few tens of megahertz. Breaking this barrier, a number of and reconfigurability for multifunctional applications. Leveraging
integrated MWP filters that combine the strengths of photonic and on the strong push towards programmable photonics from related
electronic filters, namely tens of gigahertz frequency tuning with fields, including quantum photonics85–87, researchers have strived
megahertz spectral resolution and an ultra-high extinction, have towards programmable and multipurpose integrated MWP devices,
been reported19,75–77. This unique performance metric was achieved with two particular routes explored.
by harnessing SBS — a coherent interaction between optical waves The first approach relies on circuits based on traditional inter-
and high-frequency acoustic waves (hypersound)78. Stimulated ferometric and photonic waveguide structures designed for flexible
Brillouin scattering manifested in the generation of gain and loss programming of its key parameters. Implementation using unit cells
resonances with narrow linewidths of the order of 10 MHz. of Mach–Zehnder interferometers and ring resonators that can be
While SBS has been well studied in long lengths of optical fibres, activated individually, for example, was explored in the context of
it has only been observed recently in integrated waveguides79–83. filtering16,88, waveform generation7,89,90, reconfigurable delay lines15
Achieving a strong SBS response in integrated waveguides requires and frequency measurement91,92. More recently, a fully reconfigu-
material platforms with strong electrostriction and elasto-optic rable indium phosphide photonic integrated signal processor has
coefficients and large overlap between optical and acoustic modes. been demonstrated93 made of three active ring resonators and a
To date, the strongest on-chip SBS response was demonstrated in bypass waveguide as a processing unit cell (Fig. 4a). With this cir-
chalcogenide (As2S3) waveguides79,80, while achieving high gain in cuit, reconfigurable signal processing functions, including filter-
versatile materials such as silicon is still challenging. Critically, one ing, temporal integration, temporal differentiation and Hilbert
must combat acoustic phonon leakage from silicon to the silica sub- transformation, can be performed. Another implementation of this
strate using suspended structures81–83. An alternative approach is to approach includes a photonic signal processor using reconfigurable
explore heterogeneous integration where a material that supports silicon Bragg gratings94.
efficient generation of SBS, such as chalcogenides, is embedded in a The second approach looks at the possibility of making a generic
CMOS-compatible circuit84. signal processor from a mesh of uniform tunable building blocks
On-chip SBS leads to a number of advantages for integrated that can be programmed to support multiple functions20,95–98. This
MWP filters. The ultra-narrow linewidth allows high-resolution fil- concept is inspired by field-programmable gate arrays in electronics
tering unmatched by most on-chip devices. By combining on-chip and by software-defined networks in telecommunications. In two
SBS with a cancellation filter (Fig. 3c), one can simultaneously recent demonstrations20,98, the tunable building block of choice is a

84 Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics

NaTuRe PHOTOnics Review Article
a b
Ground (G) Device L = 20 mm
Signal (S) 2
RF in Ground (G) RF Vπ = 1.4 V
RF terminate 0

Electro-optic S21 (dB)

500 m –2
–3 dB

–4
c
Signal pad –6 dB
Ground pad –6

–8
Suspended bridge

–10
10 20 30 40
Frequency (GHz)

d
0
Island

–10

Transmission (dB)
Ground pad

1 µm –20

–30
e
–40
fc–170 fc fc+170
Frequency (GHz)

n-InP
H3 H1 f 20
H2
0
H4 –20
Output power (dBm)

–40 57 dB
–60 –69.0 dBm
P
–80 Noise floor at 1 GHz
InP
N N III–V –100
SiO2 QWs
Fundamental
Si SOI –120 Intermodulation
BOX
–140
–10 0 10 20 30 40
Input power (dBm)

Fig. 2 | Advanced optical modulator technologies for MWP. a, Integrated lithium niobate-on-insulator (LNOI) Mach–Zehnder optical modulator with
modulator length of the order of 20 mm (ref. 47). b, The measured electro-optic response of the LNOI modulator. The 20-mm-long device has a 3-dB
bandwidth of 40 GHz and an ultra-low RF half-wave voltage of 1.4 V (ref. 47). c, Colorized scanning electron microscopy image of an all-plasmonic Mach–
Zehnder modulator. The plasmonic interferometer is formed by the metallic island and the metallic contact pads53. d, The measured optical spectrum at
the output of a plasmonic modulator showing optical sidebands at 170 GHz (ref. 55). fc, optical carrier frequency. e, Schematic of an ultra-linear optical
modulator using a structure of a heterogeneous ring-assisted Mach–Zehnder interferometer (RAMZI) modulator in III–V-on-silicon technology59.
H1–H4 indicate the locations of the extra thermal phase tuners. BOX, buried oxide. f, The measured linearity and SFDR achieved using the ultra-linear
heterogeneous RAMZI modulator59. SOI, silicon-on-insulator. Figure adapted from: a,b, ref. 47, Springer Nature Limited; c, ref. 53, Springer Nature Limited;
d, ref. 55, OSA; e,f, ref. 59, OSA.

Mach–Zehnder interferometer composed of a tunable coupler and a delay line, and both notch and bandpass filters (this last functional-
four input–output waveguides that can be arranged into a 2D unit ity is shown in Fig. 4c). A waveguide mesh composed of 7 hexagonal
cell in square-, triangular-, or hexagonal-type meshes (Fig. 4d). The cells fabricated in a silicon platform has also been demonstrated98.
choice of mesh configuration will determine a number of figures Figure 4e shows a photograph of the overall structure, which con-
of merit including the spatial tuning resolution step, the number of sisted of 30 independent Mach–Zehnder interferometer devices
switching elements per unit area and the losses in the Mach–Zehnder and 60 thermo-optic heaters. The structure is capable of imple-
interferometer interconnections. A programmable optical chip con- menting over 100 different circuits and functionalities. Figure 4f
necting Mach–Zehnder interferometer devices in a square-shaped shows the optical response of the chip when configured as an
mesh network fabricated in low-loss silicon nitride technology has add-drop ring resonator.
been reported20 (Fig. 4b). It consisted of two square cells and was Several practical issues need attention to advance this processor
used to demonstrate functionalities including a Hilbert transformer, approach. These include the scaling of the waveguide mesh that is

Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics 85

Review Article NaTuRe PHOTOnics

a RF input RF output
b RF input RF output

Multiwavelength MZI and

source Pulse shaper Delay line Laser ring filters
Optical modulator Photodetector Optical modulator

Microheater Microring

SiN microresonator InP AWG and SOAs GaInP photonic

combs crystal waveguide
InP all-integrated filter

c Probe Pump d e

0 Filter off
Normalized RF transmission (dB) Unwanted
Filter on interference
VOA –10

RF power (10 dB div–1)

Optical –20
modulator Signal 47 dB
2 dB
–30 Δνmax 88 MHz
–40 Δνmin 32 MHz
20 MHz
–50
SBS Ge PD
waveguide –60
11.85 11.95 12.05 12.15 11.97 11.98 11.99 12.00 12.01
Frequency (GHz) Frequency (GHz)

Fig. 3 | Integrated microwave photonic filtering. a, Schematic of a tapped delay line and its integration formed by three key components. Microresonator
frequency combs have been used as a multiwavelength source71. Line-by-line shaping of each tap has been done using an integrated InP pulse shaper73.
The dispersive delay lines were miniaturized using GaInP photonic crystal waveguides70. Scale bar, 1 μ​m. AWG, arrayed waveguide grating; SOA,
semiconductor optical amplifier. b, The microwave photonic filter based on an optical filter has been integrated monolithically on the InP platform21.
c, Artist’s impression of an integrated filter based on SBS19. PD, photodetector; VOA, variable optical attenuator. d, High-resolution RF photonic notch
filtering based on a low-power chalcogenide SBS device. The resolution can be tailored from 32 MHz to 88 MHz (ref. 19). e, Filtering of unwanted signal using
the SBS filter in c and d. The unwanted interference 20 MHz away from the desired signal is effectively filtered without significant reduction of the desired
signal power19. Figure adapted from: a(image of microresonator comb), ref. 66, Springer Nature Limited; a(image of InP AWG and SOAs), ref. 73, OSA; a(image
of GaInP photonic crystal waveguide), ref. 70, Springer Nature Limited; b(image of InP all-integrated filter), ref. 21, Springer Nature Limited; c–e, ref. 19, OSA.

directly related to the footprint and loss characteristics of the cho- of a Mach–Zehnder modulator while increasing the input optical
sen material platforms, the tracking and stabilization of circuit power, minimizing the loss in the optical waveguide platform, and
parameters, the power consumption of the tuning elements, and using a high-power-handling photodetector composed of multiple
the operation robustness against fabrication imperfections. Recent photodiodes to ensure linearity. Table 2 compares the filter perfor-
works have started to address some of these challenges99,100. mance with recent results of integrated and fibre-based104 MWP fil-
ters benchmarked against a high-performance RF filter105. It is clear
Challenges that while MWP filters are competitive or even superior in various
To be deployed in actual RF systems, integrated MWP devices have figures of merit, the noise figure remains the bottleneck. This is
to achieve comparable performance to RF devices. The require- because for passive RF filters, the noise figure is equal to their inser-
ments in terms of RF characteristics (Box 1) include no loss of the tion loss whereas for MWP filters it will be considerably raised by
signal of interest (often reflected as zero or positive RF link gain optical noise sources such as relative intensity noise, shot noise and
on the decibel scale), a low noise figure below 10 dB and a high optical amplifier noise.
spurious-free dynamic range (SFDR) of more than 120 dB Hz2/3 Extrapolating high link performance to fully integrated systems
(ref. 101). Such performance has been achieved in analog fibre-optic will be one of the key challenges to address in integrated MWP. This
links for RF signal transport but no other RF photonic processing will require efficient and linearized on-chip modulators58,59, low-
functionality has been implemented. In contrast, the link perfor- loss waveguides29, high-power and low-noise on-chip lasers106,107,
mance of reported integrated MWP systems is either well below the and high-power-handling photodetectors108 assembled through
required target performance, or, more often, not even considered or heterogeneous integration schemes. For a more detailed discussion
reported. If not addressed, the link performance can severely limit on this topic see ref. 109.
the uptake of integrated MWP technologies. Another important challenge to solve for integrated MWP is the
Only recently, researchers have made progress in demonstrating power consumption associated with tuning integrated photonics
integrated MWP functions simultaneously with high performance. and the signal processing operation. In this regard, alternatives to
For example, a recent report of a partially integrated MWP filter thermo-optic tuning are needed. Candidates for this include piezo-
in silicon nitride showed a positive link gain, a low noise figure of electrically tuned devices51. The use of ultra-low-loss waveguides
15.6 dB and a high SFDR of 116 dB Hz2/3 (refs. 102,103) while achiev- and resonators in nonlinear materials is also critical to reduce the
ing multiple stopbands with 60 dB extinction. Such a breakthrough power threshold of certain nonlinear processes that can lead to
in performance was achieved by putting together a number of ultra-low-power frequency combs63–65, filters19 and synthesizers35,
known approaches for link optimization, including low biasing and enable efficient nonlinear integrated MWP signal processing110.

86 Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics

NaTuRe PHOTOnics Review Article
a Input
b c
SOA
TC
PM
SOA N

SOA
R1 0

Power transmission (dB)

W N
TC Cell 1
PM
E W
SOA

SOA
SOA

R2 S Cell 2
E –15
TC S
PM
SOA

SOA
R3
Bar (switch)
PM TC SOA –30
Output Coupler –0.5 0.0 0.5
Frequency deviation (× FSR)

d e f

Norm. optical transmission (dB)

0

–10

TBU
–20

1 2 1 2
1 3 –30
2
3 4 6 3 4
5 –40
0 10 20 30 40
Mesh interconnection-node Relative optical frequency (GHz)

Fig. 4 | Programmable and general-purpose MWP processors. a, Reconfigurable processor in indium phosphide. The device can function as an optical
differentiator, integrator and Hilbert transformer93. PM, phase modulator; TC, tunable coupler; R1–R3, ring resonators. b, A programmable optical chip
connecting Mach–Zehnder interferometer devices in a square-shaped mesh network fabricated in low-loss silicon nitride technology20. c, By thermo-optic
tuning, the square-shaped mesh network can be programmed to exhibit a square-shaped bandpass filter normally achieved using a lattice two-ring filter20. d,
Three general topologies of interconnected Mach–Zehnder interferometers, namely triangular, hexagonal and square meshes97. TBU, tunable basic unit.
e, Photograph of a reconfigurable signal processor based on the hexagonal mesh. The overall structure consisted of 30 independent Mach–Zehnder
interferometer devices and 60 thermo-optic heaters98. Scale bar, 2 mm. f, The processor can be programmed to exhibit more than 100 distinct optical responses
including an optical response from an add-drop ring resonator98. The three curves show central frequency tuning of the optical responses. Figure adapted from:
a, ref. 93, Springer Nature Limited; b,c, ref. 20, OSA; d, ref. 97, OSA; e,f, ref. 98, under a Creative Commons licence (https://creativecommons.org/licenses/by/4.0/).

Table 2 | Performance comparison of integrated and fibre-based MWP filters

Fibre comb- Indium phosphide As2S3 SBS19 Silicon nitride Indium phosphide RF resonator105
based104 pulse shaper73 ring103 MZI and rings21
Class Multi-tap Multi-tap Optical filter Optical filter Optical filter Electronic filter
Integration Not integrated Partial Partial Partial Full Full
Type Bandpass Bandpass Bandstop Bandstop Lowpass Bandpass
RF link gain (dB) 0 –3 –30 8 –20 –3.1
Stopband suppression 32 36 55 50 30 40
(dB)
Resolution (MHz) 770 130 32 150 5,500 30
Frequency tuning (GHz) 6 2 30 10 4 2.6
Noise figure (dB) 24 N/A >​30 15.6 >​30 3.1
SFDR (dB Hz2/3) N/A N/A N/A 116 81.4 137
MZI, Mach–Zehnder interferometer. N/A, not applicable.

Outlook and perspectives applications from sensing to quantum information science.

Adoption of the technological tools described above has not Architectures that manipulate phonons generated optically and
only equipped MWP with advanced functionalities, but has also transduced through RF fields hold the promise of enhanced signal
expanded the field considerably to allow many intersections with processing111 (Fig. 5a). For example, concepts such as a phononic–
other growing fields in photonics, potentially creating new concepts photonic emitter–receiver and phonon routing between two opto-
and paradigms. mechanical cavities have been used to demonstrate RF photonic
The manipulation of phonons and high-frequency sound waves bandpass filters with ultra-narrow sub-MHz linewidths112,113.
in integrated devices can effectively bridge classical RF photonics Adoption of versatile integrated reconfigurable processors can
processing and cavity optomechanics111–113 with potential broadened enable the possibility of multiple input/multiple output (MIMO)

Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics 87

Review Article NaTuRe PHOTOnics

b
a
RF in Cavity L Cavity R
Photonic
cavity κi Control laser
DUT
G = g0*√N 10 µm
Probe κ RF out 5
laser e Phononic
cavity ϕ
γi Optomechanical 17 kHz at −3 dB
RF Phase cavities
Propagating VNA −5
Photonic γe amp shifter
acoustic wave
waveguide

|SRL|2 (dB)
Phononic −15
waveguide 1,550 nm
Interdigitated APD
laser
transducer Polarization Phase −25
controller modulator
RF source

−35

–0.8 –0.4 0.0 0.4 0.8

∆f (MHz)
c Optical interference unit d
Sound byte Spatiotemporal spike encoding Word recognition

Target output
‘Laser’

Complex RF signal Laser pulse encoding RF fingerprinting

Base Base
station station

0 1 00 1 1 0 1 1
SU(4) core DMMC 60 µm

Fig. 5 | Opportunities for integrated MWP. a, High-frequency phonons for bridging RF and photonic domains. An optomechanical device in piezoelectric
materials can couple phonons generated from RF sources and phonons generated optically as depicted by the level diagram (left) and the measurement
set-up (right)111. κe, optical cavity coupling rate; κi, intrinsic optical cavity decay rate; γe, phononic cavity coupling rate; γi, intrinsic phononic cavity decay
rate; APD, avalanche photodetector; DUT, device under test; VNA, vector network analyser. b, An optomechanical structure consisting of two optical
microcavities connected by a phononic waveguide (upper) can act as a high-resolution filter with 17-kHz, 3-dB bandwidth (lower)113. c, Programmable
optical processors intersect integrated MWP with emerging fields including integrated quantum photonic and neuromorphic photonics. The figure
depicts a programmable nanophotonic processor in silicon photonics used for deep-learning applications. The chip is composed of 56 Mach–Zehnder
interferometers, 213 phase shifters and 112 directional couplers118. d, Illustration of neuromorphic photonic concepts implemented for RF fingerprinting
of complex and crowded RF environments for cognitive radio applications117. DMMC, diagonal matrix multiplication core. Figure adapted from: a, ref. 111,
Springer Nature Limited; b, ref. 113, Springer Nature Limited; c, ref. 118, Springer Nature Limited; d, © ref. 117, reproduced by permission of Taylor and Francis
Group LLC, a division of Informa plc.

MWP and open the door to parallel linear processing114 and space- Received: 31 May 2018; Accepted: 5 October 2018;
division multiplexing115,116. Furthermore, analog photonic principles Published online: 21 January 2019
combined with unitary N ×​ N transformations have been proposed
as a key technology for the implementation of spiking and reservoir
neuromorphic photonic systems, deep learning (Fig. 5b) and brain- References
inspired processing117–119. These concepts can be implemented in 1. Seeds, A. J. & Williams, K. J. Microwave photonics. J. Lightw. Technol. 24,
4628–4641 (2006).
the wideband fingerprinting of crowded wireless environments for 2. Minasian, R. A. Photonic signal processing of microwave signals.
cognitive radio applications119 (Fig. 5c,d). Interestingly, integrated IEEE Trans. Microw. Theory Tech. 54, 832–846 (2006).
MWP systems based on cascaded ring modulation by RF signals 3. Capmany, J. & Novak, D. Microwave photonics combines two worlds.
have also been proposed for the implementation of a new genera- Nat. Photon. 1, 319–330 (2007).
4. Yao, J. Microwave photonics. J. Lightw. Technol. 27, 314–335 (2009).
tion of processing systems based on topological photonics120.
5. Marpaung, D. et al. Integrated microwave photonics. Laser Photon. Rev. 7,
The concept of parity–time symmetry in laser optics to achieve 506–538 (2013).
single-mode lasing121–123 has recently been employed in MWP sys- 6. Yao, J. P., Zeng, F. & Wang, Q. Photonic generation of ultra-wideband
tems for filterless single-frequency microwave generation. This signals. J. Lightw. Technol. 25, 3219–3235 (2007).
concept overcomes the long existing problems of mode competition 7. Khan, M. et al. Ultrabroad-bandwidth arbitrary radiofrequency waveform
generation with a silicon photonic chip-based spectral shaper. Nat. Photon.
and mode selection that have severely limited the development of 4, 117–122 (2009).
OEOs124,125, particularly in integrated devices where an ultra-nar- 8. Lim, C. et al. Fiber-wireless networks and subsystem technologies.
rowband optical filter is difficult to implement. J. Lightw. Technol. 28, 390–405 (2010).
These concepts in integrated MWP driven by the new under- 9. Capmany, J., Ortega, B. & Pastor, D. A tutorial on microwave photonic
standing in device physics and the tremendous growth of integrated filters. J. Lightw. Technol. 24, 201–229 (2006).
10. Supradeepa, V. R. et al. Comb-based radiofrequency photonic filters with
photonics represent a new phase in the field beyond telecom-related rapid tunability and high selectivity. Nat. Photon. 6, 186–194 (2012).
applications. Such explorations will sustain the growth of these 11. Ghelfi, P. et al. A fully photonics-based coherent radar system. Nature 507,
fields and bring further exciting opportunities. 341–345 (2014).

88 Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics

NaTuRe PHOTOnics Review Article
12. Hecht, J. The bandwidth bottleneck that is throttling the Internet. Nature 47. Wang, C. et al. Integrated lithium niobate electro-optic modulators
536, 139–142 (2016). operating at CMOS-compatible voltages. Nature 562, 101–104 (2018).
13. Iezekiel, S., Burla, M., Klamkin, J., Marpaung, D. & Capmany, J. RF 48. Rao, A. & Fathpour, S. Compact lithium niobate electrooptic modulators.
engineering meets optoelectronics. IEEE Microw. Mag. 16, 18–45 (2015). IEEE J. Sel. Top. Quantum Electron. 24, 3400114 (2018).
14. Zhuang, L. et al. Low-loss high-index-contrast Si3N4/SiO2 optical 49. Soref, R. A. & Bennett, B. R. Electrooptical effects in silicon. IEEE J.
waveguides for optical delay lines in microwave photonics signal processing. Quantum Electron. 23, 123–129 (1987).
Opt. Express 19, 23162–23170 (2011). 50. Trajkovic, M. et al. 55 GHz EAM bandwidth and beyond in InP active-
15. Burla, M. et al. On-chip CMOS compatible reconfigurable optical delay passive photonic integration platform. In Proc. CLEO Paper JTh5A.8
line with separate carrier tuning for microwave photonic signal processing. (OSA, 2018).
Opt. Express 19, 21475–21484 (2011). 51. Alexander, K. et al. Nanophotonic Pockels modulators on a silicon nitride
16. Norberg, E. J., Guzzon, R. S., Parker, J. S., Johansson, L. A. & Coldren, L. A. platform. Nat. Commun. 9, 3444 (2018).
Programmable photonic microwave filters monolithically integrated in InP/ 52. Koos, C. et al. Silicon-organic hybrid (SOH) and plasmonic-organic hybrid
InGaAsP. J. Lightw. Technol. 29, 1611–1619 (2011). (POH) integration. J. Lightw. Technol. 34, 256–268 (2016).
17. Kippenberg, T. J., Holzwarth, R. & Diddams, S. A. Microresonator-based 53. Haffner, C. et al. All-plasmonic Mach–Zehnder modulator enabling optical
optical frequency combs. Science 332, 555–559 (2011). high-speed communication at the microscale. Nat. Photon. 9, 525–528 (2015).
18. Haffner, C. et al. Plasmonic organic hybrid modulators—scaling highest 54. Ayata, M. et al. High-speed plasmonic modulator in a single metal layer.
speed photonics to the microscale. Proc. IEEE 104, 2362–2379 (2016). Science 358, 630–632 (2017).
19. Marpaung, D. et al. Low power, chip-based stimulated Brillouin scattering 55. Hoessbacher, C. et al. Plasmonic modulator with >​170 GHz bandwidth
microwave photonic filter with ultrahigh selectivity. Optica 2, 76–83 (2015). demonstrated at 100 GBd NRZ. Opt. Express 25, 1762–1768 (2017).
20. Zhuang, L., Roeloffzen, C. G. H., Hoekman, M., Boller, K. J. & Lowery, A. J. 56. Salamin, Y. et al. Direct conversion of free space millimeter waves
Programmable photonic signal processor chip for radiofrequency to optical domain by plasmonic modulator antenna. Nano Lett. 15,
applications. Optica 2, 854–859 (2015). 8342–8346 (2015).
21. Fandiño, J. S., Muñoz, P., Doménech, D. & Capmany, J. A monolithic 57. Chen, L., Chen, J., Nagy, J. & Reano, R. M. Highly linear ring modulator from
integrated photonic microwave filter. Nat. Photon. 11, 124–129 (2016). hybrid silicon and lithium niobate. Opt. Express 23, 13255–13264 (2015).
22. Smit, M. et al. An introduction to InP-based generic integration technology. 58. Zhang, C., Morton, P. A., Khurgin, J. B., Peters, J. D. & Bowers, J. E. Highly
Semicond. Sci. Technol. 29, 083001 (2014). linear heterogeneous-integrated Mach–Zehnder interferometer modulators
23. Thomson, D. et al. Roadmap on silicon photonics. J. Opt. 18, 073003 (2016). on Si. Opt. Express 24, 19040–19047 (2016).
24. Zhang, W. & Yao, J. Silicon-based integrated microwave photonics. 59. Zhang, C., Morton, P. A., Khurgin, J. B., Peters, J. D. & Bowers, J. E.
IEEE J. Quantum Electron. 52, 0600412 (2016). Ultralinear heterogeneously integrated ring assisted Mach–Zehnder
25. Leuthold, J., Koos, C. & Freude, W. Nonlinear silicon photonics. interferometer modulator on silicon. Optica 3, 1483–1488 (2016).
Nat. Photon. 4, 535–544 (2010). 60. Maleki, L. Sources: the optoelectronic oscillator. Nat. Photon. 5, 728–730 (2011).
26. Roeloffzen, C. H. G. et al. Silicon nitride microwave photonic circuits. 61. Tang, J. et al. Integrated optoelectronic oscillator. Opt. Express 26,
Opt. Express 21, 22937–22961 (2013). 12257–12265 (2018).
27. Moss, D. J., Morandotti, R., Gaeta, A. L. & Lipson, M. New CMOS- 62. Zhang, W. & Yao, J. Silicon photonic integrated optoelectronic oscillator
compatible platforms based on silicon nitride and Hydex for nonlinear for frequency-tunable microwave generation. J. Lightw. Technol. 36,
optics. Nat. Photon. 7, 597–607 (2013). 4655–4663 (2018).
28. Chang, M. P., Blow, E. C., Lu, M. Z., Sun, J. J. & Prucnal, P. R. Integrated 63. Yang, K. Y. et al. Bridging ultrahigh-Q devices and photonic circuits.
microwave photonic circuit for self-interference cancellation. IEEE Trans. Nat. Photon. 12, 297–302 (2018).
Microw. Theory Tech. 65, 4493–4501 (2017). 64. Ji, X. et al. Ultra-low-loss on-chip resonators with sub-milliwatt parametric
29. Roeloffzen, C. G. H. et al. Low-loss Si3N4 TriPleX optical waveguides: oscillation threshold. Optica 4, 619–624 (2017).
technology and applications overview. IEEE J. Sel. Top. Quantum Electron. 65. Xuan, Y. et al. High-Q silicon nitride microresonators exhibiting low-power
24, 4400321 (2018). frequency comb initiation. Optica 3, 1171–1180 (2016).
30. Heck, M. J. R. et al. Hybrid silicon photonic integrated circuit technology. 66. Xue, X. et al. Mode-locked dark pulse Kerr combs in normal-dispersion
IEEE J. Quantum Electron. 19, 6100117 (2013). microresonators. Nat. Photon. 9, 594–600 (2015).
31. Komljenovic, T. et al. Heterogeneous silicon photonic integrated circuits. 67. Marin-Palomo, P. et al. Microresonator-based solitons for massively parallel
J. Lightw. Technol. 34, 20–35 (2016). coherent optical communications. Nature 546, 274–279 (2017).
32. Hiraki, T. et al. Heterogeneously integrated III–V/Si MOS capacitor 68. Liang, W. et al. High spectral purity Kerr frequency comb radio frequency
Mach–Zehnder modulator. Nat. Photon. 11, 482–485 (2017). photonic oscillator. Nat. Commun. 6, 7957 (2015).
33. Han, J. H. et al. Efficient low-loss InGaAsP/Si hybrid MOS optical 69. Li, J., Lee, H. & Vahala, K. J. Microwave synthesizer using an on-chip
modulator. Nat. Photon. 11, 486–490 (2017). Brillouin oscillator. Nat. Commun. 4, 2097 (2013).
34. Hulme, J. et al. Fully integrated microwave frequency synthesizer on 70. Sancho, J. et al. Integrable microwave filter based on a photonic crystal
heterogeneous silicon-III/V. Opt. Express 25, 2422–2431 (2017). delay line. Nat. Commun. 3, 1075 (2012).
35. Spencer, D. T. et al. An integrated-photonics optical-frequency synthesizer. 71. Xue, X. et al. Programmable single-bandpass photonic RF filter based on
Nature 557, 81–85 (2018). Kerr comb from a microring. J. Lightw. Technol. 32, 3557–3565 (2014).
36. Sun, C. et al. Single-chip microprocessor that communicates directly using 72. Wu, J. et al. RF photonics: an optical microcombs’ perspective. IEEE J. Sel.
light. Nature 528, 534–538 (2015). Top. Quantum Electron. 24, 6101020 (2018).
37. Eggleton, B. J., Luther-Davies, B. & Richardson, K. Chalcogenide photonics. 73. Metcalf, A. J. et al. Integrated line-by-line optical pulse shaper for
Nat. Photon. 5, 141–148 (2011). high-fidelity and rapidly reconfigurable RF-filtering. Opt. Express 24,
38. Belt, M., Davenport, M., Bowers, J. & Blumenthal, D. Ultra-low-loss 23925–23940 (2016).
Ta2O5-core/SiO2-clad planar waveguides on Si substrates. Optica 4, 74. Zhang, W. & Yao, J. On-chip silicon photonic integrated frequency-tunable
532–536 (2017). bandpass microwave photonic filter. Opt. Lett. 43, 3622–3625 (2018).
39. Xiong, C., Pernice, W. H. P. & Tang, H. X. Low-loss silicon integrated 75. Choudhary, A. et al. Tailoring of the Brillouin gain for on-chip widely
aluminum nitride photonic circuits and their use for electro-optic signal tunable and reconfigurable broadband microwave photonic filters. Opt. Lett.
processing. Nano Lett. 12, 3562–3568 (2012). 41, 436–439 (2016).
40. Zhang, M., Wang, C., Cheng, R., Shams-Ansari, A. & Lončar, M. 76. Aryanfar, I. et al. Signal interference RF photonic bandstop filter.
Monolithic ultra-high-Q lithium niobate microring resonator. Optica 4, Opt. Express 24, 14995–15004 (2016).
1536–1537 (2017). 77. Morrison, B. et al. Tunable microwave photonic notch filter using on-chip
41. Phare, C. T., Daniel Lee, Y.-H., Cardenas, J. & Lipson, M. Graphene electro-optic stimulated Brillouin scattering. Opt. Commun. 313, 85–89 (2014).
modulator with 30 GHz bandwidth. Nat. Photon. 9, 511–514 (2015). 78. Eggleton, B. J., Poulton, C. G. & Pant, R. Inducing and harnessing
42. Capmany, J., Doménech, D. & Muñoz, P. Graphene integrated microwave stimulated Brillouin scattering in photonic integrated circuits. Adv. Opt.
photonics. J. Lightw. Technol. 32, 3785–3796 (2014). Photon. 5, 536–587 (2013).
43. Li, G. L. & Yu, P. K. L. Optical intensity modulators for digital and analog 79. Pant, R. et al. On-chip stimulated Brillouin scattering. Opt. Express 19,
applications. J. Lightw. Technol. 21, 2010–2030 (2003). 8285–8290 (2011).
44. Bull, J. D. et al. 40 GHz electro-optic polarization modulator for fiber optic 80. Choudhary, A. et al. Advanced integrated microwave signal processing with
communications systems. Proc. SPIE 5577, 133–143 (2004). giant on-chip Brillouin gain. J. Lightw. Technol. 35, 846–854 (2017).
45. Wooten, E. L. et al. A review of lithium niobate modulators for fiber-optic 81. Shin, H. et al. Tailorable stimulated Brillouin scattering in nanoscale silicon
communications systems. IEEE J. Sel. Top. Quantum Electron. 6, waveguides. Nat. Commun. 4, 1944 (2013).
69–82 (2000). 82. Laer, R. V., Kuyken, B., Thourhout, D. V. & Baets, R. Interaction between
46. Wang, C., Zhang, M., Stern, B., Lipson, M. & Loncar, M. Nanophotonic light and highly confined hypersound in a silicon photonic nanowire.
lithium niobate electro-optic modulators. Opt. Express 26, 1547–1555 (2018). Nat. Photon. 9, 199–203 (2015).

Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics 89

Review Article NaTuRe PHOTOnics
83. Kittlaus, E. A., Shin, H. & Rakich, P. T. Large Brillouin amplification in 108. Beling, A., Xie, X. & Campbell, J. C. High-power, high-linearity
silicon. Nat. Photon. 10, 463–467 (2016). photodiodes. Optica 3, 328–338 (2016).
84. Morrison, B. et al. Compact Brillouin devices through hybrid integration on 109. Liu, Y., Marpaung, D., Choudhary, A., Hotten, J. & Eggleton, B. J. Link
silicon. Optica 4, 847–854 (2017). performance optimization of chip-based Si3N4 microwave photonic filters.
85. Carolan, J. et al. Universal linear optics. Science 349, 711–716 (2015). J. Lightw. Technol. 36, 4361–4370 (2018).
86. Harris, N. C. et al. Quantum transport simulations in a programmable 110. Marpaung, D., Pagani, M., Morrison, B. & Eggleton, B. J. Nonlinear
nanophotonic processor. Nat. Photon. 11, 447–452 (2017). integrated microwave photonics. J. Lightw. Technol. 32, 3421–3427 (2014).
87. Wang, J. et al. Multidimensional quantum entanglement with large-scale 111. Balram, K. C., Davanço, M. I., Song, J. D. & Srinivasan, K. Coherent
integrated optics. Science 360, 285–291 (2018). coupling between radiofrequency, optical and acoustic waves in piezo-
88. Ibrahim, S. et al. Demonstration of a fast-reconfigurable silicon CMOS optomechanical circuits. Nat. Photon. 10, 346–352 (2016).
optical lattice filter. Opt. Express 19, 13245–13256 (2011). 112. Kittlaus, E. A. RF-photonic filters via on-chip photonic–phononic
89. Marpaung, D., Chevalier, L., Burla, M. & Roeloffzen, C. Impulse radio emit–receive operations. J. Lightw. Technol. 36, 2803–2809 (2018).
ultrawideband pulse shaper based on a programmable photonic chip 113. Fang, K., Matheny, M. H., Luan, X. & Painter, O. Optical transduction
frequency discriminator. Opt. Express 19, 24838–24848 (2011). and routing of microwave phonons in cavity-optomechanical circuits.
90. Wang, J. et al. Reconfigurable radio-frequency arbitrary waveforms Nat. Photon. 10, 489–496 (2016).
synthesized in a silicon chip. Nat. Commun. 6, 5957 (2015). 114. Perez, D. et al. Silicon photonics rectangular universal interferometer.
91. Marpaung, D. On-chip photonic-assisted instantaneous microwave Laser Photon. Rev. 11, 1700219 (2017).
frequency measurement system. IEEE Photon. Technol. Lett. 25, 115. Wen, H. et al. Few-mode fibre-optic microwave photonic links. Light Sci. Appl.
837–840 (2013). 6, e17021 (2017).
92. Fandiño, J. S. & Muñoz, P. Photonics-based microwave frequency 116. Annoni, A. et al. Unscrambling light—automatically undoing strong mixing
measurement using a double-sideband suppressed-carrier modulation between modes. Light Sci. Appl. 6, e17110 (2017).
and an InP integrated ring-assisted Mach–Zehnder interferometer filter. 117. Prucnal, P. R. & Shastri, B. J. Neuromorphic Photonics (CRC Press,
Opt. Lett. 38, 4316–4319 (2013). Boca Raton, 2017).
93. Liu, W. et al. A fully reconfigurable photonic integrated signal processor. 118. Shen, Y. C. et al. Deep learning with coherent nanophotonic circuits.
Nat. Photon. 10, 190–195 (2016). Nat. Photon. 11, 441–446 (2017).
94. Zhang, W. & Yao, J. A fully reconfigurable waveguide Bragg grating for 119. Cheng, Z., Rios, C., Pernice, W. H., Wright, C. D. & Bhaskaran, H. On-chip
programmable photonic signal processing. Nat. Commun. 9, 1396 (2018). photonic synapse. Sci. Adv. 3, e1700160 (2017).
95. Pérez, D., Gasulla, I. & Capmany, J. Software-defined reconfigurable 120. Ozawa, T., Price, H. M., Goldman, N., Zilberberg, O. & Carusotto, I.
microwave photonics processor. Opt. Express 23, 14640–14654 (2015). Synthetic dimensions in integrated photonics: from optical isolation to
96. Capmany, J., Gasulla, I. & Pérez, D. Microwave photonics: the four-dimensional quantum Hall physics. Phys. Rev. A 93, 043827 (2016).
programmable processor. Nat. Photon. 10, 6–8 (2016). 121. Feng, L., Wong, Z. J., Ma, R. M., Wang, Y. & Zhang, X. Single-mode laser
97. Pérez, D., Gasulla, I., Capmany, J. & Soref, R. A. Reconfigurable lattice mesh by parity-time symmetry breaking. Science 346, 972–975 (2014).
designs for programmable photonic processors. Opt. Express 24, 122. Hodaei, H., Miri, M. A., Heinrich, M., Christodoulides, D. N. &
12093–12106 (2016). Khajavikhan, M. Parity-time–symmetric microring lasers. Science 346,
98. Pérez, D. et al. Multipurpose silicon photonics signal processor core. 975–978 (2014).
Nat. Commun. 8, 636 (2017). 123. Liu, W. et al. An integrated parity-time symmetric wavelength-tunable
99. Miller, D. A. B. Perfect optics with imperfect components. Optica 2, single-mode microring laser. Nat. Commun. 8, 15389 (2017).
747–750 (2015). 124. Zhang, J. & Yao, J. Parity-time symmetric optoelectronic oscillator. Sci. Adv.
100. Grillanda, S. et al. Non-invasive monitoring and control in silicon 4, eaar6782 (2018).
photonics using CMOS integrated electronics. Optica 1, 129–136 (2014). 125. Liu, Y. et al. Observation of parity-time symmetry in microwave photonics.
101. Urick, V. J., McKinney, J. D. & Williams, K. J. Fundamentals of Microwave Light Sci. Appl. 7, 38 (2018).
Photonics (John Wiley & Sons, Hoboken, NJ, 2015).
102. Liu, Y., Marpaung, D., Choudhary, A. & Eggleton, B. J. Lossless and Acknowledgements
high-resolution RF photonic notch filter. Opt. Lett. 41, 5306–5309 (2016). D.M. wishes to acknowledge funding from NWO-TTW Vidi 15702. The work of J.Y. is
103. Liu, Y., Hotten, J., Choudhary, A., Eggleton, B. J. & Marpaung, D. supported by the Natural Sciences and Engineering Research Council of Canada. J.C.
All-optimized integrated RF photonic notch filter. Opt. Lett. 42, wishes to acknowledge funding from ERC ADG-2016-741415 UMWP-CHIP, GVA
4631–4634 (2017). PROMETEO 2017/103 and COST CA16220 EUIMWP.
104. Kim, H. J., Leaird, D. E., Metcalf, A. J. & Weiner, A. M. Comb-based RF
photonic filters based on interferometric configuration and balanced
detection. J. Lightw. Technol. 32, 3478–3488 (2014). Competing interests
105. Joshi, H., Sigmarsson, H., Moon, S., Peroulis, D. & Chappell, W. J. The authors declare no competing interests.
High-Q fully reconfigurable tunable bandpass filters. IEEE Trans. Microw.
Theory Tech. 57, 3525–3533 (2009).
106. Fan, Y. et al. 290 Hz intrinsic linewidth from an integrated optical chip Additional information
based widely tunable InP-Si3N4 hybrid laser. In Proc. CLEO Paper JTh5C.9 Reprints and permissions information is available at www.nature.com/reprints.
(OSA, 2017). Correspondence should be addressed to D.M.
107. Morton, P., Morton, M. & Morton, S. Ultra low phase noise, high power,
hybrid lasers for RF mixing and optical sensing applications. In Proc. IEEE Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
Avionics and Vehicle Fiber-Optics Photon. Conf. (AVFOP) Paper TuB.1 published maps and institutional affiliations.
(IEEE, 2017). © Springer Nature Limited 2019

90 Nature Photonics | VOL 13 | FEBRUARY 2019 | 80–90 | www.nature.com/naturephotonics