## Introduction

Photodetectors have potential for several applications, ranging from communication, aerospace and military technologies to pharmaceutical research1. However, high-performance photodetection over a range that extends into the ultraviolet (UV) range has proven to be challenging. Traditional Si-based thin-film technologies tend to underperform at short wavelengths due to a low optical absorption2,3,4. Alternative devices, such as photomultiplier tubes (PMTs), charged-coupled devices (CCDs) and semiconductor photodiodes1,5,6, can achieve higher absorption in the UV range but at the cost of a slow photoresponse, high power consumption and/or low efficiency6,7,8. Two-dimensional (2D) van der Waals (vdW) semiconductors offer an alternative solution. They have tuneable absorption and fast temporal responses at room temperature while remaining flexible and power efficient9,10,11,12,13,14,15. They also possess surfaces that are free of dangling bonds leading to a low density of interface traps and the formation of high-quality heterostructure devices without the need for lattice matching16. Thus, multi-layer vdW heterostructures could be used to enhance the performance of a photodetector by exploiting the optical properties of different materials and the unique effects that arise from their interactions, such as charge transfer, built-in electric fields, and moiré superlattices17,18,19,20.

Among the wide catalogue of 2D materials, the III–VI metal chalcogenide (MC) compounds have emerged as an ideal class of semiconductors for optoelectronics9,21,22,23. High-performance photodetectors based on individual flakes of either GaSe or In2Se3 have been reported before21,23,24,25. These two 2D MC layers have optical band gaps in the visible (GaSe)8 and in the near-infrared (In2Se3) spectral range26, and strong absorption in the UV region21,27,28,29. Thus, photodiodes based on multiple photon absorbing layers, such as GaSe and In2Se3, and/or their pn-junctions are well suited for broadband detection that extends into the UV range. However, the operation of a pn-junction based on 2D materials can differ qualitatively from that of a traditional semiconductor diode and may not be suitable for high-sensitivity photodetection due to large dark currents and interlayer carrier recombination30,31,32.

In this work, we report on graphene/p-GaSe/n-In2Se3/graphene heterostructures where a p-GaSe/n-In2Se3 junction and two graphene electrodes are vertically stacked. These devices have photoresponsivity that extends into the UV range under both forward and reverse bias, or without any externally applied voltage. The bias-dependent photoresponsivity differs from that of a conventional pn-junction and arises from photon-current conversion at the three interfaces of the heterostructure (i.e. graphene/GaSe, GaSe/In2Se3 and graphene/In2Se3 interfaces) with graphene acting as an optically transparent electrode and efficient carrier collector, and the GaSe and In2Se3 layers providing the photon absorbers. Furthermore, the type-II band alignment of the GaSe/In2Se3 heterostructure and the electric field at the three vdW interfaces are beneficial to suppress carrier recombination in the junction for enhanced photoresponsivity and detectivity, comparable to that of state-of-the-art photodetectors.

## Results and discussion

### Photodetectors based on graphene-contacted p-GaSe/n-In2Se3 junctions

Figure 1a shows a schematic of the p-GaSe/n-In2Se3 heterojunction with two graphene electrodes (left) and an optical image of a typical device (right). The heterostructure is assembled by exfoliation and mechanical stamping of nanometre-thick vdW layers of α-In2Se3 and ε-GaSe. Details of the fabrication and characterization of the layers by X-ray diffraction, Raman and photoluminescence (PL) spectroscopy are in the ‘Methods’ section and Supplementary Fig. 1. The layers are prepared from bulk Bridgman-grown crystals of α-In2Se3 and ε-GaSe polytypes that have a hexagonal (2H) crystal structure (Fig. 1b). The single α-In2Se3 vdW layer consists of five atomic layers stacked in the sequence Se–In–Se–In–Se (1 quintuplet layer (QL)). For ε-GaSe, each vdW layer consists of four closely packed, covalently bonded, monoatomic sheets in the sequence Se-Ga-Ga-Se. Along the c-axis, the primitive unit cells of α-In2Se3 and ε-GaSe have lattice parameters of cIn2Se3 = 19.210 Å and cGaSe = 15.949 Å, respectively. Within each a-b plane, atoms form hexagons with lattice parameters aIn2Se3 = 4.031 Å and aGaSe = 3.755 Å. Although the crystals are intentionally undoped, both layers contain residual dopants. For bulk α-In2Se3, Hall effect measurements indicate an electron density n = 4.9 × 1017 cm−3 at T = 300 K due to the existence of native donors33. The p-GaSe crystals contain instead native acceptors and a low concentration of holes p ~ 1013 cm–3 at T = 300 K34.

Figure 1b shows the type-II band alignment and built-in potential eVbi = 1.7 eV of a p-GaSe/n-In2Se3 junction at equilibrium at T = 300 K, as derived by solving Poisson’s equation and taking into account the band alignment and doping of the bulk layers. Here, we have used the electron affinities from the literature (χGaSe = 3.7 eV35, χIn2Se3 = 3.6 eV36) and the measured band gap energy of the bulk crystals (EGaSe = 2.0 eV, EIn2Se3 = 1.4 eV, Supplementary Fig. 1). The depletion region of the junction extends primarily in the p-GaSe layer and is negligibly small in n-In2Se3 due to the much larger carrier concentration and dielectric constant of In2Se3 (εIn2Se3 ~ 17 and εGaSe ~ 6)37,38. As shown in Fig. 1c, the room temperature (T = 300 K) current–voltage (IV) characteristics are asymmetric with respect to the polarity of the applied voltage with a rectification ratio of up to 1000 at V = ± 2 V (Fig. 1c, blue curve). The current density remains small in both forward and reverse bias. Although the I-V curves of the junctions appear qualitatively similar to those of a traditional pn-junction, important differences emerge when we examine devices with thin layers (Fig. 1c). As shown in Fig. 1c, with decreasing layer thickness the current in both forward and reverse bias tends to increase and the I-Vs become more symmetric with respect to the polarity of the applied voltage. These differences can arise from a qualitative change in the transport mechanisms. In a bulk-like junction, regions depleted of free carriers form on either side of the junction, thus generating built-in potentials. Carrier transport across the junction occurs by diffusion. However, for junctions composed of thin p- and n-type semiconductors, tunnelling-assisted interlayer recombination of majority carriers can increase the current in forward bias32; also, charge transport via tunnelling across a thin barrier between the two graphene electrodes and/or leakage via defects in the junction can dominate the current in both forward and reverse bias. Further differences emerge when we consider the photoresponse (Figs. 25), as discussed below.

Figure 2a shows the IV characteristics of a junction under illumination with unfocused laser light of energy  = 3.06 eV (λ = 405 nm). In both forward and reverse bias, the current increases with increasing incident power Pi. A photovoltaic response is also observed with a short-circuit current (V = 0) of up to 0.15 nA for Pi = 0.11 μW (inset of Fig. 2a). Figure 2b, c show the dependence of the photoresponsivity R = ΔI/Pi, on V and Pi under unfocussed light illumination with λ = 405 nm and λ = 633 nm, respectively. Here, ΔI = Ilight − Idark is the photocurrent. A large value of R of up to 145 A W−1 is observed at V = + 3 V at λ = 405 nm and Pi = 0.11 nW. The corresponding values of the external quantum efficiency (EQE = Rhν/e) and detectivity (D* = R(A/2eIdark)1/2) are EQE = 4.4 × 102 and D* = 1.0 × 1013 Jones, where A = 20 μm2 is the area of the device. Also, we estimate the noise equivalent power (NEP) as $${\rm{NEP}} = \frac{{I_{{\rm{noise}}}}}{R}$$, where Inoise is the noise current derived from the standard deviation in the temporal fluctuations of the dark current. For $$R = 145\;{{{\mathrm{A}}}}/{{{\mathrm{W}}}}$$ and Inoise = 4.8×10−12 A, $${\rm{NEP}} = 3.3 \times 10^{ - 14}$$ W. The corresponding signal-to-noise ratio is $${\rm{SNR}} = \frac{{I_{{{{\mathrm{signal}}}}}}}{{I_{{{{\mathrm{noise}}}}}}} = 6.0 \times 10^4$$. For all applied voltages, R is approximately one order of magnitude larger at λ = 405 nm than at λ = 633 nm, reflecting an increased optical absorption of the layers at shorter wavelengths. Furthermore, the junction can be switched between on and off states in a reproducible manner with rise and decay times of τ ~ 100 ms and without any photo-induced memory effects and/or hysteresis in the IVs (Supplementary Fig. 2). The trapping/detrapping of charge carriers due to the disordered potential of the heterostructure and/or crystal defects tend to slow down the photoresponse. We estimate the bandwidth from the photocurrent rise time/decay time, e.g. $$f_{3{{{\mathrm{dB}}}}} = 0.35/\tau = 4.5\;{{{\mathrm{Hz}}}}\;{{{\mathrm{and}}}}\;{{\Delta }}f = f_{3{{{\mathrm{dB}}}}}/0.886 = 5.1\;{{{\mathrm{Hz}}}}$$. Here, $$f_{3{{{\mathrm{dB}}}}}$$ is the 3-dB bandwidth of the detector. This is defined as the frequency at which the photocurrent decreases by 3 dB of its maximum value39. Using these parameters to estimate the detectivity, we find $$D^ \ast = \frac{{\sqrt {A{{\Delta }}f} }}{{{\rm{NEP}}}}\sim 3 \times 10^{10}$$ Jones. This is smaller than the value of D* ~ 1013 Jones, as derived from D* = R(A/2eIdark)1/2, which assumes that the total noise is dominated by the shot noise from the dark current.

We have investigated junctions with different thickness of the layers (Supplementary Figs. 35) and all revealed a similar behaviour with a photoresponse that extends from the NIR (near-infrared) to the VIS (visible) and the UV range even without any applied voltage. Figure 3 shows the dependence of the responsivity on the photon wavelength at V = 0 for a junction based on a relatively thin (~60 nm) p-GaSe/n-In2Se3 junction. It can be seen that the photoresponse increases with decreasing photon wavelength, with a stronger increase observed at energies larger than the band gap energy of GaSe. Although the responsivity remains high in all devices, the detectivity and photovoltaic response tend to become weaker in the shorter junctions due to larger dark currents and shorter depletion regions.

The responsivity of all devices is comparable or higher than the values for photodetectors based on individual flakes of GaSe21,24,28 or In2Se325,26,40,41,42. Also, pn-heterojunctions based on GaSe from this work and the literature (Supplementary Table 1) tend to have a photoresponsivity that is higher under forward bias conditions. This behaviour differs from that of a traditional diode. In the latter, the photocurrent is larger in reverse bias as the spatial extent of the space charge region and the magnitude of the electric field increase with increasing reverse bias, thus facilitating the separation of photocreated electrons and holes and their collection at the electrodes. However, for our junctions, as the forward voltage increases above V ~ +1.5 V (Fig. 2a–c), the photocurrent and photoresponse can become larger than that in reverse bias. This behaviour is observed for different powers and laser wavelengths. Thus, the photoresponse cannot arise just from the photon-current conversion at the In2Se3/GaSe interface (Fig. 2d, e) and additional mechanisms should be considered at large positive V. To examine these mechanisms, we calculate the energy band alignment for a graphene/p-GaSe/n-In2Se3/graphene junction at V = 0 and under forward bias (see Supplementary Note 1 and Supplementary Fig. 6). We assume that the layers are depleted and that the pn-junction and Schottky junctions can be treated in isolation. From the electron affinity of the layers (χGaSe = 3.7 eV35, χIn2Se3 = 3.6 eV36 and χGraphene = 4.5 eV43) and their doping, we estimate a Schottky barrier height of ΦIn2Se3 = 1.06 eV and ΦGaSe = 1.04 eV at the In2Se3/graphene and GaSe/graphene interfaces, respectively (Supplementary Table 2). The depletion regions of the junction extend primarily in the p-GaSe layer and are much smaller in n-In2Se3 due to the larger carrier concentration of In2Se3. When the junction is forward biased, the photocurrent can increase due to an enhanced contribution to the photoresponse from the separation of electron-hole pairs at the In2Se3/graphene (Fig. 2f) and GaSe/graphene (Fig. 2g) interfaces. Thus, in forward bias, both interfaces behave as reverse-biased Schottky contacts and can contribute to the IV with a photocurrent that has the same polarity as the dark current. Since the In2Se3 layer has a band gap energy EIn2Se3 ≈ 1.4 eV, significantly smaller than that of GaSe (Supplementary Fig. 1), the contribution from the In2Se3/graphene interface becomes dominant for photon energies below the band gap of GaSe (EGaSe = 2.0 eV). In summary, the three interfaces of the graphene/p-GaSe/n-In2Se3/graphene heterostructure contribute to the photodetection. Although our model does not capture the evolution of the multi-layer structure as the layer thickness decreases down to a few nanometres, it illustrates that the built-in potential at the three interfaces plays a critical role in the photoresponsivity. This is further examined by spatial mapping of the photoresponse of junctions with different thickness.

### Scanning photocurrent microscopy

We used a confocal optical microscope to obtain a photocurrent map of the junctions by focusing and scanning a laser beam onto the layer plane of the samples. Figure 4a shows the bias-dependence of the photocurrent under focused laser excitation of a junction (inset of Fig. 4a). The corresponding IV curves under unfocused laser excitation are shown in Fig. 2a. Figure 4b–f show the corresponding photocurrent maps of the device acquired under a range of bias voltages. To distinguish the different areas of the heterojunction, in each map, the GaSe, In2Se3 and graphene layers are marked by dashed lines. At zero bias (Fig. 4b), the photocurrent signal arises from the area where the four layers (top and bottom graphene, GaSe and In2Se3) of the heterostructure overlap, indicating a built-in potential at the pn-junction and efficient extraction of photo-created carriers at the two graphene electrodes.

Some general considerations can be drawn from the inspection of the photocurrent maps of several junctions. For example, by comparing the photocurrent maps of Fig. 4, it can be seen that at V = ±1 V (Fig. 4c, e) the amplitude of the photocurrent signal in forward bias is smaller than that in reverse bias; in contrast, at V = ± 2 V (Fig. 4d, f), the photocurrent in forward bias is larger than in reverse bias. This behaviour is illustrated by the bias dependence of the photocurrent in Fig. 4a and is consistent with the IV characteristics acquired under unfocused laser light in Fig. 2. The responsivity values in the focused and unfocused mode versus the laser power density are plotted in Supplementary Fig. 7. The data show that the responsivity in the focused laser mode is in line with data for the unfocused mode. In particular, the value of R tends to decrease with increasing laser power due to the increase in carrier recombination rate for increasing densities of photogenerated carriers25,26,27,28. Although the behaviour of junctions with different layer thickness is qualitatively similar, in the thinner junctions the photoresponsivity becomes more symmetric with respect to a reversal of bias polarity (Supplementary Fig. 4).

In general, the photocurrent can be non-uniform in the plane of the junction, revealing hot spots of photoresponse. These can arise not only from the non-uniformity of the individual layers, but also from their interlayer coupling, which can vary in the layer plane due to defects and/or surface roughness, and a non-uniform spatial distribution of dopants at each interface. Thus, the relative contribution of the three vdW interfaces (e.g. GaSe/In2Se3, GaSe/graphene and In2Se3/graphene) to a photocurrent map can vary from device to device (Supplementary Figs. 35). Figure 5 illustrates a junction in which the dominant contribution to the photocurrent arises from a large overlapping area between GaSe and graphene. Figure 5a shows the bias-dependence of the photocurrent under focused laser excitation of the junction (inset of Fig. 5a). As the carrier density of In2Se3 is much larger than GaSe, the depletion region extends primarily in the p-GaSe layer and is negligibly small in n-In2Se3. Similarly, we expect a charge redistribution at the GaSe/graphene interface, leading to the formation of a GaSe/graphene Schottky contact. The separation of photocreated carriers at the GaSe/graphene interface generates a photocurrent with opposite sign to that of the pn-junction at V = 0 (Fig. 5b) and same sign in forward and reverse bias (Fig. 5c, d). For this junction, it can be seen that the photocurrent signal is observed outside the region of overlap between the two graphene layers, where the electric field should be effectively zero. This current can arise from a non-uniform electric field potential in the layer plane and a temperature gradient induced by non-uniform heating under the focused laser beam through the photo-thermoelectric effect44,45.

In summary, these photodetectors enable an efficient collection of photo-generated carriers resulting into a high photoresponsivity. Bulk junctions are more preferable for photodetection as the dark current is lower in both forward and reverse bias, leading to a large detectivity. In reverse bias, the low dark current indicates a small leakage current due to negligible thermal excitation of carriers across the band gaps of GaSe and In2Se3; in forward bias, the low dark current suggests instead a small recombination current. The mechanisms for photodetection in forward bias and the recombination of carriers are further examined in the following section.

### Carrier recombination and its bias dependence

We investigated the recombination of carriers by conducting electroluminescence (EL) and PL experiments under an applied voltage. No EL signal was detected in these junctions. However, a PL signal from both GaSe and In2Se3 was measured and found to be sensitive to the applied bias. Figure 6a shows the RT PL spectra at V = 0 for the junction of Figs. 2 and 4. The PL spectrum comprises a weak band edge emission from the In2Se3 (band I) and GaSe (band III) layers, and a broad, structured emission (band II) centred at around 1.8 eV, at energies between band I and III, with amplitude that tends to saturate with increasing laser power. This emission is spatially localized (see micro-PL imaging in Fig. 6b), indicative of an extrinsic carrier recombination in GaSe.

The energy peak position of band II does not reveal any significant shift with the applied bias. However, the intensity of this band is modulated by the applied voltage, reaching a maximum at V ~ 1.5 V (Fig. 6c, d). As V increases, the built-in electric field at the In2Se3/GaSe interface decreases and electron–hole recombination becomes more favourable. On the other hand, as V increases, the electric field at the graphene/GaSe interface also increases, thus causing a bias-induced quenching of the PL emission at large V. The quenching of the PL emission over a wide range of biases is pronounced (~100%), supporting an effective charge separation and light-current conversion at the In2Se3/GaSe and GaSe/graphene interfaces. We note that the bias-dependence of the PL emission due to In2Se3 (band I, Fig. 6d) is slightly different. Although the initial increase of PL intensity with increasing V is observed also in this case, this is then followed by a saturation of the PL signal at large V. We assign this different behaviour to the larger carrier concentration in In2Se3 (n ~ 1017 cm−3) compared to that in GaSe (p ~ 1013 cm−3). Combined with a large dielectric contrast of In2Se3 (εIn2Se3 ~ 1737,38), this leads to more effective screening of the electric field in In2Se3.

While the separation of charges at three interfaces can explain both the photoresponse and carrier recombination, the origin of band II remains unclear. In particular, the observation of this band is suggestive of a recombination channel that is not beneficial for photodetection. Theoretical studies of the energetics of crystal defects in GaSe have shown that acceptor levels due to Ga vacancies (VGa) have energies at approximately 0.2 eV above the valence band (VB) edge, while interstitial Ga (Gai) create donor levels at 0.2 eV below the conduction band (CB) edge46. Large scale defects, including clusters of point defects and cracks/local stacking faults, can also form as the result of mechanical stress applied to a crystal during fabrication47. These can induce localized regions with enhanced PL emission in the range 1.7–2.0 eV48, as observed in our micro-PL imaging studies (Fig. 6b). Similar, but broader extrinsic PL emissions can be generated via controlled thermal annealing of individual flakes of GaSe (see thermal annealing studies in Supplementary Fig. 8). The sensitivity of these junctions to charges trapped by defects of GaSe indicates that efficient optoelectronic devices based on this material require an optimization of the fabrication process. A crucial requirement is the in-plane spatial homogeneity of the layers, which can affect interlayer coupling and band alignments. Furthermore, GaSe can oxidize in air leading to morphological and chemical changes of its surface. Thus, an effective passivation strategy should also be adopted to mitigate these effects. To avoid the oxidation of GaSe during our device studies, all electrical studies were conducted in vacuum and at laser powers that avoid light-induced changes of GaSe. The stability tests of our devices under these conditions are shown in Supplementary Fig. 9. The measured current remains stable over time, switching between high to low (dark current) values in response to the impinging laser power.

In conclusion, we have exploited the vdW interfaces of the graphene/p-GaSe/n-In2Se3/graphene heterostructure to fabricate high-responsivity UV photodetectors. Their operation exploits the optical absorption of the metal chalcogenide layers and the built-in potential at the three vdW interfaces of the heterostructure. The junctions can have high photoresponsivity and low dark current in both forward and reverse biases, and without any externally applied bias. Thus, these multi-layer vdW heterostructures could play a crucial role in the development of high performance 2D semiconductor-based devices for a broad range of applications. Scanning photocurrent microscopy revealed that the contribution of the three interfaces to the overall responsivity can differ from device to device due to the non-uniformity of the layers and their coupling. For example, defects can act as recombination centres in GaSe leading to a photoluminescence emission that is modulated by the applied voltage. Thus, these devices could be improved by optimization of their fabrication to reduce the contribution of crystal defects and optimize interlayer coupling. Further optimization of the layer thickness and doping, and scalable growth of the layers could also lead to optimal designs for even larger responsivity and detectivity, offering an alternative to conventional UV detection technologies.

## Methods

### Materials and device fabrication

We used Bridgman-grown α-In2Se3 and ε-GaSe crystals. The crystal structures of the α-In2Se3 and ε-GaSe bulk crystals were determined from X-ray diffraction studies on a DRON-3M diffractometer. The room temperature PL emission of bulk α-In2Se3 and ε-GaSe are centred at about 1.4 eV and 2.0 eV respectively. Chemical Vapour Deposition (CVD) graphene, exfoliated α-In2Se3 and exfoliated ε-GaSe flakes form the graphene/In2Se3/GaSe/graphene heterostructure. The thicknesses of the GaSe and α-In2Se3 flakes were estimated from their colour profiles in the optical images. Each heterostructure was assembled onto a SiO2/p-Si substrate (SiO2 layer thickness of 300 nm). For the stamping of the top and bottom graphene layers, we used a needle-assisted transfer process of patterned graphene microsheets, as described in ref. 49. The bottom graphene flakes were first deposited onto the SiO2/p-Si substrate with pre-patterned metal contacts (Au/Ti ~ 100 nm/10 nm) using a dry transfer method under ambient conditions. Flakes of In2Se3 and GaSe were exfoliated onto polydimethylsiloxane (PDMS) from a low-tack, low-residue tape and identified using transmission and reflection mode optical microscopy. Then, using a micromanipulator stage, an exfoliated flake of α-In2Se3 on PDMS was aligned to the pre-deposited bottom graphene flake and brought into contact. The PDMS was then slowly retracted in order to deposit α-In2Se3. A GaSe flake was stamped on top of the In2Se3 flake by the PDMS-assisted method mentioned above. Finally, the top graphene contact was aligned to the In2Se3/GaSe heterostructure and Au/Ti metal contact.

### Electrical, optical and microscopy measurements

Electrical and opto-electrical measurements were acquired using a Keithley 2400 Sourcemeter, an Agilent Technology B1500A Semiconductor Analyser and a lock-in amplifier (SR830) with a light chopper. Unfocused laser beams of wavelength λ = 405 nm and 633 nm were used for the photocurrent measurements. The continuous spectrum from NIR to VIS and UV monochromatic illumination was provided by a Fianium WhiteLase Supercontinuum Laser Source and a Xe lamp, respectively. The output laser wavelength was tuned by an Omni-λ300 monochromator. For photocurrent mapping experiments, a focused laser beam (λ = 405 nm or 425 nm and spot diameter d ≈ 1 μm) was used. All transport measurements were conducted in a vacuum environment (~10−6 mbar). The experimental setup for the micro-Raman and micro-PL spectroscopy studies at room temperature comprised of a frequency doubled Nd:YVO4 laser (λ = 532 nm and d ≈ 1 μm), an xyz motorized stage and an optical confocal microscope system equipped with a 0.5-m long monochromator with 1200 g/mm and 150 g/mm gratings. The signal was detected by a Si-charge-coupled device camera. The optical studies were performed at low excitation power to avoid excessive heating. The laser annealing studies were conducted using this set up and different laser powers.