Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Nature Communications volume 17, Article number: 2096 (2026) Cite this article
4000 Accesses
Metrics details
Conventional photovoltaic devices based on PN or Schottky junctions are inherently constrained by the material band gap, resulting in limited photoelectric conversion efficiency. The van der Waals ferroelectric material CuInP₂S₆ offers strong potential for efficient photocurrent generation via bulk photovoltaic effect, while heterostructure engineering provides additional degrees of freedom for performance modulation. Yet, the specificity in asymmetric CuInP₂S₆ heterojunctions remains poorly understood. Here, we design and fabricate an asymmetric photovoltaic device, consisting of Pt/CuInP₂S₆/Graphene heterostructure. The ferroelectric photovoltaic effect couples with the Cu⁺ migration dynamics, which could be manipulated by asymmetric interface barriers. Remarkably, the photovoltaic current is increased by 10 times and exhibits a positive-negative switching through polarization modulation. The heterojunction device achieves on-demand photovoltaic programming for in-sensor computing tasks, including edge detection with recognition rate F-score ~ 1 and binary pattern classification with 100% accuracy, which provides new insights for next-generation visual technology.
The development of a neuromorphic visual system demands photoelectric sensors that are ultrafast, high-responsivity, and reconfigurable for dynamic information processing^1,2,3. In recent years, diverse reconfigurable optoelectronic devices have been developed to detect optical stimuli and process image information in parallel^4,5,6,7,8, offering significant opportunities for enhanced image processing capabilities. However, conventional photovoltaic devices based on PN and Schottky junctions remain fundamentally limited by the material bandgap-manifested in the Shockley–Queisser limit and by carrier recombination losses⁹, which together constrain their photoelectric conversion efficiency. The limitations underscore the need for alternative material platforms capable of breaking bandgap restrictions and achieving superior photoelectric conversion performance.
The bulk photovoltaic effect (BPVE) of two-dimensional (2D) van der Waals (vdW) ferroelectric materials enables the separation of photogenerated carriers without a built-in electric field required by conventional PN or type II heterojunctions. This intrinsic capability can surpass the bandgap-limited photovoltage of traditional semiconductor devices, generating outputs well beyond the theoretical threshold^{10,11,12,13,14,15}. Moreover, the spontaneous polarization and ultrafast switching dynamics of ferroelectric materials can emulate synaptic plasticity in biological vision systems, facilitating highly efficient and intelligent neuromorphic visual perception and processing systems¹⁶. Over the past decade, various 2D ferroelectric materials have exhibited outstanding photovoltaic performances^17,18,19. Among them, CuInP₂S₆ (CIPS) stands out as a promising candidate due to its room-temperature ferroelectricity and unique long-range Cu⁺ ions migration^{20,21,22,23,24}, which significantly enhances the BPVE^16,25. Beyond its intrinsic photovoltaic properties of CIPS, constructing asymmetric heterojunction interfaces offers a powerful route to regulate the interface potential barrier, leading to distinct photovoltaic behaviors and regulatory mechanisms. Such tunability not only deepens the mechanistic understanding of photovoltaic technologies, but also paves a new way for the design and optimization of multi-state photoelectronic applications. In particular, CuInP₂S₆/Graphene (CIPS/Gr) heterostructure exhibit intriguing properties, such as large interface barrier tunability and high carrier-regulation capability^26,27, which indicates great application prospects in photovoltaic applications. Recently, theoretical calculations show that this heterostructure exhibits exceptional ferroelectric polarization-adjustable photovoltaic properties²⁸. However, experimental evidence directly linking ferroelectric polarization states to photovoltaic performance in asymmetric CIPS heterostructures is still lacking, and the microscopic mechanisms governing polarization–carrier transport coupling remain to be elucidated.
In this study, the ferroelectric photovoltaic (FE-PV) characteristics of an asymmetric Pt/CuInP₂S₆/Graphene (Pt/CIPS/Gr) heterostructure are systematically investigated. The polarization modulation enhances the photovoltaic current by approximately ten times with a low light intensity. The device exhibits a remarkable voltage polarity-dependent bipolar switching behavior driven by residual polarization. Combining the experimental results with density functional theory (DFT) calculations, we have proposed a physical mechanism based on the synergistic interaction among Cu⁺ ions, interface barriers, and photogenerated charge carriers. Cu⁺ interlayer migration within vdW gaps could regulate photovoltaic performance, thereby inducing the programmable and non-volatile optoelectronic response. Leveraging this capability, the device performs in-sensor computing tasks-including edge detection with recognition rate F-score ~1 and binary pattern classification with 100% accuracy—by directly mapping polarization states to synaptic weights. These findings not only elucidate the FE-PV physics of CIPS-based heterojunctions but also establish a versatile platform for next-generation functional neuromorphic-vision systems.
The designed asymmetric FE-PV heterojunction device features a simple two-terminal structure. A 25 nm Pt bottom electrode was deposited on a SiO₂ (300 nm)/Si substrate via magnetron sputtering, followed by sequential dry transfer of graphene and CIPS to form a vertical Pt/CIPS/Gr structure, as illustrated in Fig. 1a and Supplementary Fig. 1. Structural and optical characterization of CIPS and graphene confirms their suitability for subsequent photoresponse experiments (Supplementary Figs. 2–4). The photovoltaic effect of the device is modulated synergistically by laser illumination and a V_p applied to the bottom Pt electrode. This device is fabricated and applied to emulate the functionality of the human visual system, which could enable bio-inspired operation analogous to retinal photoreceptors. It could convert optical stimuli into neural impulses and realize preprocessing and temporary storage before relaying visual information to the cerebral cortex (Fig. 1b). As the key factor of a FE-PV device, the robust out-of-plane ferroelectricity of CIPS is confirmed by Piezoelectric force microscopy (PFM) measurements (Supplementary Figs. 5 and 6). Figure 1c indicates that the asymmetric interfacial barrier could be modulated by ferroelectric polarization, resulting in a nonlinear dependence of photogenerated carriers. This results in a nonlinear relationship between the scanning voltage and the photocurrent, thereby providing a more comprehensive data set for image edge detection. As shown in Fig. 1d, the anomalous switching effect in the V_p pulse dependence of short-circuit current (I_sc) is observed. It indicates the positive and negative inversion as increasing unidirectional positive polarization, forming the basis for in-device functions such as the binary pattern classification in brain-like neural computing.
a Schematic representation of the structural configuration of our device. b The application of our device as a retinal photoreceptor in cortex computing. It is demonstrated to realize the functionalities of edge detection and pattern classification in this work. c The mechanism of image edge detection. The regulation of interfacial barriers by ferroelectricity affects the transport of photogenerated carriers. It results in a nonlinear relationship between the scanning voltage and the photocurrent, thereby providing a more comprehensive data set in image edge detection. d The mechanism that implements binary pattern classification. The V_p pulse prompts an anomalous switching effect in the I_sc, with the realization of positive and negative inversion as the positive V_p increases, which originates from the Cu⁺ migration.
To characterize the photovoltaic performance of the device, low-power-density laser irradiation was employed. The intrinsic photoelectric and photovoltaic properties of heterojunction devices were systematically investigated. As shown in Fig. 2a, the photocurrent increases steadily with optical power density from 0 to 2.516 mW/cm². Under pulsed illumination at 0.505 mW/cm², the device demonstrates stable photovoltaic sensitivity with a prominent photocurrent response I_light/I_dark ratio of ~900 (Fig. 2b). As shown in Supplementary Fig. 7, the device response remains stable over a testing period of 1000 seconds, which corresponds to 2500 on/off cycles. To systematically characterize neuromorphic-vision characteristics, we have performed measurements of the write kinetics, transient response, read non-destructiveness and −3 dB bandwidth, and noise spectral density, as detailed in Supplementary Figs. 8–10.
a Variation of I-V curves in the range of ±1 V dependent on different optical power densities. b Time dependence of photocurrent characteristic curves stimulated by low-power density optical pulses with 0.505 mW/cm² light intensity. c Typical I-V curves after polarized by V_p = 0 V, V_p = +2 V and V_p = −2 V, respectively for 60 s. d Variation of I-V curves with different V_p for 30 s within the scanning range of ±1 V. e V_p dependence of |I_sc | , revealing the modulated photovoltaic effect by ferroelectric polarization. f Temperature dependence of I_sc after a V_p = −2 V pluse for 60 s under 0.505 mW/cm² light intensity irradiation.
We further investigate ferroelectric modulation of the photoresponsive behavior. As shown in Fig. 2c, the switchable photocurrent varies systematically with the applied V_p, indicating that the photovoltaic effect can be either enhanced or suppressed depending on the electric polarization state of CIPS. A V_p = +2 V pluse reverses the photocurrent to a negative value, whereas a V_p = −2 V pluse significantly enhanced the positive I_sc. To elucidate the influence of remanent polarization on photovoltaic effect in Pt/CIPS/Gr heterojunction, distinct polarization states were induced by applying voltage pulses of opposite polarity. The 60-s polarizing time is intentionally adopted to ensure that Cu⁺ migration reaches a fully stabilized state (Supplementary Fig. 11). Current-voltage (I–V) curves and photovoltaic signals are then measured under 0.505 mW/cm² illumination (Fig. 2d, e). The I–V curve exhibits opposite rectifying behaviors after polarization with negative and positive voltage. As V_p = +2 V is applied, the current in the negative voltage scanning region is relatively higher than that in the positive voltage scanning region, whereas V_p = −2 V produces the reverse trend (Fig. 2d). Notably, the absolute value of short-circuit current ( | I_sc | ) after −2.5 V polarizing is approximately ten times larger than that without polarization (Fig. 2e), and the enhancement under negative V_p is significantly greater than that under positive V_p. These results confirm the generation of an asymmetric polarity-related photocurrent, the origin of which will be discussed later. To further verify the ferroelectric manipulation of the photovoltaic behavior, temperature-dependent photocurrent measurements were performed (Fig. 2f). Below the Curie temperature (T_C), I_sc increases gradually with temperature, whereas above T_C it drops sharply due to the ferroelectric-paraelectric phase transition²⁹. These observations provide additional evidence that the photovoltaic effect in Pt/CIPS/Gr heterojunction is governed by ferroelectricity and help to distinguish the mechanism from photothermoelectric, pyroelectric, Schottky/Dember, trap-mediated photogating or interfacial charging effects at the graphene/CIPS interface.
The nonzero I_sc observed experimentally confirms the existence of the FE-PV effect in heterojunction, which is significantly affected by a forward or reverse polarizing voltage V_p applied to the device. For polyvinylidene fluoride (PVDF) based devices, the external bias towards the polarization direction enhances the current shift caused by the FE-PV effect³⁰, thereby increasing the photoelectric conversion efficiency. However, for CIPS, the external electric field causes the interlayer migration of Cu⁺ ions, which leads to a reversal of the local polarization direction of CIPS, making the correlation of external electric field and photoelectric conversion efficiency more complex. From the traditional perspective, as we apply a V_p pointing towards Gr from CIPS, Cu⁺ ions accumulate at the interface, with the polarization direction pointing towards Gr (P ↑ ) (Fig. 3a). The DFT calculation results show that a classical type II heterojunction is formed for this case, and there is an offset of CIPS conduction band points towards Gr (V_oc ＞ 0) (Fig. 3c and Supplementary Fig. 12). It enhances the transfer of photogenerated hot electrons from CIPS to Gr (Fig. 3b). When a V_p is applied pointing towards CIPS from Gr, the polarization direction points away Gr (P ↓ ) (Fig. 3d) and electrons in Gr transfer to CPIS (V_oc < 0), pushing the Dirac point of Gr to a higher energy level and causing a type II heterojunction with reduced offset (Fig. 3f and Supplementary Fig. 12). Additionally, the conduction band offset of CIPS deviates from Gr, which hinders the transfer of photogenerated hot electrons from CIPS to Gr (Fig. 3e). The polarization reversal caused by the interlayer migration of Cu⁺ ions under an electric field has overturned the traditional FE-PV effect. Therefore, we proposed the model that explains the observed polarization-modulated FE-PV effect in heterojunction by changing the polarization in Fig. 2.
a Schematic structure of Pt/CIPS/Gr device with polarization direction pointing towards Gr (P ↑ ). b Simulation structure of Pt/CIPS/Gr (P ↑ ) heterojunction and the corresponding local density of states (LDOS). c Electronic state diagram of Pt/CIPS/Gr (P ↑ ) interface. The description of (d–f) is the same as that of (a–c), respectively, but for the case of Pt/CIPS/Gr (P ↓ ) heterojunction.
Furthermore, we investigate the kinetic process of polarization switching via the migration of Cu⁺ ions systematically. Figure 4a shows the I-V curves obtained under gradually increased positive polarization, revealing the sign reversals of I_sc and the open-circuit voltage (V_oc). The photovoltaic current changes from positive to negative, and further to positive (Supplementary Fig. 13), while the V_oc varies from negative to positive and further to negative. As no V_p is applied, the curve I in Fig. 4a indicates V_oc <0 and I_sc ＞ 0. When the positive V_p increases, the V_oc shifts towards positive, gradually reaches a maximum value as V_p = +2.0 V. As positive V_p is continuously increased, the V_oc shifts towards negative, and the V_oc crosses zero point again and becomes negative for V_p = +3.0 V. The V_p dependence of photovoltaic current is plotted in Fig. 4b.
a Individual I-V curves of the heterojunction device as V_p is set from 0 V to +3 V in log scale. b Plot of I_sc dependent on positive V_p. c Schematic diagram of the Cu⁺ ions migration mechanism under positive voltage. d Various I-V curves of the heterojunction device as V_p is set from 0 V to −3 V in log scale. e Plot of I_sc dependent on negative V_p. f Schematic diagram of the Cu⁺ ions migration mechanism under negative voltage.
The physical model of kinetic process to explain the anomalous switchable FE-PV effect is set up in Fig. 4c. At zero bias shown in Fig. 4c-Ⅰ, the ordered occupation is downward polarization (P ↓ ) of Cu⁺ ions in CIPS. It leads to the existence of the FE-PV effect and current shift, which results in a nonzero dark current while V_oc ≠ 0. Under a low positive V_p, Cu⁺ ions undergo intralayer migration, causing an upward local polarization (P ↑ ) in CIPS. When the increased polarization cancels out the initial polarization, the overall polarization is zero (P = 0), while both the FE-PV effect and current shift disappear, thus V_oc = 0. As the V_p continues to increase, the overall polarization reverses to the upward polarization (P ↑ ), so V_oc becomes positive. After applying a relatively large V_p, Cu⁺ ions can achieve interlayer migration, and the polarization reverses back to the initial downward polarization (P ↓ ) while V_oc returns to the original state.
Unlike the polarity reversal induced by positive V_p, no sign reversal of the V_oc is observed as increasing negative V_p. This discrepancy originates from the initial ferroelectric state of the heterojunction is downward polarization (P ↓ ). Figure 4d displays the (I–V) curves measured under linearly increasing negative V_p, revealing a distinct drift of the curve apex. The dependence of photovoltaic current on negative V_p is presented in Fig. 4e, with the I_sc states similarly categorized into I, II, III, and IV.
State I corresponds to the initial ferroelectric state with downward polarization (P ↓ ), where V_oc <0, I_sc ＞ 0. As increasing the negative V_p, the majority of Cu⁺ ions migrate downward into the vdW gaps, while a fraction occupies the upper sites in the neighboring layer (indicated by the blue arrow). This generates localized upward polarization (P ↑ ), attenuating the overall downward polarization and consequently reducing V_oc in state II compared to state I. But it cannot completely offset the downward polarization of Cu⁺ ions that initially exist at the intralayer VDW layers. As the negative V_p further increases, more Cu⁺ ions undergo interlayer and intralayer migration to the neighboring layer, locating the downward polarization state (P ↓ ). The resultant increase in the interface potential barrier elevates V_oc (State III). Subsequently, under higher negative V_p, Cu⁺ ions repeat this migration behavior, driving V_oc undergoing a cyclic evolution of weakening—strengthening—weakening. The positive and negative offsets determined by the initial polarization state can be switched through an applied external voltage bias. The reproducibility of this behavior across different devices, measurement cycles, and independent experimental runs is shown in Supplementary Figs. 14–16, certifying that the anomalous switchable FE-PV effect of heterojunction is robust, stable and highly repetitive.
To confirm the existence of the Cu⁺ ion migration process within CIPS, consecutive switching spectroscopy PFM (SS-PFM) hysteresis loops were measured at a fixed location by tuning the tip bias. As shown in Supplementary Fig. 17, by applying a DC bias, the phase exhibits three switching loops with 180° difference. When the positive bias voltage increases, the polarization sequentially switches from P↑ to P↓ and then reverts to P ↑ , completing a polarization switching cycle. In contrast, a negative bias induces a P↑ → P↓ polarization switching. Notably, increasing the bias magnitude can generate a polarization direction antiparallel to the electric field, i.e., a positive bias can generate an abnormal P↓ state, or a negative bias can generate an abnormal P↑ state. It primarily stems from Cu⁺ ions migration within the vdW gap, consistent with the proposed mechanism to reveal the switchable FE-PV effect under unidirectional voltage. In order to prove the Cu⁺ ions migration directly, we have performed cross-sectional scanning electron microscopy–energy dispersive X-ray spectroscopy (SEM-EDS) to visually confirm Cu⁺ ions migration (Supplementary Fig. 18). Specifically, we prepared a bulk CIPS sample with ~15-μm thick, sandwiched between two Au electrodes for in-situ bias tests. The cross-sectional structure is shown in Supplementary Fig. 18a. After applying a 3 V bias pulse for 4 min, we immediately conducted EDS elemental mapping on the cross-section. As shown in Supplementary Fig. 18b, In, P, and S remain uniformly distributed, while Cu exhibits pronounced accumulation toward the cathode side, forming a clear concentration gradient. These results provide direct evidence that Cu⁺ ions migrate and redistribute within CIPS under an applied electric field.
We also fabricated a symmetric Gr/CIPS/Gr device and characterized its photovoltaic properties (Supplementary Fig. 19) under the same measurement conditions as used in Fig. 4. The results reveal a clear contrast in photocurrent behavior between the symmetric structure and the asymmetric Pt/CIPS/Gr device. As the positive or negative polarizing voltage increases gradually from 0 to 3 V, both the short-circuit current (I_sc) and open-circuit voltage (V_oc) of the symmetric Gr/CIPS/Gr device exhibit polarity reversal. In contrast, there is an asymmetric switching behavior in the asymmetric Pt/CIPS/Gr device, manifesting as that I_sc reverses as increasing positive polarizing voltage, but does not reverse with changes in negative bias voltage. To explain the difference in photovoltaic behavior between symmetric and asymmetric structures from an energy-band perspective, we have performed DFT calculations for the Gr/CIPS/Gr configuration (Supplementary Fig. 20). The results clearly show that when both contacts are graphene, their Fermi levels are aligned, and the band bending induced by Cu⁺ ions migration under an external electric field exhibits symmetric characteristics. This comparison strongly indicates that the previously reported anomalous current reversal originates from the structural asymmetry of the device interface contact, rather than the initial ferroelectric state.
Edge detection is a fundamental step in image processing and computer vision. To demonstrate the potential of Pt/CIPS/Gr-based photovoltaic devices for machine vision, we simulated image edge detection using a convolutional neural network (CNN). Figure 5a indicates the schematic illustration of the edge detection process, while a 100 × 100 image showing a petal shape is used as the input image. The pixel values in this image are preprocessed by converting to grayscale, then normalized and binarized to 0 and 1, whose pixel values of black and white are defined as 1 and 0, respectively. When mapping the input image to the FE-PV heterojunction devices, the pixel value of 1 and 0 are represented by applying and removing illumination stimuli of an optical power of 48.3 μW/cm² to the FE-PV heterojunction devices. To implement the convolution process, two 3 × 3 convolution kernels are applied with a stride of 1, where the kernel weights correspond to the photoresponsivity of the FE-PV heterojunction devices (see calculation method in Supplementary Note 1). This responsivity is directly modulated via ferroelectric polarization by adjusting pulse duration, pulse amplitude or light intensity that is applied to the FE-PV heterojunction devices. Figure 5b shows two Prewitt kernels, where the kernel 1 extracts the gradient in x direction (G_x) and the kernel 2 extracts the gradient in y direction (G_y), respectively. The output current I_x and I_y related to kernel 1 and 2 fully captures the G_x and G_y features, respectively, clearly extracting the edges between the petal and the background along the x and y direction, as shown in Fig. 5c, d. These two sets of output currents I_x and I_y can then be merged to find the absolute magnitude of the gradient at each pixel point (Fig. 5e). After these processes, the current values are further normalized and binarized to obtain the final output image (Supplementary Note 2). The edges of the input image are successfully detected in Fig. 5c–e. According to Fig. 5f, the F-score, which is a figure-of-merit in terms of true and false positives and negatives, is calculated to be 1, showing good performance of edge detection by using our heterostructure devices (Supplementary Fig. 23). Moreover, the symmetry of photoresponsivity weight can be improved by a programming algorithm to increase the photoresponsivity accuracy³¹. As the identical values of positive and negative photoresponsivity for +1 and −1 are programmed along with near-zero photoresponsivity for 0, the nonzero value of output currents beyond the detected edges could be eliminated (Supplementary Figs. 24 and 25). Details regarding reproducibility, preprocessing, and post‑processing associated with the image edge detection are summarized in Supplementary Figs. 26–28.
a Schematic illustration of the edge detecting process. b From left to right: theoretical dimensionless weights, photoresponsivity weights in units of mA/W. c, d Edge gradient components for x direction (I_x) and y direction (I_y), respectively. e Output result after merging I_x and I_y. f Final output after normalization and binarization showing the detected edge.
Pattern classification is another fundamental task in computer vision and machine learning. In Fig. 6a, two pattern sets representing the letters “X” and “T”-along with their noisy variants are used, each consisting of a 3 × 3 pixel array³². The training and test sets are identical, containing 50 patterns with possible repetitions. For each pattern, the pixel values of the body pattern pixels are defined as logical +1, while the pixel values of the background pixels are defined as logical −1. The sets of “X” and “T” correspond to the binary outputs of y = +1 and y = −1, respectively. A single-layer perceptron can be hardware-implemented with the FE-PV heterojunction devices (see Supplementary Note 3).
a Two sets of patterns representing letters “X” and “T’. b From left to right: theoretical dimensionless weights, photoresponsivity weights in units of mA/W. c Output current when classifying different input patterns.
The classification weights w that yields correct classification can be obtained through the training process in the framework of an artificial neural network. Herein, we adopt the so-called ex situ method, the training is performed in software. As shown in Fig. 6b, the theoretical dimensionless weights +1 and −1 are programmed to the positive and negative photoresponsivity values of 9.5 and −10.5 mA/W, while the calculated weights +2 and −2 are programmed to the positive and negative photoresponsivity values of 19.3 and −21.5 mA/W, respectively. Using the MAC operation in a way like that used for edge detection, Fig. 6c presents the output current I₁ during the inference of different input patterns, by using an illumination source with an optical power of 2.156 mW/cm² to the FE-PV heterojunction devices. A pattern from the “X” class results in a positive output current, while a pattern from the “T” class results in a negative output current. Like that in the edge detection, the asymmetry for positive and negative photoresponsivity weights still exists for both ±1 and ±2. Moreover, the ratio of photo responsivities corresponding to dimensionless 1 and 2 are slightly derived from 2. Despite the existence of nonideal factors, all the patterns are correctly separated, and thus the accuracy for this binary classification task is 100%.
In conclusion, we have achieved a switchable photovoltaic effect in a Pt/CuInP₂S₆/Graphene asymmetric heterojunction through ferroelectric polarization modulation. The device exhibits polarity-dependent rectifying behavior and a distinct positive-to-negative reversal of the photovoltaic current. DFT calculations reveal that this anomalous FE-PV effect originates from the coupling of Cu⁺ ion migration and interface barrier regulation. Furthermore, we demonstrate that the asymmetric heterojunction could be used in image processing, such as image edge detection and pattern classification. This work provides deep mechanistic insights and design strategies for exploiting advantageous heterojunction devices in next-generation neuromorphic and visual technologies.
CIPS single crystals were synthesized by the chemical vapor transport method³³. Stoichiometric amounts of high-purity Cu, In, P, and S powder were sealed into a vacuumed quartz tube and loaded into a dual-zone horizontal tube furnace. Initially, the hot and cold zones were maintained at 665 and 635 °C, respectively, for 34 h. The temperature gradient was then inverted, with the cold zone heated to 665 °C and the hot zone lowered to 635 °C for an additional 196 h. Following this thermal treatment, bulk CIPS single crystals were successfully obtained. Raman spectroscopy is utilized to confirm the high-quality crystal structure of the CIPS layer. Furthermore, a UV-vis absorption spectrum is obtained to characterize the optical absorption of CIPS.
The thickness of CIPS and Graphene, measured via atomic force microscope (AFM) measurements (Bruker, Dimension Icon), is ~26 and 42 nm, respectively (Supplementary Fig. 2). Furthermore, Raman (Horiba, XploRA PLUS) spectroscopy of the CIPS (Supplementary Fig. 3) reveals characteristic peaks at 101, 162, 262, 315, 375, and 445 cm⁻¹, which is assigned to P₂S₆ librations, the δ(S-P-P) mode, the δ(S-P-S) mode, cation oscillations, the δ(P-P) mode, and the δ(P-S) mode, respectively³⁴. This demonstrates that the materials possess a well-defined crystal structure. To characterize the optical absorption of CIPS, UV-vis absorption spectra (Shimadzu, UV-3600)  were measured. As shown in Supplementary Fig. 4, the absorption edge (λ_g) is determined to be 477 nm. Therefore, a 450 nm laser was selected for all subsequent experiments to stimulate the photoresponse of the heterojunction device.
Following the patterns printed onto a SiO₂ (300 nm)/Si substrate by photolithography, a 25 nm Pt electrode was deposited via the ultra-vacuum magnetron sputtering system, with a deposition rate of 0.1 nm/s. To minimize contamination, substrates were ultrasonically cleaned with ethanol and acetone before lithography. Photoresist was prebaked for 3 min to avoid charring, and electrodes were cleaned with fresh acetone to eliminate the residual photoresist. Subsequently, graphene and CuInP₂S₆ (CIPS) thin layers are obtained through mechanical exfoliation using scotch tape and attached to PDMS films, respectively. Then, under an optical microscope, we use a precise micro-manipulation system to precisely stack the target material in thin layers on the predefined electrode positions. Only flakes with smooth surfaces, clean edges, and uniform thickness were used. Graphene is used as the upper electrode by connecting a parallel Pt electrode on the periphery using a sputtering system. A vertical and asymmetrical Pt/CuInP₂S₆/Graphene heterojunction device was fabricated. The thickness of CIPS and Graphene is measured via an atomic force microscope (AFM).
Each heterojunction was optically inspected post-transfer. Only devices with uniform, bubble-free interfaces and full substrate contact were selected, ensuring the high quality of interfaces. Through statistics of approximately more than 100 instances of device fabrication experiments, the final transfer success rate (the proportion of complete heterojunction devices successfully fabricated and available for electrical testing) is more than 85%, providing a basis for the reproducibility of the research.
The FE-PV characterizations of the heterojunction device were obtained with a semiconductor characterization system (Keithley B1500A). Photovoltaic response measurements were conducted under a 450 nm wavelength optical excitation, with an optical power density ranging from 0.505 mW/cm² to 2.516 mW/cm². The incident laser power was precisely calibrated using a commercial optical power meter. Additionally, the time-dependent short-circuit current (I_sc) under zero bias voltage was obtained. Under 405 nm laser illumination, a high density of photocarriers was generated, causing I_sc to rapidly increase from the noise level (8 × 10⁻¹⁴ A). The specific measurement procedure for the FE-PV effect was as follows: In the dark, a polarizing voltage (V_p) was applied to the bottom Pt electrode to establish the polarization state, while the top graphene electrode was grounded. Subsequently, under laser illumination, the FE-PV characteristics were obtained by performing I–V scans within a ±1 V range. Furthermore, the I-V characteristics could be modulated by varying the magnitude, polarity, and duration of the applied V_p.
All calculations in this work are performed based on the density functional theory (DFT) within the Vienna ab initio simulation package (VASP)^35,36,37, with the projector augmented wave (PAW) potentials³⁸ to describe the electron-ionic core interaction. Based on the previous parameter test³⁹, we chose the Perdew–Burke–Ernzerhof formulation (PBE)⁴⁰ of the generalized gradient approximation (GGA) to describe the exchange-correlation interaction of electrons, and the DFT-D3 (Becke–Johnson) formulation⁴¹ was used to describe vdW interactions between layers. For structural relaxation of individual two-dimensional material, the wave functions are expanded in a plane-wave basis set with an energy cutoff of 650 eV for structural relaxation. The force on each ion is converged to be less than 0.01 eV/Å, and a precision of 10⁻⁶ eV is adopted to minimize the total energy of the system. The k spacing of 0.02 Å⁻¹ grid in reciprocal space is used to ensure the convergence for the total energy self-consistent calculations. The Cu 3p⁶3d¹⁰4s¹, In 4d¹⁰5s²5p¹, P 3s²3p³, S 3s²3p⁴, and C 2s²2p² electrons are treated as valence electrons. VESTA software is used for visualization of crystal structures⁴². The experimental lattice parameters are used as the initial structure, and after fully relaxation, including both the external (lattice constants and including angles) and internal (atomic coordinates) parameters, the lattice parameters are a = 4.303 Å, b = 3.189 Å, γ = 90° for orthorhombic Graphene, and a = 6.091 Å, b = 10.546 Å, c = 13.788 Å, β = 107.27° for CIPS, agreement with the experimental lattice parameters (a = 6.096 Å, b = 10.565 Å, c = 13.623 Å, β = 107.10°)²⁰. Detailed information is available at 4.
In mapping the input image to the FE-PV heterojunction devices, pixel values of 1 and 0 correspond to the application and removal of 48.3  μW/cm² illumination. Convolution is performed by sliding 3 × 3 kernels (e.g., Prewitt kernels for edge detection) over the image with a stride of 1. Kernel weights are mapped to the photoresponsivity (R_kl) of each device, adjustable through ferroelectric polarization and varying light intensity. The fundamental multiply-accumulate (MAC) operation of convolution is physically realized by summing, according to Kirchhoff’s law, the photocurrents generated from the illuminated 3 × 3 device group (Supplementary Note 5). For edge detection, convolution outputs are merged, normalized to the 0–1 range, and converted to binary using a threshold of 0.5. Performance is evaluated using the F-score. Pattern classification employs a single-layer perceptron: A flattened 3 × 3 pixel array with bias input feeds into an output neuron that applies a sign function to weighted inputs. Weights are trained externally using the perceptron learning rule on datasets, achieving perfect test accuracy. Detailed information is available at Supplementary Notes 2 and 3.
The authors declare that all data supporting the findings of this study are contained within the article and its supplementary information. Additional raw data are available from the corresponding author upon reasonable request.
Pei, Y. et al. Artificial visual perception nervous system based on low-dimensional material photoelectric memristors. ACS Nano 15, 17319–17326 (2021).
Article  CAS  PubMed  Google Scholar 
Zhang, X. et al. An artificial spiking afferent nerve based on Mott memristors for neurorobotics. Nat. Commun. 11, 51 (2020).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Yan, X. et al. Vacancy-induced synaptic behavior in 2D WS₂ nanosheet–based memristor for low-power neuromorphic computing. Small 15, 1901423 (2019).
Article  Google Scholar 
Zhu, Q.-B. et al. A flexible ultrasensitive optoelectronic sensor array for neuromorphic vision systems. Nat. Commun. 12, 1798 (2021).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Lee, D. et al. In-sensor image memorization and encoding via optical neurons for bio-stimulus domain reduction toward visual cognitive processing. Nat. Commun. 13, 5223 (2022).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Prezioso, M. et al. Training and operation of an integrated neuromorphic network based on metal-oxide memristors. Nature 521, 61–64 (2015).
Article  ADS  CAS  PubMed  Google Scholar 
Wang, Z. et al. Fully memristive neural networks for pattern classification with unsupervised learning. Nat. Electron. 1, 137–145 (2018).
Article  Google Scholar 
Yao, P. et al. Fully hardware-implemented memristor convolutional neural network. Nature 577, 641–646 (2020).
Article  ADS  CAS  PubMed  Google Scholar 
Pal, S., Sarath, N. V., Priya, K. S. & Murugavel, P. A review on ferroelectric systems for next generation photovoltaic applications. J. Phys. D: Appl. Phys. 55, 283001 (2022).
Article  CAS  Google Scholar 
Wan, S. et al. Room-temperature ferroelectricity and a switchable diode effect in two-dimensional α-In₂Se₃ thin layers. Nanoscale 10, 14885–14892 (2018).
Article  CAS  PubMed  Google Scholar 
Yuan, S. et al. Room-temperature ferroelectricity in MoTe₂ down to the atomic monolayer limit. Nat. Commun. 10, 1775 (2019).
Article  ADS  PubMed  PubMed Central  Google Scholar 
Zhou, Y. et al. Out-of-plane piezoelectricity and ferroelectricity in layered α-In₂Se₃ nanoflakes. Nano Lett. 17, 5508–5513 (2017).
Article  ADS  CAS  PubMed  Google Scholar 
Zhang, Y. J. et al. Enhanced intrinsic photovoltaic effect in tungsten disulfide nanotubes. Nature 570, 349–353 (2019).
Article  ADS  CAS  PubMed  Google Scholar 
Guan, Z. et al. Recent progress in two-dimensional ferroelectric materials. Adv. Electron. Mater. 6, 1900818 (2020).
Article  CAS  Google Scholar 
Qi, L., Ruan, S. & Zeng, Y.-J. Review on recent developments in 2D ferroelectrics: theories and applications. Adv. Mater. 33, 2005098 (2021).
Article  CAS  Google Scholar 
Liang, L. et al. Configurable kinetics of polarization switching via ion migration in ferroionic CuInP2S6. Nat. Commun. 16, 4462 (2025).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Yang, M.-M., Kim, D. J. & Alexe, M. Flexo-photovoltaic effect. Science 360, 904–907 (2018).
Article  ADS  CAS  PubMed  Google Scholar 
Yang, S. Y. et al. Above-bandgap voltages from ferroelectric photovoltaic devices. Nat. Nanotechnol. 5, 143–147 (2010).
Article  ADS  CAS  PubMed  Google Scholar 
Han, X., Ji, Y. & Yang, Y. Ferroelectric photovoltaic materials and devices. Adv. Funct. Mater. 32, 2109625 (2022).
Article  CAS  Google Scholar 
Maisonneuve, V., Cajipe, V. B., Simon, A., Von Der Muhll, R. & Ravez, J. Ferrielectric ordering in lamellar CuInP₂S₆. Phys. Rev. B 56, 10860–10868 (1997).
Article  ADS  CAS  Google Scholar 
You, L. et al. Origin of giant negative piezoelectricity in a layered van der Waals ferroelectric. Sci. Adv. 5, eaav3780 (2019).
Brehm, J. A. et al. Tunable quadruple-well ferroelectric van der Waals crystals. Nat. Mater. 19, 43–48 (2020).
Article  ADS  CAS  PubMed  Google Scholar 
Zhou, S. et al. Anomalous polarization switching and permanent retention in a ferroelectric ionic conductor. Mater. Horiz. 7, 263–274 (2020).
Article  CAS  Google Scholar 
Neumayer, S. M. et al. Ionic control over ferroelectricity in 2D layered van der Waals capacitors. ACS Appl. Mater. Interfaces 14, 3018–3026 (2022).
Article  CAS  PubMed  Google Scholar 
Li, Y. et al. Enhanced bulk photovoltaic effect in two-dimensional ferroelectric CuInP₂S₆. Nat. Commun. 12, 5896 (2021).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Wu, J. et al. High tunnelling electroresistance in a ferroelectric van der Waals heterojunction via giant barrier height modulation. Nat. Electron. 3, 466–472 (2020).
Article  Google Scholar 
Zhong, Z. et al. Robust threshold-switching behavior assisted by Cu migration in a ferroionic CuInP₂S₆ heterostructure. ACS Nano 17, 12563–12572 (2023).
Article  CAS  PubMed  Google Scholar 
Yang, K. et al. Interfacial polarization enhanced ultrafast carrier dynamics in ferroelectric CuInP₂S₆. Nano Lett. 25, 1890–1897 (2025).
Article  ADS  CAS  PubMed  Google Scholar 
Xu, T.-X. et al. Emerging van der Waals material CuInP₂S₆: physical properties, theories and applications. J. Mater. Chem. A 12, 31028–31058 (2024).
Article  ADS  CAS  Google Scholar 
Wu, S. et al. Polarization photodetectors with configurable polarity transition enabled by programmable ferroelectric-doping patterns. Nat. Commun. 15, 8743 (2024).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Alibart, F., Gao, L., Hoskins, B. D. & Strukov, D. B. High precision tuning of state for memristive devices by adaptable variation-tolerant algorithm. Nanotechnology 23, 075201 (2012).
Article  ADS  PubMed  Google Scholar 
Hertz, J., Krogh, A., Palmer, R. G. & Horner, H. Introduction to the theory of neural computation. Phys. Today 44, 70 (1991).
Article  ADS  Google Scholar 
Liu, F. et al. Room-temperature ferroelectricity in CuInP₂S₆ ultrathin flakes. Nat. Commun. 7, 12357 (2016).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Liu, Y. et al. CuInP₂S₆-based electronic/optoelectronic synapse for artificial visual system application. Adv. Funct. Mater. 34, 2306945 (2024).
Article  CAS  Google Scholar 
Kresse, G. & Hafner, J. Ab initio molecular dynamics for liquid metals. Phys. Rev. B 47, 558–561 (1993).
Article  ADS  CAS  Google Scholar 
Kresse, G. & Furthmüller, J. Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set. Comput. Mater. Sci. 6, 15–50 (1996).
Article  CAS  Google Scholar 
Kresse, G. & Furthmüller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169–11186 (1996).
Article  ADS  CAS  Google Scholar 
Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave method. Phys. Rev. B 59, 1758–1775 (1999).
Article  ADS  CAS  Google Scholar 
Jiang, X. et al. Dual-role ion dynamics in ferroionic CuInP₂S₆: revealing the transition from ferroelectric to ionic switching mechanisms. Nat. Commun. 15, 10822 (2024).
Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 
Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple. Phys. Rev. Lett. 77, 3865–3868 (1996).
Article  ADS  CAS  PubMed  Google Scholar 
Grimme, S., Ehrlich, S. & Goerigk, L. Effect of the damping function in dispersion corrected density functional theory. J. Computat. Chem. 32, 1456–1465 (2011).
Article  ADS  CAS  Google Scholar 
Momma, K. & Izumi, F. VESTA: a three-dimensional visualization system for electronic and structural analysis. J. Appl. Crystallogr. 41, 653–658 (2008).
Article  ADS  CAS  Google Scholar 
Download references
X.L.W. acknowledges the support of the National Natural Science Foundation of China (Grant No. 12474101). X.Y.W. acknowledges the support by the National Natural Science Foundation of China (Grant Nos. 12374080 and 92163101). P.Y.C. acknowledges the support by Beijing Natural Science Foundation (Grant No. 4232061), National Natural Science Foundation of China (Grant No. 62574020), and Beijing Municipal Education Commission Fund (Grant No. BPHR202203024). L.L.W. acknowledges the support of the National Natural Science Foundation of China (NSFC Grant No. 62422409). Y.Y.X. acknowledges the support by Beijing Natural Science Foundation (Z220005) and Beijing Outstanding Young Scientist Program (JWZQ20240102009). Z.J.Z. acknowledges the support by the National Natural Science Foundation of China (No. 52501307) and Beijing Natural Science Foundation (No. 2254085). X.A.J. acknowledges the support of the Ningbo Yongjiang Talent Introduction Program (2023A-390-G).
Zunyi Deng
Present address: Engineering Research Center of Integrated Circuit Packaging and Testing, Ministry of Education, Gansu Integrated Circuit Packaging and Testing Industry Research Institute, Tianshui Normal University, Tianshui, P. R. China
These authors contributed equally: Meijiao Men, Zunyi Deng, Zijing Zhao.
School of Physics and Optoelectronic Engineering, Beijing University of Technology, Beijing, China
Meijiao Men, Zijing Zhao & Xiaolei Wang
School of Aerospace Engineering & State Key Laboratory of Environment Characteristics and Effects for Near-space, Beijing institute of technology, Beijing, China
Zunyi Deng, Yuanyuan Cui, Shuaizhao Jin, Jiawang Hong & Xueyun Wang
State Key Laboratory of Semiconductor Physics and Chip Technologies, Institute of Semiconductors, Chinese Academy of Sciences, Beijing, China
Lingchen Liu & Lili Wang
Key Laboratory of Optoelectronics Technology, Ministry of Education, Beijing University of Technology, Beijing, China
Yiyang Xie & Pengying Chang
Institute of Micro/Nano Materials and Devices, Ningbo University of Technology, Ningbo, China
Xingan Jiang
Institute of Fundamental and Frontier Sciences, University of Electronic Science and Technology of China, Chengdu, China
Wuyang Ren & Jiang Wu
Department of Physics and Astronomy, Rutgers University, Piscataway, NJ, USA
Sang-Wook Cheong
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
Search author on:PubMed Google Scholar
X.L.W., X.Y.W., L.L.W., and Y.Y.X. conceived the idea and supervised the research work. M.J.M. developed the sample fabrication. M.J.M., L.C.L., W.Y.R., and J.W. characterized photovoltaic performance. Y.Y.C., S.Z.J., and X.A.J. proceeded with PFM measurements. Z.Y.D. performed the DFT calculation. P.Y.C. and Y.Y.X. performed the in-sensor computing task calculation. M.J.M., Z.J.D., and P.Y.C. wrote the manuscript. X.L.W., Z.J.Z., X.Y.W., L.L.W., J.W.H., and S.W.C. revised the manuscript. All authors discussed the results and contributed to the final manuscript.
Correspondence to Pengying Chang, Lili Wang, Xueyun Wang or Xiaolei Wang.
The authors declare no competing interests.
Nature Communications thanks Soheil Ghods, Rajesh Ulaganathan and Hao Wang for their contribution to the peer review of this work. A peer review file is available.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissions
Men, M., Deng, Z., Zhao, Z. et al. Polarization-modulated programmable photovoltaic performance of a designed ferroelectric heterojunction. Nat Commun 17, 2096 (2026). https://doi.org/10.1038/s41467-026-68853-y
Download citation
Received: 29 August 2025
Accepted: 19 January 2026
Published: 28 January 2026
Version of record: 02 March 2026
DOI: https://doi.org/10.1038/s41467-026-68853-y
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
Nature Communications (Nat Commun)
ISSN 2041-1723 (online)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source