Self-Adaptive Infrared Vision via Neural-Controlled Gain Compression in a Single Photodetector

Self-Adaptive Infrared Vision via Neural-Controlled Gain Compression in a Single Photodetector | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Self-Adaptive Infrared Vision via Neural-Controlled Gain Compression in a Single Photodetector Guanhai Li, Yuxin Song, Guanhai Li, Junzhe Gu, Jin Chen, Feilong Yu, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7289222/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Biological vision relies on eye-mediated gain control to adapt across lighting conditions—but remains fundamentally blind to infrared wavelengths and polarization. Here, we report a neuromorphic photodetector that not only emulates this self-adaptive functionality, but surpasses human vision by enabling dynamic gain regulation across the infrared–polarization domain. Using a gate-tunable Au/BP/PdSe 2 van der Waals heterostructure (vdWH), we achieve eye-like nonlinear gain compression via electrostatic barrier reconfiguration, which enables dynamic modulation of both the response area and responsivity. Integrated with a neural-network-based microcontroller, the system forms a device-level closed-loop that autonomously adjusts optical gain in real time. This expands the linear dynamic range (LDR) by three orders of magnitude, reaching ~ 80 dB at 1550 nm, with sub-millisecond response and intrinsic polarization sensitivity (PR ≈ 8)—all without external optics or analog circuitry. These results establish a scalable, intelligent optoelectronic platform that augments biological perception and advances chip-scale self-adaptive vision for autonomous sensing and edge photonic intelligence. Physical sciences/Optics and photonics/Applied optics/Optical sensors Physical sciences/Optics and photonics/Applied optics/Optoelectronic devices and components Self-adaptive photodetector Van der Waals heterostructure Electrostatic barrier engineering Nonlinear response Neural control Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction Human vision excels at self-adapting to dynamic lighting through pupil control of light intake and retina control of photoreceptors activation. Yet, despite its sophistication, biological vision remains blind to infrared wavelengths and insensitive to polarization—two critical modalities for emerging applications in autonomous sensing, edge intelligence, and human–machine interaction. 1 – 3 Bridging this gap demands vision hardware that not only self-adapts in real time, but also extends beyond the spectral and polarization boundaries of human perception. 4 – 8 Existing image sensors rely heavily on system-level compensation—such as mechanical apertures, analog gain control, or digital post-processing—to mitigate underexposure or overexposure under variable illumination. 9 , 10 While effective in controlled environments, these solutions introduce latency, increase power consumption, and hinder integration, making them unsuitable for next-generation autonomous platforms where fast, compact, and self-adaptive sensory systems are essential. 11 – 14 To address this challenge, there is a growing need for solid-state photodetectors that exhibit intrinsic, reversible optical gain modulation—capable of continuously compressing LDR without relying on external optics or auxiliary electronics. Two-dimensional (2D) material heterostructures have emerged as promising candidates, offering broadband responsivity, structural flexibility, and scalable integrability. 15 – 31 However, most 2D photodetectors still exhibit static gain responses and saturate under strong illumination, limiting their applicability in real-world, high-contrast scenarios. 32 , 33 Moreover, previously reported adaptation schemes remain either passive, externally driven, or non-generalizable, lacking the device-level intelligence and closed-loop control necessary for true autonomous vision. 34 – 42 Here, we demonstrate a neuromorphic photodetector that not only emulates the self-adaptive functionality of the human eye, but surpasses it by enabling dynamic gain regulation across infrared and polarization domains— regime fundamentally inaccessible to biological systems. Our design is based on a gate-tunable Au/BP/PdSe 2 vdWH, wherein electrostatically reconfigurable barriers enable continuous and reversible modulation of both response area and responsivity. By coupling this mechanism with an embedded neural-network-based microcontroller, we realize a hardware-in-the-loop closed feedback system that autonomously adjusts optical gain in real time. This architecture expands LDR by three orders of magnitude (~ 80 dB at 1550 nm), achieves sub-millisecond response, and offers intrinsic polarization sensitivity (PR ≈ 8)—all without external optics, analog amplifiers, or post-processing. This work introduces a scalable optoelectronic platform that tightly integrates physical tunability with embedded intelligence, offering a pathway toward chip-scale, self-adaptive vision systems that augment and exceed biological perception. Results Biological visual systems achieve robust perception across diverse lighting conditions via low-latency feedback loops between the pupil and retina (Fig. 1 a). The pupil adjusts light intake by changing its aperture, while retina regulate responsivity by altering the degree of photoreceptors activation to maintain consistent contrast and expand LDR. To emulate this dual-modulation strategy on a chip, we design a gate-tunable Au/BP/PdSe 2 vdWH that supports real-time, device-level gain control (Fig. 1 b). The device comprises a BP flake sandwiched between a gold electrode (bottom) and a PdSe 2 layer (top), forming an asymmetric vertical heterojunction. This vertical stacking enables electrostatic barrier engineering via a single back-gate terminal (V g ). Under weak illumination, the Au/BP interface—with its larger space charge region—supports stronger field-assisted carrier separation and higher responsivity. Under strong illumination, the BP/PdSe 2 interface becomes dominant, but reduced space charge region and field strength lead to suppressed responsivity. This intrinsic asymmetry enables dynamic reallocation of the photoactive region, analogous to how pupil diameter and retinal photoreceptors activation shift in response to ambient light. Due to the atomically thin nature of 2D materials, the Fermi level of BP can be finely tuned by the gate voltage, which modulates both the spatial extent of the active region and the local responsivity (Fig. 1 c). These two parameters—response area (A) and responsivity (R)—together determine the net optical gain and allow continuous compression of the photocurrent response, all within a monolithic structure and without external gain control modules. As illustrated in Fig. 1 d, the band alignment and field distribution evolve significantly under different gate biases, enabling active control over the internal gain landscape. This barrier reconfiguration serves as the physical basis for eye-like nonlinear gain compression, offering high LDR operation without external optics or multi-pixel architectures. To close the loop, we integrate the photodetector with a microcontroller-based neural network (Fig. 1 e). A FCNN infers the required gate voltage in real time based on the incident light intensity and the desired output current. The inferred voltage is then applied to the device to stabilize the output across variable lighting—mimicking the closed-loop self-adaptation of biological vision, but extending it into the infrared and polarization domain. The self-adaptive behavior of our photodetector arises from the asymmetric band alignment and electrostatic reconfigurability of the Au/BP/PdSe 2 vertical heterostructure. As shown in Fig. 2 a, pre-contact energy diagrams indicate built-in Schottky barriers at both metal–semiconductor (Au/BP) and semiconductor–semimetal (BP/PdSe 2 ) interfaces. The larger barrier at the Au/BP contact (~ 0.23 eV) provides rectifying behavior and facilitates unidirectional photocarrier separation, while the smaller barrier (~ 0.17 eV) at the BP/PdSe 2 junction offers a secondary charge extraction path under gate modulation. AFM measurements confirm the layered architecture (Fig. 2 b), with well-controlled flake thicknesses that ensure vertical field coupling while preserving interfacial sharpness and device reproducibility. The device is encapsulated with hBN to maintain air stability during ambient operation. These structural features establish a robust physical platform for field-driven gain control. The dark state characterization test results of the Au/BP/PdSe 2 device are shown in Supplementary Figures S1 and S2, these results establish a robust electrostatic tuning framework that underpins the dynamic gain modulation demonstrated in the self-adaptive photodetection system. The transfer characteristics in Fig. 2 c show strong gate-tunable photocurrent enhancement at zero source-drain bias. Illumination induces nonlinear amplification that intensifies with increasing gate voltage, indicating field-assisted carrier separation and enhanced band bending. This nonlinear gating behavior underpins the gain compressive response required for wide LDR operation and reflects the intrinsic self-adaptive nature of the device. To visualize how the absorption zone and gain evolve with gating, we performed spatial photocurrent mapping under 1550 nm excitation (Figs. 2 d–i). At V g = 0 V, the photocurrent is localized near the Au/BP interface, where the built-in field efficiently separates photocarriers. At + 40 V, this region expands due to increased Schottky barrier height and depletion width, enhancing responsivity—functionally akin to the eye reflex under dim light. In contrast, applying − 40 V shifts the junction field toward the BP/PdSe 2 side. However, PdSe 2 screens the BP layer from most excitation. Combined with reduced response area and responsivity, this results in significantly reduced photocurrent and earlier saturation—analogous to the eye reflex under strong light to suppress overexposure. These results demonstrate that both the spatial origin and the magnitude of photoresponse are continuously programmable through gate voltage, forming a tunable gain landscape that self-adapts responsively to varying illumination. This electrostatic control serves as the physical foundation for device-intrinsic, eye-like gain compression. The full band evolution under illumination is shown in Supplementary Figure S3. To evaluate the functional capabilities of the Au/BP/PdSe 2 heterostructure, we characterize its optoelectronic performance under near-infrared illumination. Figure 3a shows the output current–voltage (I–V) characteristics under various optical powers at 1550 nm. The observed photovoltaic shift—with increasing short-circuit current (I sc ) and open-circuit voltage (V oc )—confirms Schottky junction-based carrier separation, consistent with the internal field distribution in our asymmetric device. For practical sensing and imaging, fast and reliable photoswitching is critical. As shown in Fig. 3b and Supplementary Figure S4, the device demonstrates robust on–off behavior across a wide power range, with on/off ratios exceeding 87 and sharp rise/fall edges. Time-resolved measurements (Supplementary Figures S5 and S6) reveal sub-millisecond response speeds, attributed to the built-in field and high carrier mobility of BP, highlighting compatibility with video-rate applications. Other optoelectronic test results are shown in Supplementary Figures S7-S10, demonstrating the excellent optoelectronic properties of the device. Responsivity measurements at zero source-drain bias (Fig. 3c) show a peak responsivity of 45,260 V W − 1 at V g = 40 V and P = 0.62 mW mm − 2 . The detectivity (D * ) reaches 5×10 9 cm Hz 1/2 W − 1 (Supplementary Figure S11), arising from the combined effects of strong photovoltage generation, low dark current, and efficient electrostatic control. To assess LDR regulation, we analyze the power-dependent photocurrent behavior across gate voltages. As shown in Supplementary Figure S12, the photocurrent follows a power-law form (I ds ∝ P α ), where the exponent α increases with gate bias due to enhanced field-assisted separation and expanded depletion width (Supplementary Figure S13). This tunable nonlinearity enables photocurrent compression under strong illumination, emulating the eye reflex to prevent overexposure. Critically, by integrating real-time gate adjustment through our neural control loop, this nonlinear response is reshaped into a strictly linear profile spanning four orders of magnitude, corresponding to a LDR of ~ 80 dB (Fig. 3d). Compared to fixed-gate operation, this represents a three-order-of-magnitude LDR expansion, without external optics or analog gain modules. Figure 3e benchmarks our performance against representative adaptive photodetectors across spectral range and LDR compression capabilities. 32 , 35 , 43 – 51 The Au/BP/PdSe 2 system outperforms previously reported platforms in both near-infrared responsivity and LDR range, particularly within the 1550 nm window relevant to telecommunications and eye-safe imaging. Beyond intensity self-adaptation, the detector also exhibits intrinsic polarization sensitivity due to the in-plane anisotropy of BP. 52 , 53 As shown in Fig. 3f, the photocurrent varies with polarization angle, while maintaining a stable polarization ratio (PR ≈ 8) across gate biases. This anisotropic response enables full-Stokes detection and supports multifunctional imaging applications (Supplementary Figures S14, S15). Together, these results establish the Au/BP/PdSe 2 heterostructure as a multifunctional, neuromorphic photodetector capable of real-time, self-adaptive gain control, high-speed response, and polarization-resolved sensing—offering a compact, scalable platform for chip-scale artificial vision and edge photonic intelligence. To validate real-time self-adaptive vision based on our device-level gain tuning mechanism, we develop a single-pixel imaging system that couples the photodetector with an embedded neural controller (Fig. 4a and Supplementary Figure S16). This setup emulates the biological eye’s reflex arc, where light perception and gate modulation are connected through a closed feedback loop. The self-adaptive control logic operates in two steps (Fig. 4b): (1) initial illumination estimation is performed by measuring the photocurrent at V g = 0 V; (2) the inferred light intensity is used to calculate and apply the gate voltage needed to achieve the target photocurrent output. The entire loop runs in real time and is implemented fully in hardware, with convergence achieved within milliseconds. Importantly, data acquisition is decoupled from exposure integration, allowing seamless feedback during scanning—a key requirement for dynamic imaging. Figure 4c maps the relationship between gate voltage and incident power required to maintain constant output current. The inverse nonlinear trend confirms that our device architecture supports continuous, autonomous gain regulation across wide illumination levels—without the need for external optics or analog circuitry. The photocurrent is sampled and digitized by a signal collector, then fed into an STM32 microcontroller executing a pre-trained FCNN (Fig. 4d). The FCNN is trained on experimental data linking photocurrent, light intensity, and gate voltage, using Bayesian regularization to enhance generalization and mitigate overfitting. Once trained, it performs real-time inference to predict the optimal gate voltage based on the measured photocurrent and desired target current (Fig. 4e). To demonstrate the practical advantage of this self-adaptive platform, we conduct single-pixel imaging under varying optical powers. As shown in Fig. 4f, the fixed-gate configuration (left panels) leads to severe underexposure or overexposure depending on illumination conditions. In contrast, the self-adaptive configuration (right panel) maintains uniform contrast and high-fidelity image reproduction across all scenarios. This hardware-in-the-loop architecture—integrating gate-tunable materials, neuromorphic inference, and photodetector physics—realizes a fully autonomous exposure regulation system, functionally analogous to the biological eye reflex, yet extending well beyond it in spectral and polarization domains. Our results provide a blueprint for scalable, chip-level self-adaptive vision, enabling future applications in neuromorphic sensing, intelligent robotics, and edge optical intelligence. Conclusion In summary, we report a neuromorphic photodetector system that emulates and surpasses biological vision through gate-tunable, eye-like gain control across the infrared–polarization domain. By engineering an electrostatically reconfigurable Au/BP/PdSe 2 vdWH, we establish a physically intrinsic mechanism for spatially reconfigurable photoresponse and dynamic responsivity modulation. Integrated with an embedded neural network controller, the system forms a real-time hardware-in-the-loop feedback loop that autonomously adjusts optical gain, achieving a LDR of ~ 80 dB at 1550 nm, sub-millisecond response, and intrinsic polarization sensitivity (PR ≈ 8)—all without external optics or analog circuitry. Beyond device performance, this architecture exemplifies a unified framework that couples material-level tunability with embedded intelligence, enabling chip-scale, self-regulating visual functions. The demonstrated system not only restores perceptual stability under fluctuating illumination, but also extends the perceptual envelope of vision systems into spectral and polarization domains inaccessible to the human eye. These results lay the foundation for next-generation self-adaptive photonic sensors with broad applications in neuromorphic computing, autonomous robotics, and edge artificial intelligence. Method Device fabrication. Au/BP/PdSe 2 vdWH photodetectors were fabricated via standard electron-beam lithography (EBL) and deterministic dry transfer. Cr/Au (5 nm/15 nm) electrodes were patterned on Si/SiO 2 substrates (285 nm oxide) using EBL (FEI F50, NPGS system), followed by electron-beam evaporation (EBE) and lift-off in acetone. Thin BP flakes were mechanically exfoliated and transferred onto the substrates via PDMS stamping. Subsequently, PdSe 2 flakes were exfoliated and vertically stacked atop the BP under optical alignment, forming the asymmetric vdWH. All exfoliation and transfer steps were conducted in a nitrogen-filled glovebox to suppress ambient degradation. A few-layer hBN flake was transferred last to serve as an encapsulation layer. Characterizations and measurements. Optical microscopy (Zeiss Lab5) was used to inspect device morphology and layer alignment. Thicknesses of the exfoliated flakes were confirmed via AFM (Cypher S, Asylum Research) in tapping mode using Ti/Ir-coated conductive probes. Electrical measurements were performed under ambient conditions using a semiconductor parameter analyzer (Keithley 4200A-SCS). All optoelectronic measurements were carried out under 1550 nm laser illumination (spot size ≈ 1 µm). Spatially resolved photocurrent mapping was conducted using the MStarter 200 scanning photoresponse platform. Time-resolved photoresponse waveforms were captured with a mixed-domain oscilloscope (Tektronix MDO34) under modulated illumination. Neural-network-based self-adaptive control was implemented via an STM32 microcontroller interfaced with a signal collector, using custom LabVIEW routines for real-time signal synchronization and feedback operation. Declarations Competing interests The authors declare no competing interests. Acknowledgement This work was supported by Strategic Priority Research Program (B) of Chinese Academy of Sciences (XDB0580000); National Key Research and Development Program of China (2023YFA1406900); National Natural Science Foundation of China (62222514, 62550006, 62350073, U2341226, 12227901, U23A6002); China National Postdoctoral Program for Innovative Talent (BX20250352); Shanghai Science and Technology Committee (23ZR1482000, 22JC1402900, 2019SHZDZX01); This work was partially carried out at the Soft Matter Nanofab (SMN180827) in ShanghaiTech University and the robotic AI-Scientist platform of Chinese Academy of Sciences. Data availability Relevant data supporting the key findings of this study are available in the article. All raw data generated in this study are available from the corresponding author upon reasonable request. References Liu S, Liu L, Tang J, Yu B, Wang Y, Shi W (2019) Edge Computing for Autonomous Driving: Opportunities and Challenges. Proceedings of the IEEE 107 (8), 1697–1716 Tang J, Liu S, Liu L, Yu B, Shi W (2020) LoPECS: A Low-Power Edge Computing System for Real-Time Autonomous Driving Services. IEEE Access 8:30467–30479 Ibn-Khedher H, Laroui M, Moungla H, Afifi H, Abd-Elrahman E (2022) Next-Generation Edge Computing Assisted Autonomous Driving Based Artificial Intelligence Algorithms. IEEE Access 10:53987–54001 Kim MS, Kim MS, Lee GJ, Sunwoo S-H, Chang S, Song YM, Kim D-H (2022) Bio-Inspired Artificial Vision and Neuromorphic Image Processing Devices. Adv Mater Technol 7(2):2100144 Elouardi A, Bouaziz S, Dupret A, Lacassagne L, Klein JO, Reynaud R (2007) Image Processing Vision Systems: Standard Image Sensors Versus Retinas. IEEE Trans Instrum Meas 56(5):1675–1687 Son J, Kim S, Sohn K (2015) A multi-vision sensor-based fast localization system with image matching for challenging outdoor environments. Expert Syst Appl 42(22):8830–8839 Wang Z, Wan T, Ma S, Chai Y (2024) Multidimensional vision sensors for information processing. Nat Nanotechnol 19(7):919–930 Choi C, Lee GJ, Chang S, Song YM, Kim D-H (2024) Inspiration from Visual Ecology for Advancing Multifunctional Robotic Vision Systems: Bio-inspired Electronic Eyes and Neuromorphic Image Sensors. Adv Mater 36(48):2412252 Shan TJ, Kailath T (1988) Adaptive algorithms with an automatic gain control feature. IEEE Trans Circuits Syst 35(1):122–127 Liu C, Yan YP, Goh WL, Xiong YZ, Zhang LJ, Madihian M (2012) A 5-Gb/s Automatic Gain Control Amplifier With Temperature Compensation. IEEE J Solid-State Circuits 47(6):1323–1333 Li X, Wang J, Yu F, Chen J, Chen X, Lu W, Li G (2025) Nonlinear memristive computational spectrometer. Light: Sci Appl 14(1):47 Yang J, Li X, Gu J, Yu F, Chen J, Liu J, Song Y, Lin X, Chen X, Lu W, Li G (2025) Reconfigurable Ion-Migration Driven Memristor for Multistate Neuromorphic Associative Learning. ACS Photonics 12(5):2574–2582 Li S, Zhang Z, Chen X, Deng W, Lu Y, Sui M, Gong F, Xu G, Li X, Liu F, You C, Chu F, Wu Y, Yan H, Zhang Y (2022) A High-Performance In-Memory Photodetector Realized by Charge Storage in a van der Waals MISFET. Adv Mater 34(10):2107734 Li T, Miao J, Fu X, Song B, Cai B, Ge X, Zhou X, Zhou P, Wang X, Jariwala D, Hu W (2023) Reconfigurable, non-volatile neuromorphic photovoltaics. Nat Nanotechnol 18(11):1303–1310 Li X, Chen J, Yu F, Chen X, Lu W, Li G (2024) Achieving a Noise Limit with a Few-layer WSe2 Avalanche Photodetector at Room Temperature. Nano Lett 24(42):13255–13262 Wu F, Li Q, Wang P, Xia H, Wang Z, Wang Y, Luo M, Chen L, Chen F, Miao J, Chen X, Lu W, Shan C, Pan A, Wu X, Ren W, Jariwala D, Hu W (2019) High efficiency and fast van der Waals hetero-photodiodes with a unilateral depletion region. Nat Commun 10(1):4663 Wang F, Zhang T, Xie R, Liu A, Dai F, Chen Y, Xu T, Wang H, Wang Z, Liao L, Wang J, Zhou P, Hu W (2024) Next-Generation Photodetectors beyond Van Der Waals Junctions. Adv Mater 36(3):2301197 Yang J, Li X, Gu J, Yu F, Chen J, Lu W, Chen X (2025) High-Stability WSe2 Homojunction Photodetectors via Asymmetric Schottky and PIN Architectures Coatings [Online] Xu L, Xiong H, Fu Z, Deng M, Wang S, Zhang J, Shang L, Jiang K, Li Y, Zhu L, He L, Hu Z, Chu J (2021) High Conductance Margin for Efficient Neuromorphic Computing Enabled by Stacking Nonvolatile van der Waals Transistors. Phys Rev Appl 16(4):044049 Zou Y, Zhang Z, Yan J, Lin L, Huang G, Tan Y, You Z, Li P (2022) High-temperature flexible WSe2 photodetectors with ultrahigh photoresponsivity. Nat Commun 13(1):4372 Zhang J, Tse P-L, Jalil A-R, Kölzer J, Rosenbach D, Luysberg M, Panaitov G, Lüth H, Hu Z, Grützmacher D, Lu JG, Schäpers T (2021) Flux periodic oscillations and phase-coherent transport in GeTe nanowire-based devices. Nat Commun 12(1):754 Song W, Chen J, Li Z, Fang X (2021) Self-Powered MXene/GaN van der Waals Heterojunction Ultraviolet Photodiodes with Superhigh Efficiency and Stable Current Outputs. Adv Mater 33(27):2101059 Jeong M-H, Ra H-S, Lee S-H, Kwak D-H, Ahn J, Yun WS, Lee J, Chae W-S, Hwang DK, Lee J-S (2022) Multilayer WSe2/MoS2 Heterojunction Phototransistors through Periodically Arrayed Nanopore Structures for Bandgap Engineering. Adv Mater 34(8):2108412 Yang X, Zhou X, Li L, Wang N, Hao R, Zhou Y, Xu H, Li Y, Zhu G, Zhang Z, Wang J, Feng Q (2023) Large-Area Black Phosphorus/PtSe2 Schottky Junction for High Operating Temperature Broadband Photodetectors. Small 19(28):2206590 Jiao H, Wang X, Chen Y, Guo S, Wu S, Song C, Huang S, Huang X, Tai X, Lin T, Shen H, Yan H, Hu W, Meng X, Chu J, Zhang Y, Wang J HgCdTe/black phosphorus van der Waals heterojunction for high-performance polarization-sensitive midwave infrared photodetector. Sci Adv 8 (19), eabn1811 Zhu Y, Wang Y, Pang X, Jiang Y, Liu X, Li Q, Wang Z, Liu C, Hu W, Zhou P (2024) Non-volatile 2D MoS2/black phosphorus heterojunction photodiodes in the near- to mid-infrared region. Nat Commun 15(1):6015 Kim H, Uddin SZ, Lien D-H, Yeh M, Azar NS, Balendhran S, Kim T, Gupta N, Rho Y, Grigoropoulos CP, Crozier KB, Javey A (2021) Actively variable-spectrum optoelectronics with black phosphorus. Nature 596(7871):232–237 Lee S, Peng R, Wu C, Li M (2022) Programmable black phosphorus image sensor for broadband optoelectronic edge computing. Nat Commun 13(1):1485 Wu S, Chen Y, Wang X, Jiao H, Zhao Q, Huang X, Tai X, Zhou Y, Chen H, Wang X, Huang S, Yan H, Lin T, Shen H, Hu W, Meng X, Chu J, Wang J (2022) Ultra-sensitive polarization-resolved black phosphorus homojunction photodetector defined by ferroelectric domains. Nat Commun 13(1):3198 He T, Ma H, Wang Z, Li Q, Liu S, Duan S, Xu T, Wang J, Wu H, Zhong F, Ye Y, Wu J, Lin S, Zhang K, Martyniuk P, Rogalski A, Wang P, Li L, Lin H, Hu W (2024) On-chip optoelectronic logic gates operating in the telecom band. Nat Photonics 18(1):60–67 Wang S, Jiang C, Yu Y, Zhang Z, Quhe R, Yang R, Tian Y, Chen X, Fan W, Niu Y, Yan B, Jiang C, Wang Y, Wang Z, Liu C, Hu W, Zhang J, Zhou P Tellurium nanowire retinal nanoprosthesis improves vision in models of blindness. Science 388 (6751), eadu2987 Zhou S, Fan H, Wen S, Wei Y, Chen H, Zhang Y, Zou T, Gao P, Wei D, Yin Y, Lan C, Li C, Liu Y (2025) Dual-Mode Photodetectors Mimicking Retinal Rod and Cone Cells for High Dynamic Range Image Sensor. Laser & Photonics Reviews n/a (n/a), 2402192 Kocdemir IH, Akyuz AO, Koz A, Chalmers A, Alatan A, Kalkan S (2022) In Object Detection for Autonomous Driving: High-Dynamic Range vs. Low-Dynamic Range Images , 2022 IEEE 24th International Workshop on Multimedia Signal Processing (MMSP), 26–28 Sept. ; 2022; pp 1–5 Ma C, Yuan S, Cheung P, Watanabe K, Taniguchi T, Zhang F, Xia F (2022) Intelligent infrared sensing enabled by tunable moiré quantum geometry. Nature 604(7905):266–272 Liao F, Zhou Z, Kim BJ, Chen J, Wang J, Wan T, Zhou Y, Hoang AT, Wang C, Kang J, Ahn J-H, Chai Y (2022) Bioinspired in-sensor visual adaptation for accurate perception. Nat Electron 5(2):84–91 Pi L, Wang P, Liang S-J, Luo P, Wang H, Li D, Li Z, Chen P, Zhou X, Miao F, Zhai T (2022) Broadband convolutional processing using band-alignment-tunable heterostructures. Nat Electron 5(4):248–254 Yoshioka K, Wakamura T, Hashisaka M, Watanabe K, Taniguchi T, Kumada N (2022) Ultrafast intrinsic optical-to-electrical conversion dynamics in a graphene photodetector. Nat Photonics 16(10):718–723 Yoon HH, Fernandez HA, Nigmatulin F, Cai W, Yang Z, Cui H, Ahmed F, Cui X, Uddin MG, Minot ED, Lipsanen H, Kim K, Hakonen P, Hasan T, Sun Z (2022) Miniaturized spectrometers with a tunable van der Waals junction. Science 378(6617):296–299 Wang Y, Gu Y, Cui A, Li Q, He T, Zhang K, Wang Z, Li Z, Zhang Z, Wu P, Xie R, Wang F, Wang P, Shan C, Li H, Ye Z, Zhou P, Hu W (2022) Fast Uncooled Mid-Wavelength Infrared Photodetectors with Heterostructures of van der Waals on Epitaxial HgCdTe. Adv Mater 34(6):2107772 Wang H, Wang W, Zhong Y, Li D, Li Z, Xu X, Song X, Chen Y, Huang P, Mei A, Han H, Zhai T, Zhou X (2022) Approaching the External Quantum Efficiency Limit in 2D Photovoltaic Devices. Adv Mater 34(39):2206122 Moon G, Min SY, Han C, Lee S-H, Ahn H, Seo S-Y, Ding F, Kim S, Jo M-H (2023) Atomically Thin Synapse Networks on Van Der Waals Photo-Memtransistors. Adv Mater 35(4):2203481 Lai H, Lu Z, Lu Y, Yao X, Xu X, Chen J, Zhou Y, Liu P, Shi T, Wang X, Xie W, Fast (2023) Multi-Bit, and Vis-Infrared Broadband Nonvolatile Optoelectronic Memory with MoS2/2D-Perovskite Van der Waals Heterojunction. Adv Mater 35(6):2208664 Liu X, Huang M, Zou X, Ali W, Rehman SU, Li J, Li Z, Xiang L, Pan A (2025) Alcohol-sensitive MoS2 optoelectronic synapses for mimicking human-like visual adaptation. InfoMat n/a (n/a), e70019 Jiang J, Xu W, Guo F, Yang S, Ge W, Shen B, Tang N (2023) Polarization-Resolved Near-Infrared PdSe2 p-i-n Homojunction Photodetector. Nano Lett 23(20):9522–9528 Chen Q, Zhang Y, Liu S, Han T, Chen X, Xu Y, Meng Z, Zhang G, Zheng X, Zhao J, Cao G, Liu G (2020) Switchable Perovskite Photovoltaic Sensors for Bioinspired Adaptive Machine Vision. Adv Intell Syst 2(9):2000122 Hong S, Choi SH, Park J, Yoo H, Oh JY, Hwang E, Yoon DH, Kim S (2020) Sensory Adaptation and Neuromorphic Phototransistors Based on CsPb(Br1–xIx)3 Perovskite and MoS2 Hybrid Structure. ACS Nano 14(8):9796–9806 Xie D, Wei L, Xie M, Jiang L, Yang J, He J, Jiang J (2021) Photoelectric Visual Adaptation Based on 0D-CsPbBr3-Quantum-Dots/2D-MoS2 Mixed-Dimensional Heterojunction Transistor. Adv Funct Mater 31(14):2010655 Liu W, Yang X, Wang Z, Li Y, Li J, Feng Q, Xie X, Xin W, Xu H, Liu Y (2023) Self-powered and broadband opto-sensor with bionic visual adaptation function based on multilayer γ-InSe flakes. Light: Sci Appl 12(1):180 He Z, Shen H, Ye D, Xiang L, Zhao W, Ding J, Zhang F, Di C-a, Zhu D (2021) An organic transistor with light intensity-dependent active photoadaptation. Nat Electron 4(7):522–529 Li L, Li S, Wang W, Zhang J, Sun Y, Deng Q, Zheng T, Lu J, Gao W, Yang M, Wang H, Pan Y, Liu X, Yang Y, Li J, Huo N (2024) Adaptative machine vision with microsecond-level accurate perception beyond human retina. Nat Commun 15(1):6261 Zhang S, Han L, Xiao K, Zhang L, Shi C, Xu L, Deng K, Zou Y, Jiang M, Lv X, Zhu Y, Li Q, Liu C, Tang W, Li G, Ding S, Chen X, Lu W (2023) H-BN-Encapsulated Uncooled Infrared Photodetectors Based on Tantalum Nickel Selenide. Adv Funct Mater 33(48):2305380 Xin W, Li X-K, He X-L, Su B-W, Jiang X-Q, Huang K-X, Zhou X-F, Liu Z-B, Tian J-G (2018) Black-Phosphorus-Based Orientation-Induced Diodes. Adv Mater 30(2):1704653 Zhao T, Chen Y, Xu T, Zhong F, Yu Y, Ma H, Zhang K, Duan S, Hu J, Wang S, Guo J, Wang Z (2024) Topological Insulator Bi2Se3 Heterojunction with a Low Dark Current for Midwave Infrared Photodetection. ACS Photonics 11(6):2450–2458 Additional Declarations There is NO Competing Interest. Supplementary Files SUPPLEMENTARYINFORMATION.docx Self-Adaptive Infrared Vision via Neural-Controlled Gain Compression in a Single Photodetector Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7289222","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":506109009,"identity":"6757fd2d-82e2-4934-bc9e-9ceedcbe5ade","order_by":0,"name":"Guanhai Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAklEQVRIie3RP4oCMRQG8AwBt3lgGxnRKzyxWlj0INskDGgjIthYSWRhKw8wMMcY2DpD7iCCFusNUgourC9aiWS0XNh8VQL55csfxmJi/mIE10biGzRpbL4ZgydIoo2bjTotTUQSeWyIJLmzfTQ0IfK4pvu+WllArsqNPVDLsj1MdeKObDANkd6+0kQa6ms3QiIWoG14a82yeZDkyhMgIj0xAEKylO6ldD0RqizGzh/ME36qI12hdJUj9jGd+BbuSaO2hfanR0bZEbvJjD6I7rKvPl/XmIVb8sw6+fMLzWJcOrdYDl+KD7s9LgbhFnM7ZZefug5CLfd7ifDqmJiYmP+ZM76MXAZOQJneAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0000-0002-2745-055X","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":true,"prefix":"","firstName":"Guanhai","middleName":"","lastName":"Li","suffix":""},{"id":506109010,"identity":"4fcb6bac-2e77-49d3-82b8-28c6ffb1ceb5","order_by":1,"name":"Yuxin Song","email":"","orcid":"","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Yuxin","middleName":"","lastName":"Song","suffix":""},{"id":506109011,"identity":"b316c1f1-5393-4506-b022-c8da19d99ca8","order_by":2,"name":"Guanhai Li","email":"","orcid":"","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Guanhai","middleName":"","lastName":"Li","suffix":""},{"id":506109012,"identity":"6e248a44-5a15-46cc-80af-5d7d7a05d73c","order_by":3,"name":"Junzhe Gu","email":"","orcid":"","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Junzhe","middleName":"","lastName":"Gu","suffix":""},{"id":506109013,"identity":"49172f6e-2f1f-4c8e-86f5-9fafc6e0c3bb","order_by":4,"name":"Jin Chen","email":"","orcid":"https://orcid.org/0000-0003-4183-4629","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Jin","middleName":"","lastName":"Chen","suffix":""},{"id":506109014,"identity":"3960093a-e6e2-4d80-acd2-d481596a8405","order_by":5,"name":"Feilong Yu","email":"","orcid":"","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Feilong","middleName":"","lastName":"Yu","suffix":""},{"id":506109015,"identity":"6bae37eb-980c-4c80-97e8-9ba696afb7e1","order_by":6,"name":"Juntong Liu","email":"","orcid":"","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Juntong","middleName":"","lastName":"Liu","suffix":""},{"id":506109016,"identity":"1ca3710a-50a7-4690-91d5-3b9c06cb3bce","order_by":7,"name":"Jiaji Yang","email":"","orcid":"","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Jiaji","middleName":"","lastName":"Yang","suffix":""},{"id":506109017,"identity":"0fb6a3eb-0a85-46da-97e7-b3c72d28148f","order_by":8,"name":"Xiaoshuang Chen","email":"","orcid":"https://orcid.org/0000-0003-0131-9454","institution":"Shanghai Institute of Technical Physics, Chinese Academy of Sciences","correspondingAuthor":false,"prefix":"","firstName":"Xiaoshuang","middleName":"","lastName":"Chen","suffix":""},{"id":506109018,"identity":"dd44a5fe-c60c-48f1-be4e-0db5f65874dc","order_by":9,"name":"Wei Lu","email":"","orcid":"https://orcid.org/0000-0001-9859-8394","institution":"Shanghai Institute of Technical Physics","correspondingAuthor":false,"prefix":"","firstName":"Wei","middleName":"","lastName":"Lu","suffix":""}],"badges":[],"createdAt":"2025-08-04 09:15:37","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7289222/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7289222/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":90081943,"identity":"8e60ec33-ffad-4831-a63b-23978697e24d","added_by":"auto","created_at":"2025-08-28 09:09:49","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1362476,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eBio-inspired photodetector with eye-like self-adaptive gain control enabled by electrostatic barrier reconfiguration. a,\u003c/strong\u003e Schematic illustration of biological visual self-adaptation: the pupil dynamically modulates light intake under varying illumination, while the retina adjusts responsivity by activating photoreceptors to maintain perceptual stability. \u003cstrong\u003eb,\u003c/strong\u003e Device architecture of the gate-tunable Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e vdWH. A vertical stack is formed by placing a BP flake (primary absorber) on a gold electrode and capping it with PdSe\u003csub\u003e2\u003c/sub\u003e, forming an asymmetric junction. hBN encapsulation ensures ambient stability, and a global back-gate voltage (V\u003csub\u003eg\u003c/sub\u003e) applied to a heavily p-doped Si substrate emulates self-adaptive function of the eye. \u003cstrong\u003ec,\u003c/strong\u003e Gate-tunable electrostatic control enables simultaneous modulation of the active photoresponse area (A) and responsivity (R), mimicking pupil scaling and photoreceptors activation, respectively. \u003cstrong\u003ed,\u003c/strong\u003e Schematic diagram of the transfer curve showing how reconfigurable field distributions under varying gate voltages reshape the photoelectric conversion characteristics and enable device-intrinsic nonlinear gain compression. \u003cstrong\u003ee,\u003c/strong\u003e Embedded neural controller based on a fully connected neural network (FCNN) infers and applies the optimal gate voltage based on incident light intensity and desired output current, completing a closed-loop feedback system for real-time optical self-adaptation.\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7289222/v1/85b9c712b99b10258f84c1ce.png"},{"id":90081949,"identity":"11855cf5-8045-4c7c-b18f-75b28912b578","added_by":"auto","created_at":"2025-08-28 09:09:51","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":2064937,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eGate-tunable gain regulation and spatially reconfigurable photoresponse in the Au/BP/PdSe\u003c/strong\u003e\u003csub\u003e\u003cstrong\u003e2\u003c/strong\u003e\u003c/sub\u003e\u003cstrong\u003e heterostructure. a,\u003c/strong\u003e Pre-contact energy band alignment of Au, BP, and PdSe\u003csub\u003e2\u003c/sub\u003e, illustrating asymmetric Schottky barriers that enable gate-tunable charge separation. \u003cstrong\u003eb,\u003c/strong\u003e Atomic force microscopy (AFM) height profile of the vertical heterostructure confirming well-defined layer stacking and thicknesses: ~10 nm BP, ~240 nm PdSe\u003csub\u003e2\u003c/sub\u003e, and ~50 nm hBN. Optical micrograph inset shows the clean junction interface and encapsulated structure. Scale bar (white): 5 μm. \u003cstrong\u003ec,\u003c/strong\u003e Transfer curves under dark and illuminated conditions (λ = 1550 nm, V\u003csub\u003eds\u003c/sub\u003e = 0 V), revealing strong gate-dependent photocurrent amplification and nonlinear gain characteristics. \u003cstrong\u003ed–f,\u003c/strong\u003e Spatial photocurrent mapping under 1550 nm illumination (P = 40 μW) at different gate voltages: -40 V (\u003cstrong\u003ed\u003c/strong\u003e), 0 V (\u003cstrong\u003ee\u003c/strong\u003e), +40 V (\u003cstrong\u003ef\u003c/strong\u003e). The photoactive region shifts from the BP/PdSe₂ interface (−40 V) to the Au/BP interface (0 V and +40 V), demonstrating electrostatic control over absorption location and gain intensity.\u003cstrong\u003e g–i,\u003c/strong\u003e Section current diagram in Figure 2d-f, inset indicate the position of the section plane.\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7289222/v1/be61647d59bf66b68fa11ee0.png"},{"id":90081940,"identity":"734b3b09-b5d9-4cd6-bf1c-49ecdad6ce11","added_by":"auto","created_at":"2025-08-28 09:09:49","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":757349,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFunctional performance of the self-adaptive Au/BP/PdSe\u003c/strong\u003e\u003csub\u003e\u003cstrong\u003e2\u003c/strong\u003e\u003c/sub\u003e\u003cstrong\u003e photodetector: LDR, fast response, and polarization selectivity. a,\u003c/strong\u003e Output I–V characteristics under varying 1550 nm illumination, showing photovoltaic behavior and monotonic increase of short-circuit current (I\u003csub\u003esc\u003c/sub\u003e) with incident power. \u003cstrong\u003eb,\u003c/strong\u003e Photoswitching dynamics at V\u003csub\u003eg\u003c/sub\u003e = 40 V under different input powers, demonstrating high on/off ratios, reproducibility, and fast transitions. \u003cstrong\u003ec,\u003c/strong\u003e Responsivity as a function of optical power at different gate voltages. A peak responsivity of 45,260 V W\u003csup\u003e-1\u003c/sup\u003e is achieved at V\u003csub\u003eg\u003c/sub\u003e = 40 V and P = 0.62 mW mm\u003csup\u003e-2\u003c/sup\u003e. \u003cstrong\u003ed,\u003c/strong\u003e Comparison between fixed-bias and real-time self-adaptive gating. Self-adaptive control linearizes the response across four orders of magnitude, expanding the LDR to ~80 dB. \u003cstrong\u003ee,\u003c/strong\u003e Benchmarking against representative adaptive photodetectors across wavelength and LDR domains, confirming leading performance in the near-infrared. \u003cstrong\u003ef,\u003c/strong\u003e Polarization-resolved photoresponse. A polarization ratio (PR ≈ 8) is consistently maintained across gate voltages, enabling intrinsic anisotropic detection.\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7289222/v1/c2ba4ceb6af893443c67493a.png"},{"id":90081942,"identity":"417414d8-9282-4cad-b1b0-34adcea0619e","added_by":"auto","created_at":"2025-08-28 09:09:49","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":3035575,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eNeural feedback control and real-time exposure compensation in a hardware-integrated self-adaptive vision system. a,\u003c/strong\u003e System architecture of the hardware-in-the-loop control platform. The photocurrent is sampled and processed by a microcontroller running an embedded neural network to infer and apply the optimal gate voltage in real time. \u003cstrong\u003eb,\u003c/strong\u003e Two-step control logic implemented on the microcontroller: (1) light intensity estimation at V\u003csub\u003eg\u003c/sub\u003e = 0 V, and (2) gate voltage prediction for target current output. \u003cstrong\u003ec,\u003c/strong\u003e Gate voltage required to maintain a constant photocurrent under variable illumination, revealing an inverse nonlinear dependence and confirming device-level gain self-regulation. \u003cstrong\u003ed,\u003c/strong\u003e Structure of the FCNN used for inference, comprising two input nodes, one hidden layer with 90 units, and one output node. \u003cstrong\u003ee, \u003c/strong\u003eThe real (red dots) and the predicted (blue dots) gate voltages as a function of the power densities of 1550 nm laser and source-drain current. \u003cstrong\u003ef, \u003c/strong\u003eSingle-pixel imaging under eight illumination levels: fixed-bias operation (left) leads to underexposure or overexposure, while self-adaptive control (right) maintains consistent brightness and image fidelity.\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7289222/v1/8737258dabaecbf9f346f5e0.png"},{"id":91151491,"identity":"53205f4b-570d-49d0-b6c0-6e31c7275b38","added_by":"auto","created_at":"2025-09-12 07:14:05","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6055958,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7289222/v1/af0cf71f-9c73-4af0-8229-50e8d41ba7d5.pdf"},{"id":90081935,"identity":"ed5f656f-bedf-4983-add6-6579dc512705","added_by":"auto","created_at":"2025-08-28 09:09:45","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":3276868,"visible":true,"origin":"","legend":"Self-Adaptive Infrared Vision via Neural-Controlled Gain Compression in a Single Photodetector","description":"","filename":"SUPPLEMENTARYINFORMATION.docx","url":"https://assets-eu.researchsquare.com/files/rs-7289222/v1/57c043a49e640786d6e32069.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Self-Adaptive Infrared Vision via Neural-Controlled Gain Compression in a Single Photodetector","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHuman vision excels at self-adapting to dynamic lighting through pupil control of light intake and retina control of photoreceptors activation. Yet, despite its sophistication, biological vision remains blind to infrared wavelengths and insensitive to polarization\u0026mdash;two critical modalities for emerging applications in autonomous sensing, edge intelligence, and human\u0026ndash;machine interaction.\u003csup\u003e\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e Bridging this gap demands vision hardware that not only self-adapts in real time, but also extends beyond the spectral and polarization boundaries of human perception.\u003csup\u003e\u003cspan additionalcitationids=\"CR5 CR6 CR7\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eExisting image sensors rely heavily on system-level compensation\u0026mdash;such as mechanical apertures, analog gain control, or digital post-processing\u0026mdash;to mitigate underexposure or overexposure under variable illumination.\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e While effective in controlled environments, these solutions introduce latency, increase power consumption, and hinder integration, making them unsuitable for next-generation autonomous platforms where fast, compact, and self-adaptive sensory systems are essential.\u003csup\u003e\u003cspan additionalcitationids=\"CR12 CR13\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eTo address this challenge, there is a growing need for solid-state photodetectors that exhibit intrinsic, reversible optical gain modulation\u0026mdash;capable of continuously compressing LDR without relying on external optics or auxiliary electronics. Two-dimensional (2D) material heterostructures have emerged as promising candidates, offering broadband responsivity, structural flexibility, and scalable integrability.\u003csup\u003e\u003cspan additionalcitationids=\"CR16 CR17 CR18 CR19 CR20 CR21 CR22 CR23 CR24 CR25 CR26 CR27 CR28 CR29 CR30\" citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e However, most 2D photodetectors still exhibit static gain responses and saturate under strong illumination, limiting their applicability in real-world, high-contrast scenarios.\u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e Moreover, previously reported adaptation schemes remain either passive, externally driven, or non-generalizable, lacking the device-level intelligence and closed-loop control necessary for true autonomous vision.\u003csup\u003e\u003cspan additionalcitationids=\"CR35 CR36 CR37 CR38 CR39 CR40 CR41\" citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e\u003cp\u003eHere, we demonstrate a neuromorphic photodetector that not only emulates the self-adaptive functionality of the human eye, but surpasses it by enabling dynamic gain regulation across infrared and polarization domains\u0026mdash; regime fundamentally inaccessible to biological systems. Our design is based on a gate-tunable Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e vdWH, wherein electrostatically reconfigurable barriers enable continuous and reversible modulation of both response area and responsivity. By coupling this mechanism with an embedded neural-network-based microcontroller, we realize a hardware-in-the-loop closed feedback system that autonomously adjusts optical gain in real time. This architecture expands LDR by three orders of magnitude (~\u0026thinsp;80 dB at 1550 nm), achieves sub-millisecond response, and offers intrinsic polarization sensitivity (PR\u0026thinsp;\u0026asymp;\u0026thinsp;8)\u0026mdash;all without external optics, analog amplifiers, or post-processing. This work introduces a scalable optoelectronic platform that tightly integrates physical tunability with embedded intelligence, offering a pathway toward chip-scale, self-adaptive vision systems that augment and exceed biological perception.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eBiological visual systems achieve robust perception across diverse lighting conditions via low-latency feedback loops between the pupil and retina (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ea). The pupil adjusts light intake by changing its aperture, while retina regulate responsivity by altering the degree of photoreceptors activation to maintain consistent contrast and expand LDR. To emulate this dual-modulation strategy on a chip, we design a gate-tunable Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e vdWH that supports real-time, device-level gain control (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003eb).\u003c/p\u003e\n\u003cp\u003eThe device comprises a BP flake sandwiched between a gold electrode (bottom) and a PdSe\u003csub\u003e2\u003c/sub\u003e layer (top), forming an asymmetric vertical heterojunction. This vertical stacking enables electrostatic barrier engineering via a single back-gate terminal (V\u003csub\u003eg\u003c/sub\u003e). Under weak illumination, the Au/BP interface\u0026mdash;with its larger space charge region\u0026mdash;supports stronger field-assisted carrier separation and higher responsivity. Under strong illumination, the BP/PdSe\u003csub\u003e2\u003c/sub\u003e interface becomes dominant, but reduced space charge region and field strength lead to suppressed responsivity. This intrinsic asymmetry enables dynamic reallocation of the photoactive region, analogous to how pupil diameter and retinal photoreceptors activation shift in response to ambient light.\u003c/p\u003e\n\u003cp\u003eDue to the atomically thin nature of 2D materials, the Fermi level of BP can be finely tuned by the gate voltage, which modulates both the spatial extent of the active region and the local responsivity (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ec). These two parameters\u0026mdash;response area (A) and responsivity (R)\u0026mdash;together determine the net optical gain and allow continuous compression of the photocurrent response, all within a monolithic structure and without external gain control modules.\u003c/p\u003e\n\u003cp\u003eAs illustrated in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ed, the band alignment and field distribution evolve significantly under different gate biases, enabling active control over the internal gain landscape. This barrier reconfiguration serves as the physical basis for eye-like nonlinear gain compression, offering high LDR operation without external optics or multi-pixel architectures.\u003c/p\u003e\n\u003cp\u003eTo close the loop, we integrate the photodetector with a microcontroller-based neural network (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003ee). A FCNN infers the required gate voltage in real time based on the incident light intensity and the desired output current. The inferred voltage is then applied to the device to stabilize the output across variable lighting\u0026mdash;mimicking the closed-loop self-adaptation of biological vision, but extending it into the infrared and polarization domain.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe self-adaptive behavior of our photodetector arises from the asymmetric band alignment and electrostatic reconfigurability of the Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e vertical heterostructure. As shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ea, pre-contact energy diagrams indicate built-in Schottky barriers at both metal\u0026ndash;semiconductor (Au/BP) and semiconductor\u0026ndash;semimetal (BP/PdSe\u003csub\u003e2\u003c/sub\u003e) interfaces. The larger barrier at the Au/BP contact (~\u0026thinsp;0.23 eV) provides rectifying behavior and facilitates unidirectional photocarrier separation, while the smaller barrier (~\u0026thinsp;0.17 eV) at the BP/PdSe\u003csub\u003e2\u003c/sub\u003e junction offers a secondary charge extraction path under gate modulation.\u003c/p\u003e\n\u003cp\u003eAFM measurements confirm the layered architecture (Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003eb), with well-controlled flake thicknesses that ensure vertical field coupling while preserving interfacial sharpness and device reproducibility. The device is encapsulated with hBN to maintain air stability during ambient operation. These structural features establish a robust physical platform for field-driven gain control.\u003c/p\u003e\n\u003cp\u003eThe dark state characterization test results of the Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e device are shown in Supplementary Figures \u003cspan class=\"InternalRef\"\u003eS1\u003c/span\u003e and S2, these results establish a robust electrostatic tuning framework that underpins the dynamic gain modulation demonstrated in the self-adaptive photodetection system.\u003c/p\u003e\n\u003cp\u003eThe transfer characteristics in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ec show strong gate-tunable photocurrent enhancement at zero source-drain bias. Illumination induces nonlinear amplification that intensifies with increasing gate voltage, indicating field-assisted carrier separation and enhanced band bending. This nonlinear gating behavior underpins the gain compressive response required for wide LDR operation and reflects the intrinsic self-adaptive nature of the device.\u003c/p\u003e\n\u003cp\u003eTo visualize how the absorption zone and gain evolve with gating, we performed spatial photocurrent mapping under 1550 nm excitation (Figs.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003ed\u0026ndash;i). At V\u003csub\u003eg\u003c/sub\u003e = 0 V, the photocurrent is localized near the Au/BP interface, where the built-in field efficiently separates photocarriers. At +\u0026thinsp;40 V, this region expands due to increased Schottky barrier height and depletion width, enhancing responsivity\u0026mdash;functionally akin to the eye reflex under dim light. In contrast, applying \u0026minus;\u0026thinsp;40 V shifts the junction field toward the BP/PdSe\u003csub\u003e2\u003c/sub\u003e side. However, PdSe\u003csub\u003e2\u003c/sub\u003e screens the BP layer from most excitation. Combined with reduced response area and responsivity, this results in significantly reduced photocurrent and earlier saturation\u0026mdash;analogous to the eye reflex under strong light to suppress overexposure. These results demonstrate that both the spatial origin and the magnitude of photoresponse are continuously programmable through gate voltage, forming a tunable gain landscape that self-adapts responsively to varying illumination. This electrostatic control serves as the physical foundation for device-intrinsic, eye-like gain compression. The full band evolution under illumination is shown in Supplementary Figure S3.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo evaluate the functional capabilities of the Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e heterostructure, we characterize its optoelectronic performance under near-infrared illumination. Figure\u0026nbsp;3a shows the output current\u0026ndash;voltage (I\u0026ndash;V) characteristics under various optical powers at 1550 nm. The observed photovoltaic shift\u0026mdash;with increasing short-circuit current (I\u003csub\u003esc\u003c/sub\u003e) and open-circuit voltage (V\u003csub\u003eoc\u003c/sub\u003e)\u0026mdash;confirms Schottky junction-based carrier separation, consistent with the internal field distribution in our asymmetric device.\u003c/p\u003e\n\u003cp\u003eFor practical sensing and imaging, fast and reliable photoswitching is critical. As shown in Fig.\u0026nbsp;3b and Supplementary Figure S4, the device demonstrates robust on\u0026ndash;off behavior across a wide power range, with on/off ratios exceeding 87 and sharp rise/fall edges. Time-resolved measurements (Supplementary Figures S5 and S6) reveal sub-millisecond response speeds, attributed to the built-in field and high carrier mobility of BP, highlighting compatibility with video-rate applications. Other optoelectronic test results are shown in Supplementary Figures S7-S10, demonstrating the excellent optoelectronic properties of the device.\u003c/p\u003e\n\u003cp\u003eResponsivity measurements at zero source-drain bias (Fig.\u0026nbsp;3c) show a peak responsivity of 45,260 V W\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e at V\u003csub\u003eg\u003c/sub\u003e = 40 V and P\u0026thinsp;=\u0026thinsp;0.62 mW mm\u003csup\u003e\u0026minus;\u0026thinsp;2\u003c/sup\u003e. The detectivity (D\u003csup\u003e*\u003c/sup\u003e) reaches 5\u0026times;10\u003csup\u003e9\u003c/sup\u003e cm Hz\u003csup\u003e1/2\u003c/sup\u003e W\u003csup\u003e\u0026minus;\u0026thinsp;1\u003c/sup\u003e (Supplementary Figure S11), arising from the combined effects of strong photovoltage generation, low dark current, and efficient electrostatic control.\u003c/p\u003e\n\u003cp\u003eTo assess LDR regulation, we analyze the power-dependent photocurrent behavior across gate voltages. As shown in Supplementary Figure S12, the photocurrent follows a power-law form (I\u003csub\u003eds\u003c/sub\u003e \u0026prop; P\u003csup\u003e\u0026alpha;\u003c/sup\u003e), where the exponent \u0026alpha; increases with gate bias due to enhanced field-assisted separation and expanded depletion width (Supplementary Figure S13). This tunable nonlinearity enables photocurrent compression under strong illumination, emulating the eye reflex to prevent overexposure.\u003c/p\u003e\n\u003cp\u003eCritically, by integrating real-time gate adjustment through our neural control loop, this nonlinear response is reshaped into a strictly linear profile spanning four orders of magnitude, corresponding to a LDR of ~\u0026thinsp;80 dB (Fig.\u0026nbsp;3d). Compared to fixed-gate operation, this represents a three-order-of-magnitude LDR expansion, without external optics or analog gain modules.\u003c/p\u003e\n\u003cp\u003eFigure 3e benchmarks our performance against representative adaptive photodetectors across spectral range and LDR compression capabilities.\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e35\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e43\u003c/span\u003e\u0026ndash;\u003cspan class=\"CitationRef\"\u003e51\u003c/span\u003e\u003c/sup\u003e The Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e system outperforms previously reported platforms in both near-infrared responsivity and LDR range, particularly within the 1550 nm window relevant to telecommunications and eye-safe imaging.\u003c/p\u003e\n\u003cp\u003eBeyond intensity self-adaptation, the detector also exhibits intrinsic polarization sensitivity due to the in-plane anisotropy of BP.\u003csup\u003e\u003cspan class=\"CitationRef\"\u003e52\u003c/span\u003e, \u003cspan class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e As shown in Fig.\u0026nbsp;3f, the photocurrent varies with polarization angle, while maintaining a stable polarization ratio (PR\u0026thinsp;\u0026asymp;\u0026thinsp;8) across gate biases. This anisotropic response enables full-Stokes detection and supports multifunctional imaging applications (Supplementary Figures S14, S15). Together, these results establish the Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e heterostructure as a multifunctional, neuromorphic photodetector capable of real-time, self-adaptive gain control, high-speed response, and polarization-resolved sensing\u0026mdash;offering a compact, scalable platform for chip-scale artificial vision and edge photonic intelligence.\u003c/p\u003e\n\u003cp\u003eTo validate real-time self-adaptive vision based on our device-level gain tuning mechanism, we develop a single-pixel imaging system that couples the photodetector with an embedded neural controller (Fig.\u0026nbsp;4a and Supplementary Figure S16). This setup emulates the biological eye\u0026rsquo;s reflex arc, where light perception and gate modulation are connected through a closed feedback loop.\u003c/p\u003e\n\u003cp\u003eThe self-adaptive control logic operates in two steps (Fig.\u0026nbsp;4b): (1) initial illumination estimation is performed by measuring the photocurrent at V\u003csub\u003eg\u003c/sub\u003e = 0 V; (2) the inferred light intensity is used to calculate and apply the gate voltage needed to achieve the target photocurrent output. The entire loop runs in real time and is implemented fully in hardware, with convergence achieved within milliseconds. Importantly, data acquisition is decoupled from exposure integration, allowing seamless feedback during scanning\u0026mdash;a key requirement for dynamic imaging.\u003c/p\u003e\n\u003cp\u003eFigure 4c maps the relationship between gate voltage and incident power required to maintain constant output current. The inverse nonlinear trend confirms that our device architecture supports continuous, autonomous gain regulation across wide illumination levels\u0026mdash;without the need for external optics or analog circuitry.\u003c/p\u003e\n\u003cp\u003eThe photocurrent is sampled and digitized by a signal collector, then fed into an STM32 microcontroller executing a pre-trained FCNN (Fig.\u0026nbsp;4d). The FCNN is trained on experimental data linking photocurrent, light intensity, and gate voltage, using Bayesian regularization to enhance generalization and mitigate overfitting. Once trained, it performs real-time inference to predict the optimal gate voltage based on the measured photocurrent and desired target current (Fig.\u0026nbsp;4e).\u003c/p\u003e\n\u003cp\u003eTo demonstrate the practical advantage of this self-adaptive platform, we conduct single-pixel imaging under varying optical powers. As shown in Fig.\u0026nbsp;4f, the fixed-gate configuration (left panels) leads to severe underexposure or overexposure depending on illumination conditions. In contrast, the self-adaptive configuration (right panel) maintains uniform contrast and high-fidelity image reproduction across all scenarios.\u003c/p\u003e\n\u003cp\u003eThis hardware-in-the-loop architecture\u0026mdash;integrating gate-tunable materials, neuromorphic inference, and photodetector physics\u0026mdash;realizes a fully autonomous exposure regulation system, functionally analogous to the biological eye reflex, yet extending well beyond it in spectral and polarization domains. Our results provide a blueprint for scalable, chip-level self-adaptive vision, enabling future applications in neuromorphic sensing, intelligent robotics, and edge optical intelligence.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn summary, we report a neuromorphic photodetector system that emulates and surpasses biological vision through gate-tunable, eye-like gain control across the infrared–polarization domain. By engineering an electrostatically reconfigurable Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e vdWH, we establish a physically intrinsic mechanism for spatially reconfigurable photoresponse and dynamic responsivity modulation. Integrated with an embedded neural network controller, the system forms a real-time hardware-in-the-loop feedback loop that autonomously adjusts optical gain, achieving a LDR of ~ 80 dB at 1550 nm, sub-millisecond response, and intrinsic polarization sensitivity (PR ≈ 8)—all without external optics or analog circuitry. Beyond device performance, this architecture exemplifies a unified framework that couples material-level tunability with embedded intelligence, enabling chip-scale, self-regulating visual functions. The demonstrated system not only restores perceptual stability under fluctuating illumination, but also extends the perceptual envelope of vision systems into spectral and polarization domains inaccessible to the human eye. These results lay the foundation for next-generation self-adaptive photonic sensors with broad applications in neuromorphic computing, autonomous robotics, and edge artificial intelligence.\u003c/p\u003e"},{"header":"Method","content":"\u003cp\u003e\u003cb\u003eDevice fabrication.\u003c/b\u003e Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e vdWH photodetectors were fabricated via standard electron-beam lithography (EBL) and deterministic dry transfer. Cr/Au (5 nm/15 nm) electrodes were patterned on Si/SiO\u003csub\u003e2\u003c/sub\u003e substrates (285 nm oxide) using EBL (FEI F50, NPGS system), followed by electron-beam evaporation (EBE) and lift-off in acetone. Thin BP flakes were mechanically exfoliated and transferred onto the substrates via PDMS stamping. Subsequently, PdSe\u003csub\u003e2\u003c/sub\u003e flakes were exfoliated and vertically stacked atop the BP under optical alignment, forming the asymmetric vdWH. All exfoliation and transfer steps were conducted in a nitrogen-filled glovebox to suppress ambient degradation. A few-layer hBN flake was transferred last to serve as an encapsulation layer.\u003c/p\u003e\u003cp\u003e\u003cb\u003eCharacterizations and measurements.\u003c/b\u003e Optical microscopy (Zeiss Lab5) was used to inspect device morphology and layer alignment. Thicknesses of the exfoliated flakes were confirmed via AFM (Cypher S, Asylum Research) in tapping mode using Ti/Ir-coated conductive probes. Electrical measurements were performed under ambient conditions using a semiconductor parameter analyzer (Keithley 4200A-SCS). All optoelectronic measurements were carried out under 1550 nm laser illumination (spot size ≈ 1 µm). Spatially resolved photocurrent mapping was conducted using the MStarter 200 scanning photoresponse platform. Time-resolved photoresponse waveforms were captured with a mixed-domain oscilloscope (Tektronix MDO34) under modulated illumination. Neural-network-based self-adaptive control was implemented via an STM32 microcontroller interfaced with a signal collector, using custom LabVIEW routines for real-time signal synchronization and feedback operation.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eCompeting interests\u003c/h2\u003e\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThis work was supported by Strategic Priority Research Program (B) of Chinese Academy of Sciences (XDB0580000); National Key Research and Development Program of China (2023YFA1406900); National Natural Science Foundation of China (62222514, 62550006, 62350073, U2341226, 12227901, U23A6002); China National Postdoctoral Program for Innovative Talent (BX20250352); Shanghai Science and Technology Committee (23ZR1482000, 22JC1402900, 2019SHZDZX01); This work was partially carried out at the Soft Matter Nanofab (SMN180827) in ShanghaiTech University and the robotic AI-Scientist platform of Chinese Academy of Sciences.\u003c/p\u003e\u003ch2\u003eData availability\u003c/h2\u003e\u003cp\u003eRelevant data supporting the key findings of this study are available in the article. All raw data generated in this study are available from the corresponding author upon reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eLiu S, Liu L, Tang J, Yu B, Wang Y, Shi W (2019) Edge Computing for Autonomous Driving: Opportunities and Challenges. \u003cem\u003eProceedings of the IEEE 107\u003c/em\u003e (8), 1697\u0026ndash;1716\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTang J, Liu S, Liu L, Yu B, Shi W (2020) LoPECS: A Low-Power Edge Computing System for Real-Time Autonomous Driving Services. IEEE Access 8:30467\u0026ndash;30479\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIbn-Khedher H, Laroui M, Moungla H, Afifi H, Abd-Elrahman E (2022) Next-Generation Edge Computing Assisted Autonomous Driving Based Artificial Intelligence Algorithms. IEEE Access 10:53987\u0026ndash;54001\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim MS, Kim MS, Lee GJ, Sunwoo S-H, Chang S, Song YM, Kim D-H (2022) Bio-Inspired Artificial Vision and Neuromorphic Image Processing Devices. Adv Mater Technol 7(2):2100144\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eElouardi A, Bouaziz S, Dupret A, Lacassagne L, Klein JO, Reynaud R (2007) Image Processing Vision Systems: Standard Image Sensors Versus Retinas. IEEE Trans Instrum Meas 56(5):1675\u0026ndash;1687\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSon J, Kim S, Sohn K (2015) A multi-vision sensor-based fast localization system with image matching for challenging outdoor environments. Expert Syst Appl 42(22):8830\u0026ndash;8839\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang Z, Wan T, Ma S, Chai Y (2024) Multidimensional vision sensors for information processing. Nat Nanotechnol 19(7):919\u0026ndash;930\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChoi C, Lee GJ, Chang S, Song YM, Kim D-H (2024) Inspiration from Visual Ecology for Advancing Multifunctional Robotic Vision Systems: Bio-inspired Electronic Eyes and Neuromorphic Image Sensors. Adv Mater 36(48):2412252\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShan TJ, Kailath T (1988) Adaptive algorithms with an automatic gain control feature. IEEE Trans Circuits Syst 35(1):122\u0026ndash;127\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu C, Yan YP, Goh WL, Xiong YZ, Zhang LJ, Madihian M (2012) A 5-Gb/s Automatic Gain Control Amplifier With Temperature Compensation. IEEE J Solid-State Circuits 47(6):1323\u0026ndash;1333\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi X, Wang J, Yu F, Chen J, Chen X, Lu W, Li G (2025) Nonlinear memristive computational spectrometer. Light: Sci Appl 14(1):47\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang J, Li X, Gu J, Yu F, Chen J, Liu J, Song Y, Lin X, Chen X, Lu W, Li G (2025) Reconfigurable Ion-Migration Driven Memristor for Multistate Neuromorphic Associative Learning. ACS Photonics 12(5):2574\u0026ndash;2582\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi S, Zhang Z, Chen X, Deng W, Lu Y, Sui M, Gong F, Xu G, Li X, Liu F, You C, Chu F, Wu Y, Yan H, Zhang Y (2022) A High-Performance In-Memory Photodetector Realized by Charge Storage in a van der Waals MISFET. Adv Mater 34(10):2107734\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi T, Miao J, Fu X, Song B, Cai B, Ge X, Zhou X, Zhou P, Wang X, Jariwala D, Hu W (2023) Reconfigurable, non-volatile neuromorphic photovoltaics. Nat Nanotechnol 18(11):1303\u0026ndash;1310\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi X, Chen J, Yu F, Chen X, Lu W, Li G (2024) Achieving a Noise Limit with a Few-layer WSe2 Avalanche Photodetector at Room Temperature. Nano Lett 24(42):13255\u0026ndash;13262\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu F, Li Q, Wang P, Xia H, Wang Z, Wang Y, Luo M, Chen L, Chen F, Miao J, Chen X, Lu W, Shan C, Pan A, Wu X, Ren W, Jariwala D, Hu W (2019) High efficiency and fast van der Waals hetero-photodiodes with a unilateral depletion region. Nat Commun 10(1):4663\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang F, Zhang T, Xie R, Liu A, Dai F, Chen Y, Xu T, Wang H, Wang Z, Liao L, Wang J, Zhou P, Hu W (2024) Next-Generation Photodetectors beyond Van Der Waals Junctions. Adv Mater 36(3):2301197\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang J, Li X, Gu J, Yu F, Chen J, Lu W, Chen X (2025) High-Stability WSe2 Homojunction Photodetectors via Asymmetric Schottky and PIN Architectures \u003cem\u003eCoatings\u003c/em\u003e [Online]\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXu L, Xiong H, Fu Z, Deng M, Wang S, Zhang J, Shang L, Jiang K, Li Y, Zhu L, He L, Hu Z, Chu J (2021) High Conductance Margin for Efficient Neuromorphic Computing Enabled by Stacking Nonvolatile van der Waals Transistors. Phys Rev Appl 16(4):044049\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZou Y, Zhang Z, Yan J, Lin L, Huang G, Tan Y, You Z, Li P (2022) High-temperature flexible WSe2 photodetectors with ultrahigh photoresponsivity. Nat Commun 13(1):4372\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang J, Tse P-L, Jalil A-R, K\u0026ouml;lzer J, Rosenbach D, Luysberg M, Panaitov G, L\u0026uuml;th H, Hu Z, Gr\u0026uuml;tzmacher D, Lu JG, Sch\u0026auml;pers T (2021) Flux periodic oscillations and phase-coherent transport in GeTe nanowire-based devices. Nat Commun 12(1):754\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSong W, Chen J, Li Z, Fang X (2021) Self-Powered MXene/GaN van der Waals Heterojunction Ultraviolet Photodiodes with Superhigh Efficiency and Stable Current Outputs. Adv Mater 33(27):2101059\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJeong M-H, Ra H-S, Lee S-H, Kwak D-H, Ahn J, Yun WS, Lee J, Chae W-S, Hwang DK, Lee J-S (2022) Multilayer WSe2/MoS2 Heterojunction Phototransistors through Periodically Arrayed Nanopore Structures for Bandgap Engineering. Adv Mater 34(8):2108412\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYang X, Zhou X, Li L, Wang N, Hao R, Zhou Y, Xu H, Li Y, Zhu G, Zhang Z, Wang J, Feng Q (2023) Large-Area Black Phosphorus/PtSe2 Schottky Junction for High Operating Temperature Broadband Photodetectors. Small 19(28):2206590\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiao H, Wang X, Chen Y, Guo S, Wu S, Song C, Huang S, Huang X, Tai X, Lin T, Shen H, Yan H, Hu W, Meng X, Chu J, Zhang Y, Wang J HgCdTe/black phosphorus van der Waals heterojunction for high-performance polarization-sensitive midwave infrared photodetector. Sci Adv \u003cem\u003e8\u003c/em\u003e (19), eabn1811\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhu Y, Wang Y, Pang X, Jiang Y, Liu X, Li Q, Wang Z, Liu C, Hu W, Zhou P (2024) Non-volatile 2D MoS2/black phosphorus heterojunction photodiodes in the near- to mid-infrared region. Nat Commun 15(1):6015\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKim H, Uddin SZ, Lien D-H, Yeh M, Azar NS, Balendhran S, Kim T, Gupta N, Rho Y, Grigoropoulos CP, Crozier KB, Javey A (2021) Actively variable-spectrum optoelectronics with black phosphorus. Nature 596(7871):232\u0026ndash;237\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLee S, Peng R, Wu C, Li M (2022) Programmable black phosphorus image sensor for broadband optoelectronic edge computing. Nat Commun 13(1):1485\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWu S, Chen Y, Wang X, Jiao H, Zhao Q, Huang X, Tai X, Zhou Y, Chen H, Wang X, Huang S, Yan H, Lin T, Shen H, Hu W, Meng X, Chu J, Wang J (2022) Ultra-sensitive polarization-resolved black phosphorus homojunction photodetector defined by ferroelectric domains. Nat Commun 13(1):3198\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHe T, Ma H, Wang Z, Li Q, Liu S, Duan S, Xu T, Wang J, Wu H, Zhong F, Ye Y, Wu J, Lin S, Zhang K, Martyniuk P, Rogalski A, Wang P, Li L, Lin H, Hu W (2024) On-chip optoelectronic logic gates operating in the telecom band. Nat Photonics 18(1):60\u0026ndash;67\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang S, Jiang C, Yu Y, Zhang Z, Quhe R, Yang R, Tian Y, Chen X, Fan W, Niu Y, Yan B, Jiang C, Wang Y, Wang Z, Liu C, Hu W, Zhang J, Zhou P Tellurium nanowire retinal nanoprosthesis improves vision in models of blindness. Science \u003cem\u003e388\u003c/em\u003e (6751), eadu2987\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhou S, Fan H, Wen S, Wei Y, Chen H, Zhang Y, Zou T, Gao P, Wei D, Yin Y, Lan C, Li C, Liu Y (2025) Dual-Mode Photodetectors Mimicking Retinal Rod and Cone Cells for High Dynamic Range Image Sensor. \u003cem\u003eLaser \u0026amp; Photonics Reviews n/a\u003c/em\u003e (n/a), 2402192\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKocdemir IH, Akyuz AO, Koz A, Chalmers A, Alatan A, Kalkan S (2022) In \u003cem\u003eObject Detection for Autonomous Driving: High-Dynamic Range vs. Low-Dynamic Range Images\u003c/em\u003e, 2022 IEEE 24th International Workshop on Multimedia Signal Processing (MMSP), 26\u0026ndash;28 Sept. ; 2022; pp 1\u0026ndash;5\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMa C, Yuan S, Cheung P, Watanabe K, Taniguchi T, Zhang F, Xia F (2022) Intelligent infrared sensing enabled by tunable moir\u0026eacute; quantum geometry. Nature 604(7905):266\u0026ndash;272\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiao F, Zhou Z, Kim BJ, Chen J, Wang J, Wan T, Zhou Y, Hoang AT, Wang C, Kang J, Ahn J-H, Chai Y (2022) Bioinspired in-sensor visual adaptation for accurate perception. Nat Electron 5(2):84\u0026ndash;91\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePi L, Wang P, Liang S-J, Luo P, Wang H, Li D, Li Z, Chen P, Zhou X, Miao F, Zhai T (2022) Broadband convolutional processing using band-alignment-tunable heterostructures. Nat Electron 5(4):248\u0026ndash;254\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYoshioka K, Wakamura T, Hashisaka M, Watanabe K, Taniguchi T, Kumada N (2022) Ultrafast intrinsic optical-to-electrical conversion dynamics in a graphene photodetector. Nat Photonics 16(10):718\u0026ndash;723\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYoon HH, Fernandez HA, Nigmatulin F, Cai W, Yang Z, Cui H, Ahmed F, Cui X, Uddin MG, Minot ED, Lipsanen H, Kim K, Hakonen P, Hasan T, Sun Z (2022) Miniaturized spectrometers with a tunable van der Waals junction. Science 378(6617):296\u0026ndash;299\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang Y, Gu Y, Cui A, Li Q, He T, Zhang K, Wang Z, Li Z, Zhang Z, Wu P, Xie R, Wang F, Wang P, Shan C, Li H, Ye Z, Zhou P, Hu W (2022) Fast Uncooled Mid-Wavelength Infrared Photodetectors with Heterostructures of van der Waals on Epitaxial HgCdTe. Adv Mater 34(6):2107772\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang H, Wang W, Zhong Y, Li D, Li Z, Xu X, Song X, Chen Y, Huang P, Mei A, Han H, Zhai T, Zhou X (2022) Approaching the External Quantum Efficiency Limit in 2D Photovoltaic Devices. Adv Mater 34(39):2206122\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMoon G, Min SY, Han C, Lee S-H, Ahn H, Seo S-Y, Ding F, Kim S, Jo M-H (2023) Atomically Thin Synapse Networks on Van Der Waals Photo-Memtransistors. Adv Mater 35(4):2203481\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLai H, Lu Z, Lu Y, Yao X, Xu X, Chen J, Zhou Y, Liu P, Shi T, Wang X, Xie W, Fast (2023) Multi-Bit, and Vis-Infrared Broadband Nonvolatile Optoelectronic Memory with MoS2/2D-Perovskite Van der Waals Heterojunction. Adv Mater 35(6):2208664\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu X, Huang M, Zou X, Ali W, Rehman SU, Li J, Li Z, Xiang L, Pan A (2025) Alcohol-sensitive MoS2 optoelectronic synapses for mimicking human-like visual adaptation. \u003cem\u003eInfoMat n/a\u003c/em\u003e (n/a), e70019\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiang J, Xu W, Guo F, Yang S, Ge W, Shen B, Tang N (2023) Polarization-Resolved Near-Infrared PdSe2 p-i-n Homojunction Photodetector. Nano Lett 23(20):9522\u0026ndash;9528\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen Q, Zhang Y, Liu S, Han T, Chen X, Xu Y, Meng Z, Zhang G, Zheng X, Zhao J, Cao G, Liu G (2020) Switchable Perovskite Photovoltaic Sensors for Bioinspired Adaptive Machine Vision. Adv Intell Syst 2(9):2000122\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHong S, Choi SH, Park J, Yoo H, Oh JY, Hwang E, Yoon DH, Kim S (2020) Sensory Adaptation and Neuromorphic Phototransistors Based on CsPb(Br1\u0026ndash;xIx)3 Perovskite and MoS2 Hybrid Structure. ACS Nano 14(8):9796\u0026ndash;9806\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXie D, Wei L, Xie M, Jiang L, Yang J, He J, Jiang J (2021) Photoelectric Visual Adaptation Based on 0D-CsPbBr3-Quantum-Dots/2D-MoS2 Mixed-Dimensional Heterojunction Transistor. Adv Funct Mater 31(14):2010655\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu W, Yang X, Wang Z, Li Y, Li J, Feng Q, Xie X, Xin W, Xu H, Liu Y (2023) Self-powered and broadband opto-sensor with bionic visual adaptation function based on multilayer γ-InSe flakes. Light: Sci Appl 12(1):180\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHe Z, Shen H, Ye D, Xiang L, Zhao W, Ding J, Zhang F, Di C-a, Zhu D (2021) An organic transistor with light intensity-dependent active photoadaptation. Nat Electron 4(7):522\u0026ndash;529\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi L, Li S, Wang W, Zhang J, Sun Y, Deng Q, Zheng T, Lu J, Gao W, Yang M, Wang H, Pan Y, Liu X, Yang Y, Li J, Huo N (2024) Adaptative machine vision with microsecond-level accurate perception beyond human retina. Nat Commun 15(1):6261\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang S, Han L, Xiao K, Zhang L, Shi C, Xu L, Deng K, Zou Y, Jiang M, Lv X, Zhu Y, Li Q, Liu C, Tang W, Li G, Ding S, Chen X, Lu W (2023) H-BN-Encapsulated Uncooled Infrared Photodetectors Based on Tantalum Nickel Selenide. Adv Funct Mater 33(48):2305380\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eXin W, Li X-K, He X-L, Su B-W, Jiang X-Q, Huang K-X, Zhou X-F, Liu Z-B, Tian J-G (2018) Black-Phosphorus-Based Orientation-Induced Diodes. Adv Mater 30(2):1704653\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhao T, Chen Y, Xu T, Zhong F, Yu Y, Ma H, Zhang K, Duan S, Hu J, Wang S, Guo J, Wang Z (2024) Topological Insulator Bi2Se3 Heterojunction with a Low Dark Current for Midwave Infrared Photodetection. ACS Photonics 11(6):2450\u0026ndash;2458\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Self-adaptive photodetector, Van der Waals heterostructure, Electrostatic barrier engineering, Nonlinear response, Neural control","lastPublishedDoi":"10.21203/rs.3.rs-7289222/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7289222/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eBiological vision relies on eye-mediated gain control to adapt across lighting conditions\u0026mdash;but remains fundamentally blind to infrared wavelengths and polarization. Here, we report a neuromorphic photodetector that not only emulates this self-adaptive functionality, but surpasses human vision by enabling dynamic gain regulation across the infrared\u0026ndash;polarization domain. Using a gate-tunable Au/BP/PdSe\u003csub\u003e2\u003c/sub\u003e van der Waals heterostructure (vdWH), we achieve eye-like nonlinear gain compression via electrostatic barrier reconfiguration, which enables dynamic modulation of both the response area and responsivity. Integrated with a neural-network-based microcontroller, the system forms a device-level closed-loop that autonomously adjusts optical gain in real time. This expands the linear dynamic range (LDR) by three orders of magnitude, reaching\u0026thinsp;~\u0026thinsp;80 dB at 1550 nm, with sub-millisecond response and intrinsic polarization sensitivity (PR\u0026thinsp;\u0026asymp;\u0026thinsp;8)\u0026mdash;all without external optics or analog circuitry. These results establish a scalable, intelligent optoelectronic platform that augments biological perception and advances chip-scale self-adaptive vision for autonomous sensing and edge photonic intelligence.\u003c/p\u003e","manuscriptTitle":"Self-Adaptive Infrared Vision via Neural-Controlled Gain Compression in a Single Photodetector","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-28 09:08:44","doi":"10.21203/rs.3.rs-7289222/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"8b4b5f7e-fe5f-440d-9512-6589325ded35","owner":[],"postedDate":"August 28th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":53758308,"name":"Physical sciences/Optics and photonics/Applied optics/Optical sensors"},{"id":53758309,"name":"Physical sciences/Optics and photonics/Applied optics/Optoelectronic devices and components"}],"tags":[],"updatedAt":"2025-09-12T07:05:53+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-28 09:08:44","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7289222","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7289222","identity":"rs-7289222","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}