The vibrational motion of atoms in solids is ordinarily expected to dissipate energy from electronic excitations. Strong coupling of vibrational motion to electrons, however, can significantly transform the nature of accessible excited states, dramatically enriching light–matter interactions. In soft matter, such as photosynthetic light-harvesting complexes, the interplay between atomic motion and exciton dynamics enhances electronic energy transfer¹⁻⁴ in spite of the fluctuating physical environment. In crystals, the presence of strong interactions between electronic excitations and phonons,^5,6 which are the elementary excitations of the atomic lattice, enables the trapping of excitations,⁷⁻⁹ allows mechanical control of electron transport,^10,11 and drives the formation of exotic exciton–phonon quasiparticles and excitonic complexes.^12,13

A particularly striking manifestation of strong exciton–phonon coupling is the periodic vibronic structure that appears in molecular absorption spectra.¹⁴ The well-separated and multiple peaks in these systems correspond to distinct vibronic states; resonant excitation therefore enables us to directly address each individual state. Easy access to both optical and electronic control in transition-metal dichalcogenides (TMDs) offers a unique opportunity to electrically control vibronic states in a semiconductor device. Nevertheless, the voltage-tunable vibronic structure of periodic and well-separated peaks has not been realized in semiconductor TMD heterostructure devices.

Here we report the emergence of multiple periodic and well-separated photocurrent peaks when individual van der Waals (vdW) layers are stacked to form atomically thin heterostructure photodiodes. In particular, we find that the interlayer photocurrent exhibits a rich structure with numerous photocurrent sidebands as a function of incident photon energy E_PH. Strikingly, photocurrent sidebands are manifested only for values of E_PH close to the interlayer exciton energy; they are absent for the intralayer excitons. These resonances occur periodically with an energy spacing of approximately 30 meV and are observed for two distinct interlayer excitons within the same heterostructure. This energy spacing corresponds to the frequency of a group of prominent phonon modes observed in the Raman spectrum of the heterostructure. Importantly, we demonstrate that the manifold of photocurrent peaks can be controlled by the source-drain bias voltage.

As we discuss below, this multipeaked structure of the exciton resonance can be attributed to the appearance of a manifold of well-formed vibronic coupled exciton–phonon states (mirroring those typically found in molecular systems). We note that this vibronic structure of interlayer excitons in vdW heterostructures has so far been obscured from purely optical probes such as photoluminescence (PL);¹⁵ see also the discussion below. In contrast, the multicomponent photocurrent spectroscopy that we employed provides a means to directly address and control the rich vibronic structure of interlayer excitons that results from strong electron–phonon coupling in stacked vdW materials.

Experimentally, we studied vdW heterostructures composed of bilayer tungsten diselenide (WSe₂) stacked on top of monolayer molybdenum diselenide (MoSe₂), as shown in Figure ​Figure11. 2L-WSe₂/MoSe₂ serves as an important benchmark for photocurrent spectroscopy because it exhibits two distinct interlayer exciton resonances, one which has been observed in many PL studies of WSe₂/MoSe₂ (K → K) and another that has not been previously observed (Γ → K), as discussed in detail below. Owing to this newly observed interlayer exciton and the type II band alignment (see detailed band structure calculations in ref (25)), 2L-WSe₂/MoSe₂ also fulfills the technological need for semiconductor heterostructures with near-infrared band gaps, similar to silicon. Indeed, this work establishes it as an excellent material system with a clearly evident photocurrent near 1 eV.

Open in a separate window

Figure 1

Stack engineering and characterization of encapsulated WSe₂-MoSe₂ heterojunction photodiode devices. (a) Schematic of the multistage inverted fabrication process. (b) Schematic of the hexagonal boron nitride (hBN)-encapsulated heterostructure with a multilayer graphene back gate and source-drain electrodes. (c) Electronic energy band diagram at V_SD = 0 V, showing the conduction bands (CB), valence bands (VB), and chemical potential μ_c in equilibrium (horizontal dashed gray line). (d) Raman spectroscopy (λ = 532 nm) of MoSe₂ (green), bilayer WSe₂ (blue), and the heterostructure (black). (Inset) log-scale Raman intensity versus Raman shift near 30 meV. (e) Photoluminescence (λ = 532 nm) from MoSe₂ (green), bilayer WSe₂ (blue), and the heterostructure (black). (f) Interlayer dark current vs V_SD (blue points); the device displays ordinary vdW heterostructure p–n junction behavior²⁴⁻²⁷ with a fit (solid red line) to the diode using ideality factor α = 1.82, which relates the applied voltage V_SD to the potential energy difference established across the WSe₂–MoSe₂ interface.^17,29 (Inset) Interlayer photocurrent vs V_SD; E_PH = 0.99 eV. (See the Supporting Information Section S4(16) for details.)

Using an inverted fabrication process (Figure ​Figure11a, Supporting Information Section S1(16)), we first patterned multilayer graphene gate electrodes and conventional metal source and drain contacts. The WSe₂–MoSe₂ heterostructures, assembled and characterized independently, were laminated onto the prefabricated device patterns. Combining the vdW heterostructure in this way enabled complete protective encapsulation of the WSe₂-MoSe₂ interface region between hexagonal boron nitride layers (schematic in Figure ​Figure11b). We fabricated and studied three devices, each with a random orientation between constituent layers. The results described below were consistent across all devices.

Devices were characterized using Raman, photoluminescence (PL), and photocurrent (PC) spectroscopy. As shown in Figure ​Figure11d, the Raman spectrum of the heterostructure (black line) exhibits peaks that are also evident in the individual MoSe₂ (green line) and WSe₂ (blue line) layers. When comparing Raman peaks between the heterostructure and its constituent layers, we observed a negligible shift of the peak positions as a function of energy. Importantly, we observed pronounced Raman peaks at the WSe₂-MoSe₂ interface (black line), at energies near 30 meV (29.9, 31.0, and 31.9 meV, Figure ​Figure11d inset). In our previous work,¹⁷ Raman peaks near 30 meV were used to identify the layer thickness, confirming that the heterostructure was composed of bilayer WSe₂ and monolayer MoSe₂. In MoSe₂, the A_1g peak at 241 cm^–1 (29.8 meV), E_2g¹ peak at 288 cm^–1, and lack of B_2g¹ peak between 350 and 360 cm^–1 are characteristic of monolayer thickness. The Raman spectrum of WSe₂ exhibits an A_1g mode at 250 cm^–1 (30.9 meV), an E_2g mode at 260 cm^–1 (32.2 meV), and a B_2g¹ mode at 309 cm^–1, indicating bilayer thickness.

Figure ​Figure11e compares the PL vs photon energy E_PH from MoSe₂ (green line), WSe₂ (blue line), and the heterostructure (black line). The stacked vdW heterostructure exhibits PL spectral features similar to those of the individual layers.¹⁸⁻²³ In addition to the observed excitonic resonances characteristic of the individual layers, low-energy bound interlayer e–h pairs (excitons) are known to form between the valence band of bilayer WSe₂ and the conduction band of MoSe₂ (Figure ​Figure11c). Direct access to interlayer excitons has proven to be challenging. For example, although signatures of the lowest-lying interlayer exciton near E_I ≈ 1.0 eV (arising from carriers in the momentum mismatched Γ- and K-valleys) can be indirectly inferred,¹⁷ direct PL signatures are washed out as a result of very small oscillator strengths.

To overcome the small oscillator strength that prevents strong and direct PL signatures for these interlayer excitons, here we instead employed measurements of the interlayer photocurrent I_PC with the laser focused on the WSe₂-MoSe₂ heterostructure. In the photocurrent setup, interlayer excitons that are generated by infrared laser illumination of the vdW heterostructure can be subsequently dissociated. Once dissociated, separated electrons and holes transit the device, resulting in a photocurrent that increases in reverse bias (Figure ​Figure11f inset; the main panel shows dark current characterization of the device, displaying typical diode behavior).²³⁻²⁷

Figure ​Figure22 examines the detailed dependence of the interlayer photocurrent on E_PH and V_G. Using sensitive current amplification and sweeping E_PH, we find that I_PC, which is the difference in current measured with the light on and light off, is particularly pronounced in two frequency windows occurring near E_PH ≈ 1.3 and 0.9 eV. Photoexcitation energies in the vicinity of these two hot spots correspond to interlayer excitons hosted in the WSe₂-MoSe₂ heterostructure, namely, the K → K (~1.3 eV) and Γ → K (~0.9 eV) interlayer excitons.^17,18,28

Open in a separate window

Figure 2

Multiple and periodic sidebands in the photocurrent spectra of van der Waals p–n heterojunction devices. (a) Interlayer photocurrent I_PC vs E_PH and V_G near the direct (K → K) interlayer exciton transition for device 1, T = 20 K. (b) Line cut of the photocurrent spectrum at V_G = −3.5 V. For full data, see Supporting Information Section S4.2.¹⁶ (c) 2D Fourier transform of the photocurrent vs E_PH second derivative. (d, left) Schematic of the interlayer exciton in the heterostructure; (right) schematic of the band structure with the momentum direct (K → K) and indirect (Γ → K) interlayer excitons labeled. (e) I_PC vs E_PH and V_G of the heterostructure near the Γ → K exciton at T = 20 K. (f) Line cut of the photocurrent spectrum. (g) 2D Fourier transform of the photocurrent vs E_PH second derivative.

Strikingly, when E_PH was tuned between 1.24 and 1.40 eV (Figure ​Figure22a), we observed a periodic sequence of photocurrent peaks that occur in a narrow range of gate voltages (near V_G = −3.5 V). Although the strongest peak occurs at E_PH = 1.32 eV, it is only slightly stronger than several equally spaced maxima at higher and lower E_PH (Figure ​Figure22b). We observed an average peak separation of 30 meV. To clarify this periodic modulation, we calculated the Fourier transform of the second derivative of the photocurrent data, where we find a clear periodic component at 1/Δε = (30 meV)⁻¹ (marked by the blue dashed line in Figure ​Figure22c). Interestingly, near the interlayer excitation from the WSe₂K-valley to the MoSe₂K-valley (Figure ​Figure22d), the discrete energy difference Δε = 30 meV between photocurrent peaks closely corresponds to the frequency at which a strong Raman signal is observed in the heterojunction, Ω ≈ 30 meV. Here, is Planck’s constant and Ω is the phonon frequency.

In the same fashion as above, optical excitation of the lowest-lying (Γ → K) interlayer exciton also results in a series of approximately equally spaced discrete sidebands with energy spacing Ω ≈ 30 meV. This behavior is highlighted in Figure ​Figure22e, which shows I_PC vs E_PH and V_G at infrared photon energies. In the range E_PH = 0.88–1.03 eV, we observed a set of evenly spaced photocurrent maxima, which increase in amplitude as E_PH increases. The lowest-energy peak occurs at E_PH = 0.90 eV, and line traces of I_PC vs E_PH (Figure ​Figure22f) show regularly spaced peaks that are superimposed on a photocurrent background that increases with E_PH. Taking the Fourier transform of the second derivative of the interlayer photocurrent data (Figure ​Figure22g) reveals two periodic components: a dominant component at Δε = 30 meV and a weaker component at 22 meV. In sharp contrast to this periodic structure in Figure ​Figure22, photocurrent corresponding to the excitation of intralayer excitons does not produce such sidebands (Supporting Information Section S4.1¹⁶).

The appearance of a periodic array of photocurrent sidebands measured at fixed source-drain bias V_SD can be understood through the strong coupling of phonons and interlayer excitons, with each of the sideband peaks identified with a coupled (interlayer) exciton–phonon state of frequency ω_vib,n (V_SD), where n is an index that labels the vibronic states/peaks. In these vibronic states, electronic excitations are intertwined with lattice displacements and can be understood as Franck–Condon-type progressions. (See the detailed theoretical description in Supporting Information Section S5.¹⁶) We note, parenthetically, that the 30 meV periodicity in photocurrent spectroscopy (Figure ​Figure22) closely matches a window of narrow phonon branches expected for the interlayer heterostructure. (See Supporting Information Section S3(16) for discussion of phononic origins as well as phonon dispersion calculations in the heterostructure.) Remarkably, each of these vibronic states, ω_vib,n, is addressable by tuning E_PH. This contrasts with phonon-broadened peaks wherein the action of exciton–phonon interaction broadens the exciton transition without yielding individual well-defined exciton–phonon states.⁶

Although the E_PH spectroscopy in Figure ​Figure22 revealed the vibronic structure (at fixed V_SD), we can leverage the out-of-plane electric field of the interlayer p–n junction to electrically control the registry of vibronic states. To see this, we conducted detailed V_SD-dependent measurements of the interlayer exciton photoresponse. Figure ​Figure33a shows photoconductance (dI_PC/dV_SD) maps as a function of V_SD and V_G for a reverse-biased p–n heterojunction (device 2) with E_PH = 0.99 eV. In these dI_PC/dV_SD maps, multiple vertical stripes stand out, corresponding to prominent periodic oscillations (as a function of V_SD) in the photoconductance (Figure ​Figure33b). Indeed, this periodic oscillation in photoconductance is confirmed via Fourier transform of the dI_PC/dV_SD maps and displays a well-defined V_SD periodicity. These photoconductance oscillations, observed at room temperature, can be understood to arise from V_SD shifting the entire registry of vibronic peaks so that each peak shifts in and out of resonance with the fixed E_PH.

Open in a separate window

Figure 3

Electrical control of the vibronic photoresponse in WSe₂-MoSe₂ heterojunction devices. (a) Interlayer photoconductance dI_PC/dV_SD vs V_SD and V_G for device 2 (E_PH = 0.99 eV) at room temperature. Data were extracted from a large set of spatial photocurrent maps to isolate the WSe₂-MoSe₂ heterostructure photoresponse (Supporting Information Section S4¹⁶). (b) Photoconductance vs V_SD line trace extracted from the photoconductance map (averaged over the peak response region between V_G = −0.3 and 0.4 V). (c) 2D Fourier transform of the data in a. (d) Color map of d²I_PC/dΔE² vs ΔE and V_G. ΔE = eV_SD/αη is determined by combining the phenomenological factors extracted from the data in Figure ​Figure11f with the dimensionless parameter η = t/d. V_SD values are shown on the top axis. (e) d²I_PC/dΔE² versus ΔE. (Inset) Fourier transform spectral density of the Fourier transformed and averaged data from a. (f) 2D Fourier transform of the second-derivative data in a. Additional details are in Supporting Information Sections S2.2 and S4.3.¹⁶

Electrical control of the vibronic peaks is consistent with the fact that interlayer excitons exhibit an electric dipole, p, that points out of plane. The exciton–phonon vibronic states can therefore naturally be sensitive to an out-of-plane electric field, E_i. Indeed, the electrical control of ω_vib,n (V_SD) can be understood through a Stark shift ΔE = p·E_i = edV_SD/tα = eV_SD/αη that uniformly shifts the registry of vibronic states. Here, η = t/d is a dimensionless quantity that compares the interlayer distance t to the effective dipole length d (assumed to be approximately the same for all states). The ideality factor α (obtained in fits of Figure ​Figure11f) appears in this expression to phenomenologically capture the reduction of V_SD due to leakages such as nonideal contacts.²⁹ In the vdW p–n junction photodiode, the Stark shift and thus the registry of vibronic states are directly controlled using V_SD.

Utilizing the V_SD-dependent photoresponse to probe interlayer vibronic states, as demonstrated here, differs strongly from previous experiments on vdW heterostructures. Recently, interlayer exciton dipole strengths were estimated by studying an electrically controlled Stark shift through purely optical signals,^30,31 and no evidence of a vibronic registry was reported. In our p–n junction photodiode, we can utilize the V_SD-controlled optoelectronic response to estimate the dipole strength. As an illustration, we note that the separation between periodic vibronic peaks in Figure ​Figure22f approximately corresponds to 30 meV, yielding a modest estimate η = t/d ≈ 2.5 (consistent with recent measurements of interlayer exciton dipole strengths³⁴). This yields energy-dependent d²I_PC/dΔE² vs ΔE oscillations with a 30 meV increment (Figure ​Figure33d,f). In these (Figure ​Figure33e), each pronounced dip corresponds to a vibronic state ω_vib,n (V_SD).

These results also differ from previous experiments on individual monolayer vdW materials. Although the absorption spectra of monolayer TMDs generally display broad absorption peaks,³¹⁻³⁴ optical measurements of some vdW materials have attributed sparse and isolated sidebands to individual vibrational modes.³⁵⁻³⁹ In MoSe₂^40,41 and WSe₂,^42,43 optical spectroscopy measurements have resolved sideband features attributed to strong intralayer exciton–phonon coupling yet reported only at very low temperatures. Here, photoconductance oscillations are clearly observed at room temperature (Figure ​Figure33), indicating that the photocurrent is a sensitive probe of interlayer excitons, which are particularly susceptible to exciton–phonon coupling that emerges from the stacking of two vdW crystal layers. Combining the photocurrent with other advanced optical and optoelectronic techniques may allow the further elucidation of the fine details of the electron–phonon coupling in the heterostructure (e.g., distinguishing adiabatic⁵ and nonadiabatic/resonant couplings⁴⁴) and the measurement of its strength (e.g., determining whether the vibronic coupling is strong enough to induce self-trapping of the excitons^7,9). Such advanced techniques could include coherent optical spectroscopy such as 2D electronic spectroscopy or methods to enhance the observation and identification of relevant phonons (e.g., resonant Raman).

By stacking atomically thin semiconductors, we have demonstrated a new type of device that harnesses both vibrational and electronic energy to absorb near-infrared light, an important part of the solar light spectrum. In solid-state optoelectronics, atomic vibrations are often thought of as a loss mechanism. For example, the excess energy of photoexcited electrons above the band edge is typically lost to phonons, thus lowering the efficiency of solar cells. This paradigm is turned on its head in photosynthetic complexes, wherein atomic vibrations are instead harnessed to enhance energy transport through strong vibronic coupling. Our device, a rudimentary vibronic photodiode, exhibits vibronic effects that are often observed in photosynthesis yet have not been harnessed in solid-state devices. Stack engineering, exemplified here by stacking two vdW layers, MoSe₂ and WSe₂, gives rise to an entire registry of exciton–phonon vibronic states that can be individually addressed and electrically controlled in a solid-state setting. From a broader perspective, access to vibronic states in these vdW layers may enable excitonic phenomena more traditionally found in molecule-like systems, ranging from singlet fission⁴⁵ and exciton dissociation⁴⁶ in organic compounds to long-lived coherent dynamics and enhancements to exciton transport in photosynthetic complexes.^{1−4,47−49} We anticipate that stack engineering of strong exciton–phonon coupling will establish vdW heterojunctions as a versatile platform for controlling vibronic physics in 2D semiconductors devices.