Gate‐Tunable Polarity‐Reconfigurable Photovoltaic Response With Dual Linearity in NbS₃/WSe₂ Heterostructures for In‐Sensor Image Processing

ABSTRACT

Reconfigurable optoelectronic image sensors capable of in‐sensor computing are promising for energy‐efficient edge visual perception. Self‐powered photovoltaic operation further reduces the power demand of image sensing and processing. However, it remains challenging to achieve a continuously gate‐tunable self‐powered photovoltaic response that combines reversible polarity with linear weight programmability. Here, we demonstrate a gate‐tunable photovoltaic detector based on an NbS₃/WSe₂ van der Waals heterostructure. Owing to the ambipolar transport of WSe₂ and the distinct electrostatic responses of the two constituent materials, gate voltage continuously modulates the interfacial band alignment and reversibly switches the built‐in electric field at the heterointerface. Consequently, the device exhibits gate‐controlled positive and negative photovoltaic responses. The short‐circuit photocurrent scales linearly with incident power density, while the short‐circuit photoresponsivity varies linearly with gate voltage over a broad programming window. This dual linearity allows the photoresponsivity to function as a programmable analog weight for in‐sensor convolution. As a proof of concept, several image convolution kernels are implemented for in‐sensor image processing, and the experimentally obtained outputs agree well with algorithmic simulations. This work provides a promising device‐level building block for reconfigurable and energy‐efficient vision sensors for edge computing.

An NbS₃/WSe₂ van der Waals heterostructure realizes a continuously gate‐tunable self‐powered photovoltaic response through interfacial band‐alignment modulation and reversible built‐in electric‐field switching. The short‐circuit photocurrent scales linearly with incident power density, while short‐circuit photoresponsivity varies linearly with gate voltage. This dual linearity enables programmable analog weights for reconfigurable in‐sensor convolutional image processing.

1. Introduction

Rapid advances in edge computing and the internet of things are driving increasing demand for intelligent vision systems that combine low power consumption with real‐time operation [1, 2, 3]. In conventional von Neumann architectures, image sensing and data processing are physically separated, resulting in substantial energy consumption, latency, and hardware redundancy caused by frequent data transfer and signal conversion between the sensor and the processor [4, 5]. By contrast, in‐sensor computing moves part or all of the computational workload to the sensing front end, allowing image acquisition and preliminary processing to occur within the same hardware platform [6, 7, 8]. This close integration of sensing and computation can substantially reduce redundant data movement and back‐end processing overhead, making it attractive for next‐generation machine vision, autonomous driving, and remote sensing applications [9, 10, 11, 12, 13, 14].

A central challenge in implementing in‐sensor computing is the development of optoelectronic devices that combine reliable photodetection capability with programmable response characteristics [15, 16]. In particular, such devices should provide sufficient sensitivity and fast response, while enabling linear and reversible modulation of photoresponsivity (R) across both positive and negative values. This capability permits hardware‐level emulation of positive and negative weights in neural network operations and supports a broad range of computational functions [17, 18]. Among the available modulation strategies, gate voltage control is especially attractive because it is compatible with standard semiconductor processes and offers precise electrical programmability [19, 20, 21]. However, conventional semiconductor photodetectors are constrained by fixed band structures and doping profiles, which prevent dynamic reconfiguration of the built‐in electric field and photocurrent direction and thereby limit their utility in reconfigurable in‐sensor computing [22]. To this end, research has shifted from fixed‐response photodetectors toward reconfigurable devices that enable electrical programming of sensing and computing functions. Self‐powered operation with linearly programmable bipolar photoresponses is especially attractive for low‐power machine vision, as it allows optical sensing and signed analog weighting to be integrated at the sensing front end. Yet achieving both reversible photovoltaic polarity and linear weight programmability in a single device is a critical prerequisite for building high‐fidelity in‐sensor computing systems.

Two‐dimensional (2D) van der Waals heterostructures provide a versatile materials platform for reconfigurable optoelectronic devices [23, 24, 25, 26]. Their atomic‐scale thickness enables the carrier type and concentration to be effectively modulated by gate bias, thereby shifting the Fermi level [27, 28]. When one constituent layer, such as an ambipolar semiconductor, exhibits a gate‐sensitive Fermi level while the other layer remains comparatively insensitive to the gate field, the relative Fermi‐level alignment between the two materials can be continuously tuned. This dynamic band engineering modifies the interfacial band bending and built‐in electric field, enabling reversible control over photocurrent polarity [29, 30]. Nevertheless, polarity switching alone is insufficient for accurate in‐sensor computing. The photoresponsivity also needs to vary linearly over a wide gate voltage range, because linearity is critical for precise and predictable weight programming through gate bias and for constructing reconfigurable convolution kernels [31, 32]. Recent 2D material‐based optoelectronic devices have enabled reconfigurable visual processing through different mechanisms, including gate‐tunable van der Waals heterostructures for positive and negative photoresponses and convolution [19, 31, 33], electrically programmed non‐volatile neuromorphic photovoltaics and photomemristors for stored responsivity states [16, 34], and persistent‐photoconductivity‐based heterostructures for mid‐infrared trajectory perception [35]. These advances demonstrate diverse routes toward the integration of sensing, memory, and computing, while also highlighting the importance of predictable electrical programming of photovoltaic polarity and analog photoresponsivity for accurate in‐sensor operations.

Here, we demonstrate a linearly reconfigurable photovoltaic detector based on an NbS₃/WSe₂ van der Waals heterostructure. This combination was chosen because WSe₂ is ambipolar with strong gate dependence, whereas NbS₃ maintains n‐type transport over the measured gate range. Their distinct electrostatic responses enable gate voltage to continuously tune the Fermi‐level alignment and even reverse the built‐in electric field, providing the physical basis for the polarity‐reconfigurable photovoltaic response. Optoelectronic measurements reveal reversible switching between positive and negative photovoltaic responses over a broad spectral range, while the short‐circuit photocurrent scales linearly with incident optical power. More importantly, the photoresponsivity varies linearly with gate voltage over a wide 45 V window, providing a basis for precise and programmable analog weights. Building on this dual linearity, we further demonstrate in‐sensor convolution for several image processing functions. Overall, this work experimentally validates a route toward linearly tunable bipolar photovoltaic response through gate‐controlled band engineering and suggests a promising strategy for reconfigurable and energy‐efficient in‐sensor image processing.

2. Results and Discussion

Figure 1a summarizes a simplified materials‐selection rationale for polarity‐reconfigurable photovoltaic heterostructures. Material A possesses a strongly gate‐sensitive Fermi level, whereas material B exhibits a comparatively weaker electrostatic response, such that sufficient gate modulation can reverse their relative Fermi‐level alignment and consequently the interfacial band bending and built‐in electric field. Under a sufficiently negative gate bias, the Fermi level of material A shifts downward and the channel exhibits p‐type transport. The resulting Fermi level offset relative to material B generates a built‐in electric field (E _bi) directed from material B to material A. Under illumination, this field separates photogenerated electron‐hole pairs and produces a negative photovoltaic current. As the gate voltage is increased, the Fermi level of material A shifts upward and the channel becomes n‐type. Consequently, the sign of the Fermi level offset is reversed, which reverses the built‐in electric field and gives rise to a positive photovoltaic current [36, 37]. Thus, by exploiting gate‐tunable band alignment and the different electrostatic responses of the two constituent materials, the interfacial band bending and built‐in electric field can be reversibly controlled, providing the physical basis for bipolar photovoltaic response with both positive and negative polarities [38, 39, 40]. In an in‐sensor computing architecture, the device photoresponsivity directly serves as the programmable weight in convolution operations [33, 41]. The mapping between weight and gate voltage therefore directly determines computational precision and trainability. The solid line in Figure 1b represents the ideal linear response required for in‐sensor computing, in which photoresponsivity changes linearly with gate voltage across both positive and negative values. Such a relationship allows both the sign and magnitude of the weight to be accurately set by gate bias, supporting reliable convolution operations. By contrast, the dashed line represents a more common nonlinear response. Although the photoresponsivity can still switch from negative to positive values, such nonlinear mapping makes the programmed weight less predictable and less precise. This nonlinearity can introduce computational errors and consequently limit the use of such devices in applications that require accurate weight updates and high‐fidelity image processing [31, 32].

FIGURE 1.

(a) Schematic illustration of band‐alignment modulation in the heterostructure under gate bias, enabling polarity‐reconfigurable photovoltaic response. (b) Comparison between the desired linear (solid line) and typical nonlinear (dashed line) relationships between gate‐tunable photoresponsivity and gate voltage.

Figure 2a schematically shows the NbS₃/WSe₂ heterostructure fabricated on a SiO₂/Si substrate. The electrode connected to WSe₂ is defined as the drain, the grounded electrode connected to NbS₃ is defined as the source, and the bottom Si substrate serves as the gate. Raman spectroscopy with 532 nm excitation was used to verify heterostructure formation and the crystalline quality of the constituent materials. As shown in Figure S1, Raman spectra collected from the individual flakes and the overlap region exhibit the characteristic peaks of NbS₃ and WSe₂, confirming successful assembly of the heterostructure [42, 43, 44]. The transfer characteristics of individual NbS₃ and WSe₂ devices are provided in Figure S2, showing that NbS₃ behaves as an n‐type semiconductor, whereas WSe₂ exhibits ambipolar transport with strong gate dependence. The gate‐modulated output characteristics of the heterostructure are presented in Figure 2b and show typical diode‐like behavior. Notably, the rectification direction reverses under external gate modulation, consistent with the ambipolar nature of the WSe₂ channel. The rectification ratio extracted as a function of V _g is summarized in Figure 2c. Here, the rectification ratio is defined as |I _ds(+3 V)| / |I _ds(−3 V)|, where the absolute current values are used to compare the relative conduction asymmetry under opposite V _ds. As V _g is swept from positive to negative values, the rectification ratio evolves from below unity to above unity, indicating a gate‐induced reversal of the dominant rectification direction. This gate‐reconfigurable rectification provides electrical evidence that interfacial band bending and built‐in electric field can be continuously modulated by electrostatic gating [45, 46].

FIGURE 2.

(a) Schematic diagram of the NbS₃/WSe₂ heterostructure. (b) Output curves under gate voltage modulation, showing gate‐reconfigurable bidirectional rectification. (c) Rectification ratio (|I _forward| / |I _reverse|) as a function of V _g at V _ds = ±3 V. (d) KPFM image of the NbS₃/WSe₂ heterostructure and surface potential profile extracted along the white arrow. Schematic band alignments of the NbS₃/WSe₂ heterostructure in representative gate regimes corresponding to (e) negative and (f) positive photovoltaic responses.

To further probe interfacial carrier transfer and band alignment in the NbS₃/WSe₂ heterostructure, Kelvin probe force microscopy (KPFM) was performed to map the surface potential distribution across the heterointerface. The atomic force microscopy (AFM) topography and corresponding height profiles are provided in Figure S3. Owing to the different work functions of the two constituent layers, the KPFM image in Figure 2d reveals clear potential steps across the heterointerfaces. The contact potential difference (CPD) between the AFM tip and the material surface is defined as CPD_material = (Ф _tip—Ф _material)/e, where e is the elementary charge and Ф _tip and Ф _material are the work functions of the tip and the material, respectively [44, 47]. The line profile extracted from Figure 2d indicates that the Fermi level of WSe₂ is approximately 96 meV higher than that of NbS₃. The corresponding band alignment between NbS₃ and WSe₂ is provided in Figure S4. Based on these results, the mechanisms for gate‐reconfigurable photovoltaic response are schematically illustrated in Figure 2e,f. Under sufficiently negative gate bias, the heterostructure can be approximately described as n‐type NbS₃ coupled with hole‐dominated WSe₂. This interpretation is supported by the transfer characteristics in Figure S2, where WSe₂ exhibits ambipolar transport with hole‐dominated conduction at negative V _g, whereas NbS₃ retains n‐type transport throughout the measured gate‐voltage range. In this representative negative‐response regime, the interfacial built‐in electric field drives photogenerated electrons toward NbS₃ and holes in the opposite direction under illumination, resulting in a negative short‐circuit photocurrent. As V _g is increased, the Fermi level of WSe₂ shifts upward, and the WSe₂ channel becomes increasingly electron‐dominated. The heterostructure therefore gradually evolves toward an n‐type NbS₃/electron‐dominated WSe₂ configuration. In the representative positive‐response regime, the change in relative band alignment reverses the interfacial built‐in electric field and the direction of carrier separation, leading to a positive short‐circuit photocurrent.

The transport characteristics described above indicate that gate bias effectively modulates the built‐in electric field in the NbS₃/WSe₂ heterostructure. We therefore investigated the gate‐controlled photovoltaic characteristics of the NbS₃/WSe₂ photodetector. The left panel of Figure S5a shows the I _ds‐V _ds curves measured under 810 nm illumination at V _g = 12 V for different power densities. Under these conditions, the device exhibits a positive short‐circuit current (I _sc) and a negative open‐circuit voltage (V _oc). For clarity, we use the sign of I _sc to distinguish positive and negative photovoltaic responses. This operating condition is therefore defined as the positive photovoltaic response. By contrast, the I _ds‐V _ds curves measured at V _g = −33 V under 810 nm illumination show a negative I _sc and a positive V _oc, corresponding to a negative photovoltaic response (right panel of Figure S5a). Similar gate‐dependent switching between positive and negative photovoltaic responses is observed under 532 and 450 nm illumination, as shown in Figure S5, indicating that polarity‐reconfigurable photovoltaic behavior is maintained over a broad spectral range. Figure 3a–c show the time‐resolved photoresponse of the NbS₃/WSe₂ photodetector under 810, 532, and 450 nm illumination at V _g = 12 V and V _g = −33 V, with different shading levels representing different power densities. At each wavelength, the photodetector exhibits gate‐tunable positive and negative photocurrents, and the magnitude of both responses increases with power density. The extracted I _sc values as a function of power density under different wavelengths at V _g = 12 V and V _g = −33 V are presented in Figure 3d–f. The power‐dependent I _sc was fitted using I _sc ∝ P^α , where P is the incident power density. The fitted α values are close to 1 for all tested wavelengths, indicating an approximately linear relationship between I _sc and incident power density for both positive and negative photovoltaic responses [18, 36]. Such near‐linear behavior is favorable for high‐fidelity image information acquisition.

FIGURE 3.

Positive and negative photoresponses under varying power densities (indicated by different shading levels) at V _g = 12 V and V _g = −33 V under illumination at (a) 810 nm, (b) 532 nm, and (c) 450 nm. Extracted I _sc as a function of power density at V _g = 12 V and V _g = −33 V under illumination at (d) 810 nm, (e) 532 nm, and (f) 450 nm.

To clarify the origin of the photovoltaic response, spatially resolved photocurrent mapping was performed under different gate voltages. As shown in Figure S6, positive photocurrent is predominantly localized in the NbS₃/WSe₂ overlap region at V _g = 12 V, whereas the photocurrent in the same region reverses to negative polarity at V _g = −30 V. No significant photoresponse is observed at the metal electrodes or graphene/semiconductor contact areas, supporting that the gate‐reconfigurable photovoltaic response is dominated by the gate‐modulated built‐in electric field at the NbS₃/WSe₂ heterointerface [48, 49]. Figure S7 shows the response speeds of the positive and negative photovoltaic modes under laser irradiation. The rise and fall times are 349 and 376 µs, respectively, for the positive photoresponse, and 418 and 457 µs, respectively, for the negative photoresponse. This fast response supports the use of the NbS₃/WSe₂ photodetector for rapid image processing applications. Furthermore, we evaluated the cyclic stability of the positive and negative photovoltaic modes of the NbS₃/WSe₂ heterostructure device at V _g = 12 V and V _g = −33 V, respectively. As shown in Figure S8, both modes maintain their respective polarity and stable response amplitudes under repeated 810, 532, and 450 nm illumination without obvious degradation over the measured cycles, confirming good reproducibility over the tested spectral range. Since the corresponding R _sc values serve as programmable analog weights for convolution, stable photoresponse is essential for preserving both the sign and magnitude of the weights. Variations in the photoresponse would cause deviations in the programmed weights and consequently reduce the reproducibility and fidelity of the processed images. The observed cyclic stability therefore supports the reliable use of the programmed photovoltaic responses for in‐sensor image processing.

To evaluate the continuous gate tunability of the reconfigurable photovoltaic response, we performed gate voltage‐dependent optoelectronic measurements under 810, 532, and 450 nm illumination. Figure 4a shows the I _ds‐V _ds curves measured under 810 nm illumination. Over the gate voltage range from −33 to 3 V, a reversible transition in the photovoltaic response is clearly observed. At V _g = −33 V, the heterostructure exhibits a negative I _sc, whereas the current polarity becomes positive as V _g is gradually increased to 3 V. This result confirms that the photovoltaic polarity of the NbS₃/WSe₂ heterostructure can be reversibly reconfigured by gate bias. Similar behavior is observed under 532 and 450 nm illumination, demonstrating that the gate‐tunable photovoltaic response is robust across a broad spectral range (Figure 4b,c). Under fixed power density, we further measured the time‐resolved photoresponse at different gate voltages for 810, 532, and 450 nm illumination. As shown in Figure 4d–f, distinct positive and negative photocurrents are observed at zero bias across the continuously tunable gate voltage range, in agreement with the gate‐dependent I _ds‐V _ds characteristics. These results further verify that the photovoltaic current polarity can be reversibly programmed through electrostatic gating. A similar gate‐tunable switching behavior between negative and positive zero‐bias photocurrents is also observed in another independently fabricated NbS₃/WSe₂ heterostructure device (Figure S9). Beyond the linear mapping between optical power and photocurrent, in‐sensor convolution also requires a linear mapping between gate voltage and photoresponsivity. We therefore extracted the gate‐dependent short‐circuit photoresponsivity (R _sc) at different wavelengths. As shown in Figure 4g–i, the photovoltaic current measured over the gate voltage range from −33 to 12 V was converted into the corresponding positive and negative R _sc values. Notably, R _sc exhibits a near‐linear dependence on gate voltage throughout the programmed window under all tested illumination wavelengths. Linear fitting of the R _sc‐V _g curves yields R ² values of 0.992, 0.987, and 0.982 under 810, 532, and 450 nm illumination, respectively, indicating near‐linear gate programmability of the photoresponsivity. While residual nonlinearity may introduce finite errors in the programmed analog weights and the convolution output, the high fitting coefficients confirm that these deviations are limited over the investigated programming range. For more general convolution operations, the programming accuracy can be further enhanced by calibrating the R _sc‐V _g relationship and applying a compensated voltage‐to‐weight mapping. This 45 V gate‐voltage window enables direct programming of both the sign and magnitude of the convolution weight, thereby supporting the construction of different convolution kernels for multiple image processing functions. In future devices, the required gate‐voltage range could be reduced by improving the gate‐coupling efficiency through thinner dielectrics, high‐κ gate insulators, or local‐gate architectures. Furthermore, the photoresponsivity could be enhanced through optical cavities or plasmonic structures to improve light absorption, together with interface passivation or contact engineering to reduce recombination and collection losses, while preserving the linear I _sc‐P and R _sc‐V _g relationships.

FIGURE 4.

I _ds‐V _ds curves measured at different V _g values under illumination at (a) 810 nm, (b) 532 nm, and (c) 450 nm. Gate‐tunable switching between negative and positive photovoltaic responses under illumination at (d) 810 nm, (e) 532 nm, and (f) 450 nm. Calculated R _sc as a function of V _g from −33 to 12 V under illumination at (g) 810 nm, (h) 532 nm, and (i) 450 nm.

Gate voltages of different polarities effectively modulate the built‐in electric field of the NbS₃/WSe₂ heterostructure photodetector, producing reconfigurable positive and negative photovoltaic responses. Because the I _sc scales linearly with power density and the R _sc varies linearly with gate voltage, the device can perform simultaneous optical sensing and convolution operations. Figure 5a schematically illustrates the in‐sensor image processing mechanism based on the gate‐tunable heterostructure photodetector. As a proof of concept, a single NbS₃/WSe₂ device was operated sequentially to realize the nine elements of each 3 × 3 convolution kernel. For each image patch, pixel intensities were encoded as incident power densities P _i, and the gate voltage was sequentially adjusted to program the required photoresponsivities R _i. The nine photocurrent contributions were measured sequentially and summed as I _sc = ∑P _i R _i to yield one convolution output [50, 51]. This demonstrates single‐device sequential convolution, rather than simultaneous operation of a physical array. A future parallel implementation would require independently gated detectors with their photocurrents summed on a common output line (Figure S10).

FIGURE 5.

(a) Schematic illustration of in‐sensor image processing using gate‐tunable switching between negative and positive photocurrent in the NbS₃/WSe₂ heterostructure device. The input signal, represented by an optical signal under laser illumination, is processed by a 3 × 3 reconfigurable matrix kernel using V _g modulation to generate negative and positive R _sc values. (b) The input image is a 256 × 256 grayscale image (left). A representative 3 × 3 pixel region (top, right) from the input image is converted from grayscale to incident power (bottom, right). The inset shows the equation used to convert the input image from an 8‐bit grayscale map to an incident power map. (c) Demonstration of in‐sensor convolutional image processing using experimentally reconfigurable kernels, including sharpness, top Sobel, right Sobel, edge extraction, and emboss. The upper panel shows the gate‐voltage matrices used to program the experimental kernels, and the lower panel shows the corresponding processed images.

We then carried out proof‐of‐concept convolution image processing under 810 nm illumination. Each grayscale value in the input image (0–255) was converted into a corresponding incident power density P _i, ranging from 0.1 to 10 mW cm^{− 2}. As shown in Figure 5b, the upper 3 × 3 panel corresponds to the grayscale values of the original image, whereas the lower 3 × 3 panel represents the corresponding P _i values. Figure 5c presents the convolutional image processing results for several filters, including sharpness, top Sobel, right Sobel, edge extraction, and emboss, using the reconfigurable convolution kernel programmed in the NbS₃/WSe₂ device. The upper panel displays the gate‐voltage matrices used to program the experimental convolution kernels, and the lower panel shows the corresponding processed images obtained using these kernels. Similar in‐sensor convolution operations can also be performed under 532 and 450 nm illumination, as shown in Figures S11 and S12. Comparison with algorithmic simulations using standard kernels (Figure S13) shows that the experimentally programmed kernels reproduce the key image features. These results validate the feasibility of using the gate‐programmable positive and negative photoresponsivity of NbS₃/WSe₂ heterostructures as analog convolution weights, while future array integration will be required for fully parallel in‐sensor image processing.

3. Conclusion

In summary, we have demonstrated a gate‐tunable photovoltaic detector based on an NbS₃/WSe₂ van der Waals heterostructure with reconfigurable bipolar response. Through electrostatic modulation of the WSe₂ Fermi level, the built‐in electric field at the heterointerface can be continuously reversed, enabling the photocurrent polarity to switch between negative and positive values under illumination. Optoelectronic characterization confirmed that the short‐circuit photocurrent scaled linearly with incident power density, while the photoresponsivity varied linearly with gate voltage over a 45 V programming window. This dual linearity enabled the photoresponsivity to function as a controllable analog weight in convolution operations, with the gate voltage directly defining both the sign and magnitude of the weight. We further demonstrated several image convolution operations using a single device, and the experimental results agreed well with algorithmic simulations. These results indicate that the NbS₃/WSe₂ heterostructure offers a viable platform for integrating sensing and computation within a single pixel element. Further development of device arrays and on‐chip training strategies may facilitate future in‐sensor processors for edge applications.

4. Experimental Section

4.1. Device Fabrication

Single‐crystalline WSe₂ was commercially obtained from HQ Graphene, while NbS₃ single crystals were grown according to our previous work [52]. A highly doped (100) p‐type silicon wafer with a 300 nm thermal oxide layer (Addison) was cut and used as the device substrate. The NbS₃/WSe₂ heterostructure was assembled through mechanical exfoliation and dry transfer. First, few‐layer WSe₂ flakes were mechanically exfoliated directly onto a Si substrate with a 300 nm thermal oxide layer. An NbS₃ flake was exfoliated onto a polydimethylsiloxane film supported on a glass slide, and then precisely aligned and transferred onto the target WSe₂ flake using a home‐built micromanipulator system. The same transfer procedure was repeated to transfer graphene layers as contact electrodes at both ends of the NbS₃/WSe₂ heterostructure. To improve contact quality, the fabricated devices were annealed at 200°C for 2 h under an argon atmosphere.

4.2. Characterization and Photoresponse Measurements

Optical microscopy was performed using a Nikon Y‐IDP microscope. Raman spectroscopy and photocurrent mapping were conducted using a WITec Alpha 300R confocal microscope system equipped with a 100× or 50× objective and a 532 nm excitation laser. During photocurrent mapping, the current signal was amplified and recorded with a transimpedance amplifier, while an arbitrary function generator (DG5252, Rigol) was used to apply bias to the devices. Atomic force microscopy and Kelvin probe force microscopy measurements were carried out using a Bruker Dimension Icon system. The electrical and optoelectronic characteristics of the devices were measured at room temperature under high vacuum using a Keithley 4200 semiconductor parameter analyzer combined with a Lake Shore TTPX cryogenic probe station. A supercontinuum white light laser (SC400‐8, Fianium) coupled to a monochromator (AOTF‐V1, Fianium) was used as the light source, and the incident power was calibrated using a commercial Thorlabs power meter (PM200). For time‐resolved photoresponse characterization, the device output was connected to a Femto DLPCA‐200 amplifier and recorded using a Tektronix TDS2024C digital oscilloscope. A 520 nm diode laser (Coherent OBIS) served as the excitation source, modulated by a Rigol DG5252 function generator.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting File 1: advs77288‐sup‐0001‐SuppMat.docx.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.