Threshold-tunable event-driven vision sensor for adaptive visual processing

Threshold-tunable event-driven vision sensor for adaptive visual processing | 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 Threshold-tunable event-driven vision sensor for adaptive visual processing Bowen Zhu, Guanlei Zhao, Siyu Zhang, Zhongfang Zhang, Yingjie Tang, and 9 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9334649/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract Event-driven vision sensors provide high-temporal-resolution perception with minimal data redundancy, offering significant advantages for machine vision applications. However, conventional designs suffer from high circuit complexity or limited adaptability in complex dynamic scenes. Herein, we present a neuromorphic vision sensor based on MoTe 2 /In 2 O 3 heterojunction phototransistor, enabling retina-like adaptive visual processing. The device achieves microsecond-scale response (1.2/1.0 µs) and a dynamic event threshold range (DETR) exceeding 60 dB. This tunability allows in-sensor processing, such as noise suppression and feature enhancement, eliminating the need for external circuitry. System-level validation is achieved using a 6×6 sensor array, which demonstrates denoising effect for raindrop noise in rainy environments and the enhancement of weak signals in foggy conditions. Additionally, we introduce an event-entropy metric to evaluate the effects of threshold modulation on the event stream and establish its quantitative connection to information efficiency. By co-designing materials, devices, and circuits with the novel entropy metric, this work provides a scalable framework for building adaptive, robust, and highly integrated event-based vision systems. Physical sciences/Engineering/Electrical and electronic engineering Physical sciences/Nanoscience and technology/Nanoscale devices/Electronic devices Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Machine vision 1 – 3 underpins the perceptual and decision-making capabilities of intelligent systems 4 – 8 , requiring vision sensors that offer high temporal precision, low latency, and strong adaptability. Frame-based cameras 9 , 10 , which operate on fixed-exposure and global-sampling principles, struggle with static data redundancy and computational inefficiency. In contrast, neuromorphic event-driven vision sensors 11 , 12 , inspired by the spatiotemporal processing of the human retina 13 – 15 , shift the paradigm towards dynamic, information-centric perception and a cornerstone of neuromorphic computing systems 16 – 18 . These sensors operate asynchronously, recording only local intensity changes as sparse spike signals. This approach minimizes data redundancy and markedly enhances the processing of temporal visual information. Conventional circuit-based dynamic vision sensors (DVS) provide significant advantages, including compatibility with complementary metal-oxide-semiconductor (CMOS) technology, microsecond-level temporal resolution, and strong environmental adaptability benefited from the tunable threshold of the comparator circuit 19 . However, the asynchronous circuits, composed of photoreceptors, amplifiers and comparators, require each pixel to function as an independent sensing and processing unit. This complexity limits pixel density and fill factor, while introducing thermal noise and power overhead, especially in static scenes. Recent advances in event-driven vision sensors have focused on simplifying circuit architecture within the pixel. Examples include the two-transistor-two-resistor-one-capacitor (2T2R1C) and band-engineered semiconductor devices that enable one-photosensor-one-capacitor (1P1C) or carrier-blocking event-generation mechanisms 20 – 24 . However, the fixed threshold eliminates the ability to adjust sensitivity dynamically according to environmental conditions, which creates a practical dilemma: under noisy or low-light conditions, a fixed low threshold triggers excessive spurious events, whereas in high-illumination or low-contrast scenes, a fixed high threshold suppresses meaningful signals. While achieving structural simplicity, these designs sacrifice perceptual adaptability, undermining the robustness of neuromorphic perception and encoding complex real-world environments 25 – 27 . Consequently, a longstanding challenge is the inability to integrate adaptive perception into a structurally simple device. In this work, we demonstrate an in-sensor event-driven vision system with intrinsically tunable event thresholds enabled by a defect-engineered capacitive mechanism embedded in a single MoTe 2 /In 2 O 3 heterojunction phototransistor. The device exhibits microsecond-scale response times (1.2/1.0 µs) and a high dynamic range exceeding 82 dB. It further enables continuous and wide-range event-threshold modulation, with a dynamic event threshold range (DETR) exceeding 60 dB (10 µW cm - 2 to 11.7 mW cm - 2 ). A system-level validation was achieved using a 6×6 dynamic sensor array, showing robust denoising and feature enhancement under adverse conditions such as rain and fog. Furthermore, we introduce an event entropy metric that quantitatively links threshold modulation to information efficiency, providing a unified framework for evaluating and optimizing the information density of asynchronous event streams. This work provides an effective strategy for developing bioplausible event-based visual sensors. Bio-inspired adaptive event-driven vision sensor with in-sensor threshold modulation The human visual system achieves robust perception in dynamic environments through hierarchical adaptation spanning the retina 28 and cortex 29 , 30 , a principle that neuromorphic hardware seeks to emulate 31 – 33 . As illustrated in Fig. 1a, at the retinal level, the retina spiking neuron (RSN) adjust their response gain to maintain a wide dynamic range and stable encoding under varying luminance. At higher stages, visual cortex (VC) feedback further modulates neuronal excitability and perceptual thresholds in a task-dependent manner, enabling adaptive visual processing 34 , 35 . Inspired by the biological visual system, DVS captures temporal dynamics through sophisticated circuit architecture. Threshold-control units modulate detection sensitivity across varying illumination conditions, thereby regulating event responses. However, this circuit-centric methodology increases system complexity, resulting in additional power consumption and time delay. Integrating event-spiking generation and threshold tunability within a single device addresses the architectural complexity and latency of conventional event-driven sensors (Fig. 1b). This top-down neural visual pathway can be physically realized within a single heterojunction transistor, where the heterojunction encodes visual stimuli into photo-induced spike signals, and the gate serves as a control terminal that modulates carrier density and band bending to regulate spike amplitude and thus the event threshold (ETS). Capacitive behavior is essential for spike generation; however, recent approaches 22 – 24 rely on an additional insulating layer that introduce interfacial barriers and parasitic capacitance, inevitably suppressing carrier coupling and attenuating spike amplitude. In contrast, we harness defect-state capacitance intrinsically formed within MoTe 2 36 , which directly couples with photogenerated carriers, enabling spike generation while preserving high photoresponse in a structurally simplified architecture. The In 2 O 3 37, 38 channel provides electrostatic control over carrier density and interfacial band alignment. Gate modulation alters the occupation of defect states and the associated trap-detrap dynamics, thereby tuning the effective carrier capture process. Therefore, MoTe 2 /In 2 O 3 heterojunction phototransistor intrinsically integrates event-spiking generation and ETS modulation within a single device. Figure 1 | Threshold-tunable event-driven vision sensor for high-order perceptual processing. a , Schematic illustration of hierarchical adaptation in the biological visual system. b , Device architecture and working mechanism of the MoTe 2 /In 2 O 3 heterojunction phototransistor. Electron-beam-induced defect states in MoTe 2 act as hole trapping centers and integrate it with an In 2 O 3 channel to realize a heterojunction transistor with gate-controlled threshold-tunable threshold characteristics. c , Equivalent circuit model of the device. d , Gate-controlled threshold modulation of the device. e , Representative event responses under different threshold settings, illustrating the transition between noise suppression and detail enhancement regimes through threshold modulation. f , Relationship between ETS and event entropy: event entropy increases with decreasing ETS, as a lower threshold facilitates event-spiking generation and leads to a more distributed event representation. g , Hardware array ETS processing for complex driving scenarios: increasing the threshold during rainy conditions and decreasing it during foggy conditions. h , Event-entropy-guided regulation of event imaging. In rainy conditions, increased threshold suppresses noise events and reduces entropy, while in foggy conditions, decreased threshold enhances weak signals and edge features, improving overall perception efficiency. Figure 1c illustrates the equivalent circuit model of the device. The MoTe 2 /In 2 O 3 heterojunction is represented by a trap-mediated charge storage element that behaves analogously to a capacitance, arising from the dynamic trapping and release of carriers at defect states. This effective capacitance is modulated by the gate voltage ( V GS ), which tunes the occupation of defect states and the associated trapping dynamics. The device operates under zero drain bias ( V DS = 0), with event sensing governed by the intrinsic trap-mediated charge storage behavior. Threshold modulation is achieved via V GS , which modulates the carrier density and defect-state occupation, thereby altering the effective carrier trapping dynamics. As a result, the ETS can be continuously adjusted: increasing V GS suppresses carrier trapping and raises the threshold, while decreasing V GS enhances trapping and lowers the threshold (Fig. 1d). This dynamically tunable threshold enables adaptive response to varying illumination conditions. The device enables flexible event encoding through ETS modulation, where the optimal threshold can be adaptively tuned to enable noise suppression and weak-signal enhancement for different application scenarios (Fig. 1e). To quantify the effect of threshold modulation, we introduce event entropy as a metric, drawing inspiration from the concept of image entropy 39 . Event entropy evaluates the information content of an event stream based on the joint probability distribution of events across spatial blocks and polarity. By partitioning the sensor array into spatial blocks and computing the probabilities of positive and negative events within each block, we obtain the expression for event entropy: $$\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:H(X,S)=-\sum\:_{x=1}^{N}\sum\:_{s\in\:\left\{+1,\:\:-1\right\}}p(x,s){\text{log}}_{2}p\left(x,s\right)\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(1\right)$$ where \(\:x\:\) indexes the spatial blocks, \(\:s\:\) denotes event polarity, and \(\:p\left(x,s\right)\:\) is the probability of observing an event of polarity \(\:s\:\) in block \(\:x\) . Event entropy quantifies the complexity and richness of the event distribution. The detailed definition and calculation of event entropy are provided in Supplementary Note Ⅰ. A higher event entropy corresponds to a more distributed event pattern with richer information content, whereas a lower value reflects a more concentrated distribution associated with noise suppression (Fig. 1f). This metric enables quantitative evaluation of the event stream under different threshold settings and establishes a direct link between threshold modulation and the dynamic information density of the event stream. Consequently, threshold modulation enables continuous control over event-stream encoding, transitioning between noise suppression and detail enhancement regimes, thereby improving robustness under dynamic visual conditions. Leveraging the environmental adaptability, we develop a vision sensing hardware system based on a MoTe 2 /In 2 O 3 device array, enabling real-time dynamic threshold configuration for processing complex visual scenes (Fig. 1g). For two contrasting driving scenarios, namely rainy and foggy conditions, the system adaptively increases or decreases the threshold to regulate the event response. This enables effective suppression of rain-induced noise and enhancement of blurred vehicle features, thereby improving visual perception efficiency through coordinated sensing and processing. Building on the gate-modulated threshold engineering described above, we next examine the impact of adaptive threshold control on event-based imaging at the event entropy scale (Fig. 1h). In rainy conditions, excessive rain-induced events introduce redundant noise, which can be suppressed by increasing the threshold, resulting in reduced event entropy and improved information efficiency. In contrast, under foggy conditions, lowering the threshold enhances sensitivity to weak signals, strengthening edge information and recovering fine details. Event entropy quantitatively characterizes the distribution of the event stream and serves as an effective metric for evaluating imaging performance under different threshold settings. Defect engineering in MoTe 2 MoTe 2 has attracted considerable attention due to its broadband spectral response and strong light absorption. Previous studies have shown that electron-beam processes can effectively modulate the properties of MoTe 2 thin films 40 . In this work, deep-level defect states are introduced into the MoTe 2 film via electron-beam treatment, forming hole trapping centers that enable intrinsic charge storage analogous to a capacitance, without requiring additional circuit components. The transient response arises from dynamic trapping and release of photogenerated carriers. Structural defects are intrinsically introduced during the electron-beam evaporation of the top electrode. The relatively weak Mo-Te bonding facilitates the formation of Te vacancies and associated deep-level trap states under electron-beam exposure 41 . These defect states dominate carrier relaxation and enable pronounced transient photoresponses under time-varying illumination. This defect formation is directly evidenced by both structural and chemical characterizations. Transmission electron microscopy (TEM) reveals that the pristine MoTe 2 film, which initially exhibits a well-defined layered structure, undergoes pronounced lattice misalignment and interlayer distortion after electron-beam evaporation (EBE), indicating substantial lattice disruption and disorder (Fig. 2 a, b). The relatively weak Mo-Te bonding facilitates the formation of Te vacancies and associated deep-level trap states 42 . Consistent with this, X-ray photoelectron spectroscopy (XPS) analysis reveals a decrease in the Te:Mo atomic ratio from 1.94:1 to 1.48:1, confirming the generation of Te vacancies and corresponding deep trap states (Fig. 2 c). A comparison of the elemental valence states and compositions before and after deposition is provided in Supplementary Fig. 1. These structural and compositional changes translate into markedly altered excitonic and carrier dynamics. Transient absorption measurements reveal that evaporated MoTe 2 exhibits a significantly stronger and longer-lived bleaching signal at 703 nm compared with its pristine counterpart (Fig. 2 d, e). The stronger and longer-lived bleaching signal indicates the presence of deep trap states, which significantly alter the excitonic dynamics by trapping carriers more effectively, leading to prolonged signal decay times (Fig. 2 f). The prolonged decay reflects suppressed carrier recombination due to deep trapping, which inhibits rapid relaxation back to the ground state. As a result, transient perturbations in illumination are amplified, whereas static or slowly varying backgrounds are intrinsically suppressed. Event-driven sensing mechanism and gate voltage-modulated threshold tunability Benefiting from this defect-engineering strategy, we develop a MoTe 2 /In 2 O 3 heterojunction phototransistor. In this lateral architecture, an In 2 O 3 channel bridges the source and drain, while a MoTe 2 layer partially overlaps the channel near the source to form a heterojunction. Owing to the low Mo-Te bond energy, EBE introduces a high density of defect states into the MoTe 2 layer, forming an intrinsic charge-trapping centers. Photogenerated carriers are transiently captured and released in response to illumination changes, giving rise to pronounced bipolar photocurrent spikes, while relaxing to a near-zero steady-state baseline under constant illumination. This material-level charge accumulation and release mechanism enables direct event-spiking generation within a single transistor, without relying on external capacitors or peripheral signal-conditioning circuits. The heterostructure combines narrow-bandgap MoTe 2 , which provides broadband optical absorption, with wide-bandgap In 2 O 3 , which offers high carrier mobility and strong gate electrostatic control. The resulting band alignment forms an interfacial barrier whose height can be efficiently modulated by the gate through carrier density tuning in In 2 O 3 , enabling dynamic regulation of the event-generation threshold. In this way, defect-mediated trapping dynamics and gate-controlled transport are synergistically integrated, allowing circuit-level functionalities to be intrinsically realized within a single device. The light-induced event-driven sensing process proceeds through three distinct stages (Fig. 3 a). Upon illumination onset (stage I), incident 520 nm light is predominantly absorbed by MoTe 2 , generating electron-hole pairs. Operating at zero drain-source bias ( V DS =0 V), the photoresponse arises solely from carrier separation by the built-in electric field at the MoTe 2 /In 2 O 3 heterointerface: photogenerated electrons are rapidly injected into the In 2 O 3 channel, while holes drift toward the electrodes or are trapped in defect states within MoTe 2 . This abrupt interfacial charge separation produces a positive transient photocurrent pulse, corresponding to an ON event. During continuous illumination (stage II), photocarrier generation balances defect-state trapping and detrapping. This results in a relaxation of the photocurrent towards a near-baseline level, maintaining selective sensitivity to temporal variations while rejecting static background signals. Upon illumination removal (stage III), electrons accumulated in the In 2 O 3 channel recombine with holes previously trapped in MoTe 2 , generating a negative transient pulse that marks an OFF event. Collectively, these observations indicate that the transient photocurrent spiking is primarily governed by the capacitive trapping and release of carriers at the defect states in MoTe 2 , a process reinforced by the heterojunction band alignment that directs charge transfer across the interface. This mechanism fundamentally differs from capacitor-dependent designs and enables event spiking within a heterojunction. The physical image of the device is shown in Fig. 3 b. The device exhibits pronounced transient responses over a wide range of illumination levels. Stepwise changes in light intensity generate distinct current pulses that reliably return to baseline once illumination stabilizes (Fig. 3 c). Pulse amplitude increases monotonically with illumination contrast, indicating a robust correspondence between stimulus variation and event strength (Fig. 3 d). As V GS = -10 V and under 800 nm illumination, the device exhibits a dynamic range of over 82 dB (Supplementary Fig. 2). It exceeds the range of the human retina, indicating the superior ability to detect both weak and strong stimuli without saturation. The device exhibits a rapid photoelectric pulse response, with a response time of 1.2 µs for positive pulses and 1.0 µs for negative pulses (Supplementary Fig. 3). In addition, owing to its narrow bandgap, the wide spectral response of MoTe 2 , spanning from 360 to 1064 nm, is particularly advantageous for event-driven vision sensors as it allows for reliable operation under diverse lighting conditions, from visible light to infrared, thereby enhancing the sensor’s versatility and robustness in real-world environments. (Supplementary Fig. 4). Beyond event-spiking generation, the lateral heterojunction architecture enables active electrical modulation of the ETS via V GS , a capability that is fundamentally absent in conventional vertical designs. By extracting the minimum detectable illumination change as a function of the V GS from − 10 to 0 V, the device achieves a DETR exceeding 60 dB, corresponding to illumination variations from 10 µW/cm 2 to 11.7 mW/cm 2 . This range covers the vast majority of lighting conditions encountered in natural environments (Fig. 2 e). This enhanced dynamic range is crucial for applications in real-world environments where illumination varies drastically. This gate-modulated behavior arises from the tunable electric field of the heterojunction. At negative gate biases ( V GS =-15 V), reduced carrier density in In 2 O 3 modifies the interfacial band alignment while leaving the MoTe 2 electrostatic potential largely unchanged. This enhances the effective junction field, promoting more efficient charge separation. As V GS approaches 0 V, the junction field strength diminishes (Fig. 2 f). The mechanism analysis of gate-modulated heterojunction electric field strength is shown in Supplementary Fig. 6. By modulating the strength of the heterojunction electric field, we achieve in-sensor event information processing (Fig. 2 g). The tunable response intensity allows for a functional switch between signal enhancement and noise suppression. From an event-stream perspective, threshold modulation regulates event density and temporal sparsity. Lower thresholds enhance sensitivity to weak features, making it ideal for low-contrast scenes, while higher thresholds suppress spurious events caused by noise or rapid background fluctuations. Electrically controlled modulation of the heterojunction field enables flexible regulation of event-spiking generation within a single transistor. By overcoming the fixed-threshold limitations of conventional designs, this approach provides a robust physical foundation for environmentally adaptive, threshold-tunable event-driven vision sensors. Hardware implementation and system-level validation of threshold-tunable event-driven vision sensor array To validate the practical deployability of the gate-tunable event mechanism at the system level, we integrated a 6×6 MoTe 2 /In 2 O 3 threshold-tunable event-driven vision sensor (TEVS) array with a designed printed circuit board (PCB) platform for real-time acquisition and processing (Fig. 4a). The system comprises a photosensor array, a dedicated gate-bias PCB for global threshold control, and an acquisition PCB for signal conditioning and digitization. Figure 4. | Hardware implementation and system-level validation of the TEVS array. a , Schematic of the 6×6 MoTe 2 /In 2 O 3 TEVS array integrated with a custom-designed PCB platform for real-time acquisition and processing of event data. b , Diagram of the system-level signal flow from the sensor array to host computer imaging. c , Representative signals recorded through the acquisition chain, illustrating the process of photocurrent conversion, amplification, threshold detection and pulse shaping. d , Single-pixel event imaging experiments demonstrate the motion of a car model captured by a single phototransistor. e , Event imaging results of the car model. The system generates images from 3 different positions, with event signatures encoded solely through the temporal sequence of ON and OFF events. Figure 4b illustrates the system-level signal flow from on-board PCB sampling to host-computer imaging. The photocurrent signals are first converted to voltage and amplified using trans-impedance amplifiers (TIAs), followed by polarity-selective comparators for threshold detection and pulse shaping, after which event-based images are displayed on the host computer. The layout and detailed circuit design of the PCB platform are provided in Supplementary Figs. 12 and 13. This TEVS-PCB design thereby provides a compact and scalable system-level solution, eliminating the need for per-pixel differential amplifiers and comparators. And it partially addresses the long-standing challenge of requiring independent circuit units per pixel in conventional event cameras. Representative signals recorded along the acquisition chain are presented in Fig. 4c. The MoTe 2 /In 2 O 3 phototransistor generates bipolar, spike current transients in response to illumination changes. The transient signals are captured directly using an oscilloscope. The original signals, being relatively small in amplitude, are susceptible to interference such as lead-injected noise, which could distort the waveform. To preserve the intrinsic event dynamics, the current pulses are processed with an unfiltered transimpedance amplifier that provides gain without temporal smoothing. The resulting voltage pulses exhibit a high intrinsic signal-to-noise ratio for both ON and OFF events. This indicates that the observed noise predominantly originates from the measurement setup (e.g., the noise of the oscilloscope leads), whereas the MoTe 2 /In 2 O 3 phototransistor array itself generates events with high fidelity and low intrinsic noise. A subsequent comparator stage converts these supra-threshold events into clean digital pulses, providing reliable polarity discrimination. Single-pixel event imaging of a moving object provided a stringent system-level validation of our device-circuit design (Fig. 4d, e). In the measurement, a moving car model traverses the active area of a single phototransistor, and its motion is encoded exclusively through the temporal sequence of ON and OFF events generated by illumination changes. Despite the absence of spatial multiplexing, distinct object geometries and trajectories produce reproducible and characteristic temporal event signatures. Variations in object length, shape, and traversal speed are reflected in the event density, polarity sequence, and inter-event timing, enabling discrimination based purely on temporal information. Real-time imaging of the vehicle model is presented in Supplementary Video 1. These results demonstrate that meaningful visual information can be extracted without frame reconstruction or multi-pixel correlation. Instead, perception emerges from the precise timing of sparse events at a single sensing node, directly reflecting the core operating principle of event-driven vision. In-sensor event-stream processing via threshold modulation We next demonstrate array-level in-sensor event-stream processing enabled by gate-controlled threshold modulation using a 6×6 MoTe 2 /In 2 O 3 TEVS array integrated with a custom PCB platform. The experimental configuration and representative real-time outputs are shown in Fig. 5 a. A 100 Hz modulated laser illuminates the array through a spatial mask made of optical films with different transmittances. This setup simultaneously emulates both the spatially varying attenuation (e.g., fog) and the stochastic noise (e.g., rain) found in challenging environmental conditions. The gate voltage V GS is adjusted in real time to globally tune the ETS across the array, without modifying the readout circuitry. Event entropy was computed by defining a block of pixels (2×2 for the 6×6 array) and calculating the probability of event occurrence within each block over time. This block size balances assessment accuracy with computational load. In the rain-mimicking scenario (Fig. 5 b), the mask generates abundant noise-induced events when the threshold is set low ( V GS = -12 V). Under this condition, both noise and object-related signals are indiscriminately captured, resulting in a high event entropy of 2.5 bits and a reduced imaging accuracy of 73.9% for the car model. Raising the threshold by adjusting V GS to -6 V effectively suppresses rain-induced noise, reducing event entropy to 2.0 bits and restoring imaging accuracy to 100%. This result illustrates how threshold elevation selectively filters noise while preserving salient object features. Conversely, in the fog-mimicking scenario (Fig. 5 c), a high threshold ( V GS = -5 V) renders many pixels unresponsive to weak intensity variations, leading to incomplete object representation, a low event entropy of 1.3 bits, and an imaging accuracy of 76.5%. Lowering the threshold to V GS = -12 V increases sensitivity to these subtle variations, which raises the event entropy to 2.0 bits and recovers the full structure of the car model with 100% accuracy. This illustrates how threshold reduction enhances weak feature detection. The dynamic threshold modulation process is demonstrated in the Supplementary Video 2 and 3, showing the dynamic adjustments in real-time. Together, these results demonstrate that gate-controlled threshold modulation can dynamically balance noise suppression and feature enhancement based on environmental conditions, offering task-oriented event-stream processing. To further assess the scalability and generality of threshold modulation and event entropy, we performed event-stream simulations on a 1280×720 pixel array derived from real-world driving scenes under rain and fog, respectively (Fig. 5 d, e). Given the high pixel density, a 16×16 pixel region is adopted as the block size to ensure computational efficiency. The ETS is defined as the ratio of the variation in grayscale intensity to the total grayscale intensity. In the rain scenario, a low ETS (3%) produces dense noise-dominated event streams, whereas increasing the threshold to 5% suppresses rain-induced interference and reduces event entropy from 10.8 to 10.0 bits, substantially improving perceptual clarity. In contrast, under foggy conditions, an overly high threshold (12%) leads to severe loss of structural information, while lowering the threshold to 4% recovers fine vehicle contours and enhances scene interpretability, with the event entropy increasing from 6.7 to 7.6 bits. Although obtained from large-scale simulations, the threshold values are directly mapped from experimentally validated gate modulation, highlighting the scalability of the proposed physical mechanism. Collectively, these results establish threshold modulation as a powerful in-sensor mechanism for event-stream processing. By dynamically regulating event density and information content in response to environmental conditions, the proposed sensor enables robust perception without relying on external computation or complex circuitry. Event entropy serves as a quantitative bridge linking physical threshold control to perceptual performance, providing a valuable metric and tool for evaluating and optimizing event-based vision in complex real-world environments. Conclusions In conclusion, we demonstrate a tunable-threshold event-driven vision sensor based on a single MoTe 2 /In 2 O 3 heterojunction phototransistor. The device achieves microsecond-scale response (1.2/1.0 µs) and a DETR exceeding 60 dB, which enables in-sensor event processing for noise suppression and feature enhancement without external capacitors or threshold-control circuits. System-level validation using a 6×6 sensor array confirms real-time imaging performance in complex dynamic environments. This work establishes single-transistor tunable-threshold photodetection as a minimal functional building block that collapses circuit complexity into device-level functionality, while introducing event entropy as a quantitative framework linking threshold modulation to information efficiency, forming a basis for the design of event-driven vision architectures and the characterization of information density in asynchronous event streams. Methods Materials Indium nitrate hydrate (In(NO 3 ) 3 ·xH 2 O, 99.999%) powder was purchased from Sigma-Aldrich. 2-methoxyethanol (2-ME, 99.3%), acetylacetone (AcAc, 99%) and ammonium hydroxide (NH 3 ·H 2 O, 28%) were purchased from Alfa Aesar. 2H-MoTe 2 , grown via chemical vapor deposition (CVD) was obtained from SixCarbon Technology Shenzhen. Preparation of InO precursor solution 0.3 g In(NO 3 ) 3 ·xH 2 O was dissolved in 10 mL 2-ME. Then, add 100 µL AcAc and 35 µL NH 3 ·H 2 O in the solution as additives. After that, stir the In 2 O 3 precursor solution at room temperature for 6 hours. MoTe 2 /In 2 O 3 phototransistor fabrication The In 2 O 3 precursor solution was spin coated at 3000 rpm for 40 s on SiO 2 /p-Si (100 nm/500 µm) substrate after O 2 plasma cleaning. It was then annealed at 200°C for 10 mins. And use the maskless lithography machine to design the In 2 O 3 pattern and etch the film with a solution of HCl: H 2 O (1:15 volume ratio). After that, put the film on the heating stage to anneal at 300°C for 1 hour in air. 2H-MoTe 2 film was transferred onto an In 2 O 3 /SiO 2 /Si substrate using a wet transfer method and was annealed at 200°C for 2 hours to improve the interface contact. Next, use the maskless lithography machine to design the In 2 O 3 pattern and etch the MoTe 2 film by Ar and O 2 inductively coupled plasma. Finally, the source-drain Ni/Au (8/50 nm) electrodes were deposited through EBE. The W/L ratio is 400/20 µm. Material and device characterizations Field emission high resolution images were obtained by transmission electron microscopy (Talos F200X G2). The elemental composition analysis of MoTe 2 and In 2 O 3 thin films was performed using micro-focus X-ray photoelectron spectroscopy (Thermo Scientific, Nexsa G2). Ultrafast transient absorption fluorescence microspectroscopy system (PH-Tuning, Light Conversion and TA100, Time Tech Spectra) was used to measure the corresponding defects within MoTe 2 . The transmittance of MoTe 2 and In 2 O 3 thin films was measured using a UV-VIS-NIR microspectrophotometer (CRAIC Technologies Inc, CRAIC 20/30PV). The electrical characteristics of MoTe 2 /In 2 O 3 phototransistor was measure by the Agilent B1500 semiconductor parameter analyzer. The incident light at multiple wavelengths is generated by lasers with wavelengths of 360/450/520/640/800/1064 nm (Changchun New Industries Optoelectronics Tech Co., Ltd, China) and modulated by an arbitrary waveform generator (Tektronix AFG3152C). The light intensity was calibrated by Thorlabs S120VC standard silicon photodiode. Formula for calculating the dynamic range of MoTe 2 /In 2 O 3 phototransistors $$\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:DR\left(dB\right)=20{log}_{10}\left(\frac{{S}_{max}}{{S}_{min}}\right)\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(2\right)\:$$ \(\:{S}_{max}\:\) is the maximum signal intensity that the detector can process without saturation, while \(\:{S}_{min}\:\) is the minimum signal intensity that the detector can detect. Here, the signal intensity refers to the change in light intensity. The dynamic range (DR) physically represents the span of input signal intensities over which a system can reliably detect and process signals, from the minimum discernible level above noise to the maximum level before saturation. Formula for calculating the DETR of MoTe 2 /In 2 O 3 phototransistors $$\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:DETR\left(dB\right)=20{log}_{10}\left(\frac{{ETS}_{max}}{{ETS}_{min}}\right)\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\left(3\right)$$ \(\:{ETS}_{max}\) and \(\:{ETS}_{min}\) represent the minimum detectable light intensity change under different gate biases. DETR represents the gate-controlled dynamic range of ETS which the detector can reliably modulate its minimum detectable illumination change. Declarations Data Availability The main data supporting the findings of this study are available within the Article. Source data are provided with this paper. Additional data and information are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interest. Author contributions B.Z., H.W., G.Z. conceived and designed the experiments. G.Z., Z.Z. designed and fabricated the MoTe 2 /In 2 O 3 phototransistor array and conduct material mechanism characterization. S.Z. designed the PCB acquisition circuit and an upper computer display system for event imaging. All authors contributed to the preparation of the manuscript. B.Z., H.W. supervised the project. Acknowledgments This work is supported by the National Key Research and Development Program of China (2024YFA1208800) funded by MOST, the National Natural Science Foundation of China (Grant Number. 62574169, 62504208), and Scientific Research Project of Westlake University Westlake (No. WU20248032). We thank the Westlake Centre for Micro/Nano Fabrication, the Instrumentation and Service Centre for Physical Sciences (ISCPS), and the Instrumentation and Service Centre for Molecular Sciences (ISCMS) at Westlake University for the facility support and technical assistance. And we thank Dr. Xue LOU from Instrumentation and Service Center for Molecular Sciences at Westlake University for the assistance of transient absorption spectroscopy measurement. References Mennel L et al (2020) Ultrafast machine vision with 2D material neural network image sensors. Nature 579:62–66 Tan H, van Dijken S (2023) Dynamic machine vision with retinomorphic photomemristor-reservoir computing. Nat Commun 14:2169 Chen R et al (2026) Integrated bionic LiDAR for adaptive 4D machine vision. Nat Commun 17:24 Gehrig D, Scaramuzza D (2024) Low-latency automotive vision with event cameras. Nature 629:1034–1040 Yang Z et al (2024) A vision chip with complementary pathways for open-world sensing. Nature 629:1027–1033 Pan J et al (2025) Adaptive spatial-temporal information processing based on in-memory attention-inspired devices. Nat Commun 16:7449 Huang H et al (2025) In-sensor compressing via programmable optoelectronic sensors based on van der Waals heterostructures for intelligent machine vision. Nat Commun 16:3836 Paredes-Valles F et al (2024) Fully neuromorphic vision and control for autonomous drone flight. Sci Robot 9:eadi0591 Hirayama T (2013) &IEEE. The Evolution of CMOS Image Sensors. In Asian Solid-State Circuits Conference 5–8IEEE Yang M et al (2025) Dynamic range and precision of hybrid vision sensors. Nature 642:E7–E8 Lin Q et al (2025) Event-driven retinomorphic photodiode with bio-plausible temporal dynamics. Nat Nanotechnol 20:1213–1220 Zhou Y et al (2023) Computational event-driven vision sensors for in-sensor spiking neural networks. Nat Electron 6:870–878 Euler T, Haverkamp S, Schubert T, &Baden T (2014) Retinal bipolar cells: elementary building blocks of vision. Nat Rev Neurosci 15:507–519 Gollisch T, Meister M (2010) Eve Smarter than Scientists Believed: Neural Computations in Circuits of the Retina. Neuron 65:150–164 Masland R (2012) The Neuronal Organization of the Retina. Neuron 76:266–280 Xu J et al (2025) High-order dynamics in an ultra-adaptive neuromorphic vision device. Nat Nanotechnol 20:1419–1430 Tiw PJ et al (2026) An end-to-end memristive hardware system based on single-spike coding for human–machine interfaces. Nat Electron 9:225–236 Liu Z et al (2025) A memristor-based adaptive neuromorphic decoder for brain–computer interfaces. Nat Electron 8:362–372 Lichtsteiner P, Posch C, Delbruck T (2008) A 128×128 120 dB 15 µs latency asynchronous temporal contrast vision sensor. IEEE J Solid-State Circuits 43:566–576 Jang H et al (2022) In-sensor optoelectronic computing using electrostatically doped silicon. Nat Electron 5:519–525 Zhou Y et al A 2T2R1C vision cell with 140 dB dynamic range and event-driven characteristics for in-sensor spiking neural network. In Int Electron Devices Meeting 31–34 (IEEE 2022). Kumar M, Park H, Seo HA (2025) Single-Pixel Event Photoactive Device for Real-Time, In-Sensor Spatiotemporal Optical Information Processing. Adv Mater 37:2406607 Feng S et al (2025) A Charge-Coupled Phototransistor Enabling Synchronous Dynamic and Static Image Detection. Adv Mater 37:2417675 Li K et al (2025) Thin-film event-based vision sensors for enhanced multispectral perception beyond human vision. Infomat 7:e70007 Dang B et al (2026) Spiking neural networks with fatigue spike-timing-dependent plasticity learning using hybrid memristor arrays. Nat Electron 9:213–224 Li X et al (2025) Federated learning using a memristor compute-in-memory chip with in situ physical unclonable function and true random number generator. Nat Electron 8:518–528 Song W et al (2024) Programming memristor arrays with arbitrarily high precision for analog computing. Science 383:903–910 Hahn J et al (2023) Evolution of neuronal cell classes and types in the vertebrate retina. Nature 624:415–424 Dosenbach N, Raichle M, &Gordon E (2025) The brain's action-mode network. Nat Rev Neurosci 26:158–168 Stowell R et al (2019) Noradrenergic signaling in the wakeful state inhibits microglial surveillance and synaptic plasticity in the mouse visual cortex. Nat Neurosci 22:1782–1792 Ravichandran V et al (2026) Memristive cellular neural networks for fast in-pixel computing. Nat Electron 9:189–199 Pannone A et al (2024) Robust chemical analysis with graphene chemosensors and machine learning. Nature 634:572–578 Zheng Y, Ghosh S, Das S (2024) A butterfly-inspired multisensory neuromorphic platform for integration of visual and chemical cues. Adv Mater 36:2307380 Noudoost B, Chang M, Steinmetz N, Moore T (2010) Top-down control of visual attention. Curr Opin Neurobiol 20:183–190 Reimer J et al (2016) Pupil fluctuations track rapid changes in adrenergic and cholinergic activity in cortex. Nat Commun 7:13289 Deng Y et al (2021) MoTe 2 : Semiconductor or Semimetal? ACS Nano 15:12465–12474 Li D et al (2023) Bidirectionally Photoresponsive Optoelectronic Transistors with Dual Photogates for All-Optical-Configured Neuromorphic Vision. Adv Funct Mater 33:2303198 Li D et al (2024) An Active-Matrix Synaptic Phototransistor Array for In-Sensor Spectral Processing. Adv Sci 11 Celik T (2014) Spatial Entropy-Based Global and Local Image Contrast Enhancement. IEEE Trans Image Process 23:5298–5308 Khalid S, Medasani B, Lyons J, Wickramaratne D, Janotti A (2024) The deep-acceptor nature of the chalcogen vacancies in 2D transition-metal dichalcogenides. 2d Mater 11:021001 Cheng Z et al (2024) P/N-type conversion of 2D MoTe 2 controlled by top gate engineering for logic circuits. ACS Appl Mater Interfaces 16(28):36539–36546 McDonough J et al (2008) Thermodynamic, kinetic, and computational study of heavier chalcogen (S, Se, and Te) terminal multiple bonds to molybdenum, carbon, and phosphorus. Inorg Chem 47:2133–2141 Additional Declarations There is NO Competing Interest. Supplementary Files SupplementaryInformationNN.pdf Supplementary Material SupplementaryVideo3.mp4 Supplementary Video 3 SupplementaryVideo2.mp4 Supplementary Video 2 SupplementaryVideo1.mp4 Supplementary Video 1 Cite Share Download PDF Status: Under Review 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-9334649","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":631384492,"identity":"54fba4b3-0cef-463f-93f9-5b8b2e31d3bb","order_by":0,"name":"Bowen Zhu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+klEQVRIiWNgGAWjYDACCQY2MM0PJgsYDCCCxGiRbAASBwxI0WJwgFgt8rN7zB58bLPJM76RfPjzBwM7Y34G5oO3eRjs8nBpYZxzxtxwZltasdmNtDSJAwbJZpINbMnWPAzJxbi0MEvkmEnzbjucuO12jhnQYcw2Bgd4zKR5GA4kNuDQwgbR8j9x8+z8zx8OGNTb2B/g/4ZXCw9Ey4HEDdI5DECHHTYzYOBhw6tFQiKt3HDmv+TEGfefmUmcMThuLHGYzdhyjkEyTi3yM5K3Pfhwxi6xv+fw4w8VFdWG/e3ND2+8qbDDqQULYAYRBsSrHwWjYBSMglGACQDAvlNEyw2zEQAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0001-7534-9723","institution":"Westlake University","correspondingAuthor":true,"prefix":"","firstName":"Bowen","middleName":"","lastName":"Zhu","suffix":""},{"id":631384493,"identity":"1ad25669-1c1d-4ff1-928c-4c436a8a3af3","order_by":1,"name":"Guanlei Zhao","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Guanlei","middleName":"","lastName":"Zhao","suffix":""},{"id":631384494,"identity":"7f9226ee-0119-44e5-a32c-38ce8adbdff4","order_by":2,"name":"Siyu Zhang","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Siyu","middleName":"","lastName":"Zhang","suffix":""},{"id":631384495,"identity":"7529c435-8407-4fcd-9abe-f36209ed0492","order_by":3,"name":"Zhongfang Zhang","email":"","orcid":"https://orcid.org/0009-0005-3493-4949","institution":"University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Zhongfang","middleName":"","lastName":"Zhang","suffix":""},{"id":631384496,"identity":"827e3090-c9d4-406e-85ad-191fc5084d63","order_by":4,"name":"Yingjie Tang","email":"","orcid":"https://orcid.org/0000-0002-6062-8807","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Yingjie","middleName":"","lastName":"Tang","suffix":""},{"id":631384497,"identity":"5c55b881-db1b-4c8a-b628-b3b9eedc6fd7","order_by":5,"name":"Guolei Liu","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Guolei","middleName":"","lastName":"Liu","suffix":""},{"id":631384498,"identity":"3009dfa8-e90c-4e4c-a12c-69a05717b66d","order_by":6,"name":"Fanfan Li","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Fanfan","middleName":"","lastName":"Li","suffix":""},{"id":631384499,"identity":"05ea157f-3fc6-4b5d-ada7-f74955de7dac","order_by":7,"name":"Kaixin Ge","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Kaixin","middleName":"","lastName":"Ge","suffix":""},{"id":631384500,"identity":"94249f8f-134f-4217-a969-b05926a95adf","order_by":8,"name":"Shuhui Ren","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Shuhui","middleName":"","lastName":"Ren","suffix":""},{"id":631384501,"identity":"83f5e11d-c897-48b7-ac61-ae1069559d9e","order_by":9,"name":"Jiayi Mao","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Jiayi","middleName":"","lastName":"Mao","suffix":""},{"id":631384502,"identity":"d89fb837-509d-4c3e-ab07-131c42a4d469","order_by":10,"name":"Weilong Feng","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Weilong","middleName":"","lastName":"Feng","suffix":""},{"id":631384503,"identity":"eb476c54-4be9-4510-b987-8a24a636dad6","order_by":11,"name":"Saisai Wang","email":"","orcid":"","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Saisai","middleName":"","lastName":"Wang","suffix":""},{"id":631384504,"identity":"5ca99897-c8de-4db0-8925-0d4cdde76a3f","order_by":12,"name":"Qi Huang","email":"","orcid":"https://orcid.org/0009-0008-2502-3651","institution":"Westlake University","correspondingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Huang","suffix":""},{"id":631384505,"identity":"a99cf847-a37e-4259-9d6e-1d0f5f5d2a46","order_by":13,"name":"Hong Wang","email":"","orcid":"https://orcid.org/0000-0002-0213-3523","institution":"Xidian University","correspondingAuthor":false,"prefix":"","firstName":"Hong","middleName":"","lastName":"Wang","suffix":""}],"badges":[],"createdAt":"2026-04-06 14:07:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9334649/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9334649/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108157921,"identity":"4a2385df-0285-48d7-a678-46ca6893a98c","added_by":"auto","created_at":"2026-04-30 03:09:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1440234,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThreshold-tunable event-driven vision sensor for high-order perceptual processing. a\u003c/strong\u003e, Schematic illustration of hierarchical adaptation in the biological visual system. \u003cstrong\u003eb\u003c/strong\u003e, Device architecture and working mechanism of the MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction phototransistor. Electron-beam-induced defect states in MoTe\u003csub\u003e2\u003c/sub\u003e act as hole trapping centers and integrate it with an In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e channel to realize a heterojunction transistor with gate-controlled threshold-tunable threshold characteristics. \u003cstrong\u003ec\u003c/strong\u003e, Equivalent circuit model of the device. \u003cstrong\u003ed\u003c/strong\u003e, Gate-controlled threshold modulation of the device.\u003cstrong\u003e e\u003c/strong\u003e, Representative event responses under different threshold settings, illustrating the transition between noise suppression and detail enhancement regimes through threshold modulation.\u003cstrong\u003e f\u003c/strong\u003e, Relationship between ETS and event entropy: event entropy increases with decreasing ETS, as a lower threshold facilitates event-spiking generation and leads to a more distributed event representation. \u003cstrong\u003eg\u003c/strong\u003e, Hardware array ETS processing for complex driving scenarios: increasing the threshold during rainy conditions and decreasing it during foggy conditions. \u003cstrong\u003eh\u003c/strong\u003e, Event-entropy-guided regulation of event imaging. In rainy conditions, increased threshold suppresses noise events and reduces entropy, while in foggy conditions, decreased threshold enhances weak signals and edge features, improving overall perception efficiency.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/01c68b8b18d8e0ac0a658edd.png"},{"id":108157913,"identity":"e947ed62-3a0f-4a3d-bbe6-fd956a625ace","added_by":"auto","created_at":"2026-04-30 03:09:22","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":952535,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eCharacterization of electron-beam-induced defects in MoTe\u003c/strong\u003e\u003csub\u003e\u003cstrong\u003e2\u003c/strong\u003e\u003c/sub\u003e\u003cstrong\u003e.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e, TEM image of the MoTe\u003csub\u003e2 \u003c/sub\u003efilm before EBE, showing the well-defined layered structure. \u003cstrong\u003eb\u003c/strong\u003e, TEM image after EBE, revealing pronounced lattice misalignment and interlayer distortion due to defect formation. \u003cstrong\u003ec\u003c/strong\u003e, XPS analysis showing the Te:Mo atomic ratio before and after EBE. The decrease in Te content after EBE confirms the formation of Te vacancies, which act as deep electronic trap states. Transient absorption spectra at 703 nm of the samples: \u003cstrong\u003ed\u003c/strong\u003e, Before EBE and \u003cstrong\u003ee\u003c/strong\u003e, After EBE. There is a significantly stronger and longer-lived bleaching signal after EBE. \u003cstrong\u003ef\u003c/strong\u003e, Bleaching signal plot before and after EBE at 703 nm, showing changes in the transient absorption dynamics.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/322a3ff7ca0dbb022bcc3a92.png"},{"id":108157955,"identity":"db97151c-92e4-456b-9d08-6f3f7b1d78e8","added_by":"auto","created_at":"2026-04-30 03:09:35","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":721419,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEvent-driven sensing mechanism and gate voltage-modulated threshold tuning. a\u003c/strong\u003e, Illustration of the event-generation process in the MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction phototransistor. The figure shows the photogenerated carrier dynamics including (I) electron-hole pair generation, (II) trapping, and (III) recombination, contributing to transient event spikes. \u003cstrong\u003eb\u003c/strong\u003e, The physical image of the MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistor. \u003cstrong\u003ec\u003c/strong\u003e, Photocurrent response to step changes.\u003cstrong\u003e d\u003c/strong\u003e, Response of the device under various illumination levels. The figure shows pronounced transient responses as a function of light intensity, with pulse amplitudes increasing with light intensity. \u003cstrong\u003ee\u003c/strong\u003e,\u003cstrong\u003e \u003c/strong\u003eDETR exceeds 60 dB, corresponding to illumination variations from 10 μW/cm\u003csup\u003e2\u003c/sup\u003e to 11.7 mW/cm\u003csup\u003e2\u003c/sup\u003e. \u003cstrong\u003ef\u003c/strong\u003e, Output characteristics curve of the MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistor at different gate voltages, showing that gate-modulated behavior arises from the tunable electric field of the heterojunction. \u003cstrong\u003eg\u003c/strong\u003e, Response of event pulses under varying \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e settings. The tunable response intensity allows for a functional switch between signal enhancement and noise suppression modes.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/65d79266d73f2c6ebf6f73d6.png"},{"id":108157915,"identity":"06ecbf39-eab2-43b6-a017-c8c71df703c6","added_by":"auto","created_at":"2026-04-30 03:09:23","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":873772,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHardware implementation and system-level validation of the TEVS array.\u003c/strong\u003e \u003cstrong\u003ea,\u003c/strong\u003e Schematic of the 6×6 MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e TEVS array integrated with a custom-designed PCB platform for real-time acquisition and processing of event data. \u003cstrong\u003eb\u003c/strong\u003e, Diagram of the system-level signal flow from the sensor array to host computer imaging. \u003cstrong\u003ec\u003c/strong\u003e, Representative signals recorded through the acquisition chain, illustrating the process of photocurrent conversion, amplification, threshold detection and pulse shaping. \u003cstrong\u003ed\u003c/strong\u003e, Single-pixel event imaging experiments demonstrate the motion of a car model captured by a single phototransistor. \u003cstrong\u003ee\u003c/strong\u003e, Event imaging results of the car model. The system generates images from 3 different positions, with event signatures encoded solely through the temporal sequence of ON and OFF events.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/5edc160d284785ccf23a08e2.png"},{"id":108157894,"identity":"693c39b1-4263-48b5-96b5-ab8d3358c539","added_by":"auto","created_at":"2026-04-30 03:09:20","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":2144480,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eIn-sensor event stream array imaging and effect processing via threshold modulation.\u003c/strong\u003e \u003cstrong\u003ea\u003c/strong\u003e, Experimental setup for in-sensor event stream processing using a 6×6 MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e TEVS array. A 100 Hz laser illuminates the transistor array through a spatial mask with various shapes, simulating environmental conditions such as rain and fog. The configurations of the mask patterns and the array image are shown in the figure. \u003cstrong\u003eb\u003c/strong\u003e, Event-spiking generation response in a rain-mimicking scenario. By adjusting to a higher threshold, noise events caused by raindrops are suppressed, thereby improving the imaging performance. \u003cstrong\u003ec\u003c/strong\u003e, Fog simulation showing the effect of threshold modulation on object representation, enhancing sensitivity to weak intensity variations. \u003cstrong\u003ed\u003c/strong\u003e, \u003cstrong\u003ee\u003c/strong\u003e, For event stream simulation on a 1280×720 pixel array, the impact of different thresholds on imaging performance and event entropy under rain and fog conditions was compared, demonstrating the capability of threshold modulation to improve perception clarity in complex environments with high-density pixels.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/35c1df7383b62e7277bff3b1.png"},{"id":108158196,"identity":"f2671006-1ffb-44a7-ace8-e1880aafb18a","added_by":"auto","created_at":"2026-04-30 03:10:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":7553531,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/b7299203-28fa-497d-8abb-401f6b04c444.pdf"},{"id":108158146,"identity":"99af7de2-c898-4f93-8fee-8aca3392031b","added_by":"auto","created_at":"2026-04-30 03:10:08","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":3162038,"visible":true,"origin":"","legend":"Supplementary Material","description":"","filename":"SupplementaryInformationNN.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/208684de39dc19a57d8c495f.pdf"},{"id":108157929,"identity":"8e2177b1-077c-44a2-8976-2eb847dfc258","added_by":"auto","created_at":"2026-04-30 03:09:24","extension":"mp4","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":19876369,"visible":true,"origin":"","legend":"Supplementary Video 3","description":"","filename":"SupplementaryVideo3.mp4","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/5545e4e3b3ae11aac2dfa674.mp4"},{"id":108157920,"identity":"0b645581-b701-40fc-bbc1-f5e4641b8311","added_by":"auto","created_at":"2026-04-30 03:09:23","extension":"mp4","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":22203212,"visible":true,"origin":"","legend":"Supplementary Video 2","description":"","filename":"SupplementaryVideo2.mp4","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/207b89684b543a1f16ae1591.mp4"},{"id":108157918,"identity":"2c307ccf-23dd-4ae7-9caa-60055c3098ec","added_by":"auto","created_at":"2026-04-30 03:09:23","extension":"mp4","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":24961215,"visible":true,"origin":"","legend":"Supplementary Video 1","description":"","filename":"SupplementaryVideo1.mp4","url":"https://assets-eu.researchsquare.com/files/rs-9334649/v1/35c98fcf6c583a907061a903.mp4"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Threshold-tunable event-driven vision sensor for adaptive visual processing","fulltext":[{"header":"Introduction","content":"\u003cp\u003eMachine vision\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 underpins the perceptual and decision-making capabilities of intelligent systems\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, requiring vision sensors that offer high temporal precision, low latency, and strong adaptability. Frame-based cameras\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, which operate on fixed-exposure and global-sampling principles, struggle with static data redundancy and computational inefficiency. In contrast, neuromorphic event-driven vision sensors\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e, inspired by the spatiotemporal processing of the human retina\u003csup\u003e\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, shift the paradigm towards dynamic, information-centric perception and a cornerstone of neuromorphic computing systems\u003csup\u003e\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. These sensors operate asynchronously, recording only local intensity changes as sparse spike signals. This approach minimizes data redundancy and markedly enhances the processing of temporal visual information. Conventional circuit-based dynamic vision sensors (DVS) provide significant advantages, including compatibility with complementary metal-oxide-semiconductor (CMOS) technology, microsecond-level temporal resolution, and strong environmental adaptability benefited from the tunable threshold of the comparator circuit\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. However, the asynchronous circuits, composed of photoreceptors, amplifiers and comparators, require each pixel to function as an independent sensing and processing unit. This complexity limits pixel density and fill factor, while introducing thermal noise and power overhead, especially in static scenes.\u003c/p\u003e \u003cp\u003eRecent advances in event-driven vision sensors have focused on simplifying circuit architecture within the pixel. Examples include the two-transistor-two-resistor-one-capacitor (2T2R1C) and band-engineered semiconductor devices that enable one-photosensor-one-capacitor (1P1C) or carrier-blocking event-generation mechanisms\u003csup\u003e\u003cspan additionalcitationids=\"CR21 CR22 CR23\" citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. However, the fixed threshold eliminates the ability to adjust sensitivity dynamically according to environmental conditions, which creates a practical dilemma: under noisy or low-light conditions, a fixed low threshold triggers excessive spurious events, whereas in high-illumination or low-contrast scenes, a fixed high threshold suppresses meaningful signals. While achieving structural simplicity, these designs sacrifice perceptual adaptability, undermining the robustness of neuromorphic perception and encoding complex real-world environments\u003csup\u003e\u003cspan additionalcitationids=\"CR26\" citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Consequently, a longstanding challenge is the inability to integrate adaptive perception into a structurally simple device.\u003c/p\u003e \u003cp\u003eIn this work, we demonstrate an in-sensor event-driven vision system with intrinsically tunable event thresholds enabled by a defect-engineered capacitive mechanism embedded in a single MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction phototransistor. The device exhibits microsecond-scale response times (1.2/1.0 \u0026micro;s) and a high dynamic range exceeding 82 dB. It further enables continuous and wide-range event-threshold modulation, with a dynamic event threshold range (DETR) exceeding 60 dB (10 \u0026micro;W cm\u003csup\u003e-\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e to 11.7 mW cm\u003csup\u003e-\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e). A system-level validation was achieved using a 6\u0026times;6 dynamic sensor array, showing robust denoising and feature enhancement under adverse conditions such as rain and fog. Furthermore, we introduce an event entropy metric that quantitatively links threshold modulation to information efficiency, providing a unified framework for evaluating and optimizing the information density of asynchronous event streams. This work provides an effective strategy for developing bioplausible event-based visual sensors.\u003c/p\u003e\n\u003ch3\u003eBio-inspired adaptive event-driven vision sensor with in-sensor threshold modulation\u003c/h3\u003e\n\u003cp\u003eThe human visual system achieves robust perception in dynamic environments through hierarchical adaptation spanning the retina\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e and cortex\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e, a principle that neuromorphic hardware seeks to emulate\u003csup\u003e\u003cspan additionalcitationids=\"CR32\" citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. As illustrated in Fig.\u0026nbsp;1a, at the retinal level, the retina spiking neuron (RSN) adjust their response gain to maintain a wide dynamic range and stable encoding under varying luminance. At higher stages, visual cortex (VC) feedback further modulates neuronal excitability and perceptual thresholds in a task-dependent manner, enabling adaptive visual processing\u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. Inspired by the biological visual system, DVS captures temporal dynamics through sophisticated circuit architecture. Threshold-control units modulate detection sensitivity across varying illumination conditions, thereby regulating event responses. However, this circuit-centric methodology increases system complexity, resulting in additional power consumption and time delay.\u003c/p\u003e \u003cp\u003eIntegrating event-spiking generation and threshold tunability within a single device addresses the architectural complexity and latency of conventional event-driven sensors (Fig.\u0026nbsp;1b). This top-down neural visual pathway can be physically realized within a single heterojunction transistor, where the heterojunction encodes visual stimuli into photo-induced spike signals, and the gate serves as a control terminal that modulates carrier density and band bending to regulate spike amplitude and thus the event threshold (ETS). Capacitive behavior is essential for spike generation; however, recent approaches\u003csup\u003e\u003cspan additionalcitationids=\"CR23\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e rely on an additional insulating layer that introduce interfacial barriers and parasitic capacitance, inevitably suppressing carrier coupling and attenuating spike amplitude. In contrast, we harness defect-state capacitance intrinsically formed within MoTe\u003csub\u003e2\u003c/sub\u003e\u003csup\u003e36\u003c/sup\u003e, which directly couples with photogenerated carriers, enabling spike generation while preserving high photoresponse in a structurally simplified architecture. The In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e\u003csup\u003e37, 38\u003c/sup\u003e channel provides electrostatic control over carrier density and interfacial band alignment. Gate modulation alters the occupation of defect states and the associated trap-detrap dynamics, thereby tuning the effective carrier capture process. Therefore, MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction phototransistor intrinsically integrates event-spiking generation and ETS modulation within a single device.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eFigure\u0026nbsp;1 | Threshold-tunable event-driven vision sensor for high-order perceptual processing. a\u003c/b\u003e, Schematic illustration of hierarchical adaptation in the biological visual system. \u003cb\u003eb\u003c/b\u003e, Device architecture and working mechanism of the MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction phototransistor. Electron-beam-induced defect states in MoTe\u003csub\u003e2\u003c/sub\u003e act as hole trapping centers and integrate it with an In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e channel to realize a heterojunction transistor with gate-controlled threshold-tunable threshold characteristics. \u003cb\u003ec\u003c/b\u003e, Equivalent circuit model of the device. \u003cb\u003ed\u003c/b\u003e, Gate-controlled threshold modulation of the device. \u003cb\u003ee\u003c/b\u003e, Representative event responses under different threshold settings, illustrating the transition between noise suppression and detail enhancement regimes through threshold modulation. \u003cb\u003ef\u003c/b\u003e, Relationship between ETS and event entropy: event entropy increases with decreasing ETS, as a lower threshold facilitates event-spiking generation and leads to a more distributed event representation. \u003cb\u003eg\u003c/b\u003e, Hardware array ETS processing for complex driving scenarios: increasing the threshold during rainy conditions and decreasing it during foggy conditions. \u003cb\u003eh\u003c/b\u003e, Event-entropy-guided regulation of event imaging. In rainy conditions, increased threshold suppresses noise events and reduces entropy, while in foggy conditions, decreased threshold enhances weak signals and edge features, improving overall perception efficiency.\u003c/p\u003e \u003cp\u003eFigure 1c illustrates the equivalent circuit model of the device. The MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction is represented by a trap-mediated charge storage element that behaves analogously to a capacitance, arising from the dynamic trapping and release of carriers at defect states. This effective capacitance is modulated by the gate voltage (\u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e), which tunes the occupation of defect states and the associated trapping dynamics. The device operates under zero drain bias (\u003cem\u003eV\u003c/em\u003e\u003csub\u003eDS\u003c/sub\u003e = 0), with event sensing governed by the intrinsic trap-mediated charge storage behavior. Threshold modulation is achieved via \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e, which modulates the carrier density and defect-state occupation, thereby altering the effective carrier trapping dynamics. As a result, the ETS can be continuously adjusted: increasing \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e suppresses carrier trapping and raises the threshold, while decreasing \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e enhances trapping and lowers the threshold (Fig.\u0026nbsp;1d). This dynamically tunable threshold enables adaptive response to varying illumination conditions. The device enables flexible event encoding through ETS modulation, where the optimal threshold can be adaptively tuned to enable noise suppression and weak-signal enhancement for different application scenarios (Fig.\u0026nbsp;1e).\u003c/p\u003e \u003cp\u003eTo quantify the effect of threshold modulation, we introduce event entropy as a metric, drawing inspiration from the concept of image entropy\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Event entropy evaluates the information content of an event stream based on the joint probability distribution of events across spatial blocks and polarity. By partitioning the sensor array into spatial blocks and computing the probabilities of positive and negative events within each block, we obtain the expression for event entropy:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:H(X,S)=-\\sum\\:_{x=1}^{N}\\sum\\:_{s\\in\\:\\left\\{+1,\\:\\:-1\\right\\}}p(x,s){\\text{log}}_{2}p\\left(x,s\\right)\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(1\\right)$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:x\\:\\)\u003c/span\u003e\u003c/span\u003eindexes the spatial blocks, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:s\\:\\)\u003c/span\u003e\u003c/span\u003edenotes event polarity, and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:p\\left(x,s\\right)\\:\\)\u003c/span\u003e\u003c/span\u003eis the probability of observing an event of polarity \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:s\\:\\)\u003c/span\u003e\u003c/span\u003ein block \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:x\\)\u003c/span\u003e\u003c/span\u003e. Event entropy quantifies the complexity and richness of the event distribution. The detailed definition and calculation of event entropy are provided in Supplementary Note Ⅰ. A higher event entropy corresponds to a more distributed event pattern with richer information content, whereas a lower value reflects a more concentrated distribution associated with noise suppression (Fig.\u0026nbsp;1f). This metric enables quantitative evaluation of the event stream under different threshold settings and establishes a direct link between threshold modulation and the dynamic information density of the event stream. Consequently, threshold modulation enables continuous control over event-stream encoding, transitioning between noise suppression and detail enhancement regimes, thereby improving robustness under dynamic visual conditions.\u003c/p\u003e \u003cp\u003eLeveraging the environmental adaptability, we develop a vision sensing hardware system based on a MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e device array, enabling real-time dynamic threshold configuration for processing complex visual scenes (Fig.\u0026nbsp;1g). For two contrasting driving scenarios, namely rainy and foggy conditions, the system adaptively increases or decreases the threshold to regulate the event response. This enables effective suppression of rain-induced noise and enhancement of blurred vehicle features, thereby improving visual perception efficiency through coordinated sensing and processing. Building on the gate-modulated threshold engineering described above, we next examine the impact of adaptive threshold control on event-based imaging at the event entropy scale (Fig.\u0026nbsp;1h). In rainy conditions, excessive rain-induced events introduce redundant noise, which can be suppressed by increasing the threshold, resulting in reduced event entropy and improved information efficiency. In contrast, under foggy conditions, lowering the threshold enhances sensitivity to weak signals, strengthening edge information and recovering fine details. Event entropy quantitatively characterizes the distribution of the event stream and serves as an effective metric for evaluating imaging performance under different threshold settings.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eDefect engineering in MoTe\u003csub\u003e2\u003c/sub\u003e\u003c/h2\u003e \u003cp\u003eMoTe\u003csub\u003e2\u003c/sub\u003e has attracted considerable attention due to its broadband spectral response and strong light absorption. Previous studies have shown that electron-beam processes can effectively modulate the properties of MoTe\u003csub\u003e2\u003c/sub\u003e thin films\u003csup\u003e\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. In this work, deep-level defect states are introduced into the MoTe\u003csub\u003e2\u003c/sub\u003e film via electron-beam treatment, forming hole trapping centers that enable intrinsic charge storage analogous to a capacitance, without requiring additional circuit components. The transient response arises from dynamic trapping and release of photogenerated carriers. Structural defects are intrinsically introduced during the electron-beam evaporation of the top electrode. The relatively weak Mo-Te bonding facilitates the formation of Te vacancies and associated deep-level trap states under electron-beam exposure\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. These defect states dominate carrier relaxation and enable pronounced transient photoresponses under time-varying illumination.\u003c/p\u003e \u003cp\u003eThis defect formation is directly evidenced by both structural and chemical characterizations. Transmission electron microscopy (TEM) reveals that the pristine MoTe\u003csub\u003e2\u003c/sub\u003e film, which initially exhibits a well-defined layered structure, undergoes pronounced lattice misalignment and interlayer distortion after electron-beam evaporation (EBE), indicating substantial lattice disruption and disorder (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ea, b). The relatively weak Mo-Te bonding facilitates the formation of Te vacancies and associated deep-level trap states\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e. Consistent with this, X-ray photoelectron spectroscopy (XPS) analysis reveals a decrease in the Te:Mo atomic ratio from 1.94:1 to 1.48:1, confirming the generation of Te vacancies and corresponding deep trap states (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ec). A comparison of the elemental valence states and compositions before and after deposition is provided in Supplementary Fig.\u0026nbsp;1.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThese structural and compositional changes translate into markedly altered excitonic and carrier dynamics. Transient absorption measurements reveal that evaporated MoTe\u003csub\u003e2\u003c/sub\u003e exhibits a significantly stronger and longer-lived bleaching signal at 703 nm compared with its pristine counterpart (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ed, e). The stronger and longer-lived bleaching signal indicates the presence of deep trap states, which significantly alter the excitonic dynamics by trapping carriers more effectively, leading to prolonged signal decay times (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ef). The prolonged decay reflects suppressed carrier recombination due to deep trapping, which inhibits rapid relaxation back to the ground state. As a result, transient perturbations in illumination are amplified, whereas static or slowly varying backgrounds are intrinsically suppressed.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eEvent-driven sensing mechanism and gate voltage-modulated threshold tunability\u003c/h3\u003e\n\u003cp\u003eBenefiting from this defect-engineering strategy, we develop a MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction phototransistor. In this lateral architecture, an In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e channel bridges the source and drain, while a MoTe\u003csub\u003e2\u003c/sub\u003e layer partially overlaps the channel near the source to form a heterojunction. Owing to the low Mo-Te bond energy, EBE introduces a high density of defect states into the MoTe\u003csub\u003e2\u003c/sub\u003e layer, forming an intrinsic charge-trapping centers. Photogenerated carriers are transiently captured and released in response to illumination changes, giving rise to pronounced bipolar photocurrent spikes, while relaxing to a near-zero steady-state baseline under constant illumination. This material-level charge accumulation and release mechanism enables direct event-spiking generation within a single transistor, without relying on external capacitors or peripheral signal-conditioning circuits. The heterostructure combines narrow-bandgap MoTe\u003csub\u003e2\u003c/sub\u003e, which provides broadband optical absorption, with wide-bandgap In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e, which offers high carrier mobility and strong gate electrostatic control. The resulting band alignment forms an interfacial barrier whose height can be efficiently modulated by the gate through carrier density tuning in In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e, enabling dynamic regulation of the event-generation threshold. In this way, defect-mediated trapping dynamics and gate-controlled transport are synergistically integrated, allowing circuit-level functionalities to be intrinsically realized within a single device.\u003c/p\u003e \u003cp\u003eThe light-induced event-driven sensing process proceeds through three distinct stages (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003ea). Upon illumination onset (stage I), incident 520 nm light is predominantly absorbed by MoTe\u003csub\u003e2\u003c/sub\u003e, generating electron-hole pairs. Operating at zero drain-source bias (\u003cem\u003eV\u003c/em\u003e\u003csub\u003eDS\u003c/sub\u003e=0 V), the photoresponse arises solely from carrier separation by the built-in electric field at the MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterointerface: photogenerated electrons are rapidly injected into the In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e channel, while holes drift toward the electrodes or are trapped in defect states within MoTe\u003csub\u003e2\u003c/sub\u003e. This abrupt interfacial charge separation produces a positive transient photocurrent pulse, corresponding to an ON event. During continuous illumination (stage II), photocarrier generation balances defect-state trapping and detrapping. This results in a relaxation of the photocurrent towards a near-baseline level, maintaining selective sensitivity to temporal variations while rejecting static background signals. Upon illumination removal (stage III), electrons accumulated in the In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e channel recombine with holes previously trapped in MoTe\u003csub\u003e2\u003c/sub\u003e, generating a negative transient pulse that marks an OFF event. Collectively, these observations indicate that the transient photocurrent spiking is primarily governed by the capacitive trapping and release of carriers at the defect states in MoTe\u003csub\u003e2\u003c/sub\u003e, a process reinforced by the heterojunction band alignment that directs charge transfer across the interface. This mechanism fundamentally differs from capacitor-dependent designs and enables event spiking within a heterojunction.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe physical image of the device is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003eb. The device exhibits pronounced transient responses over a wide range of illumination levels. Stepwise changes in light intensity generate distinct current pulses that reliably return to baseline once illumination stabilizes (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003ec). Pulse amplitude increases monotonically with illumination contrast, indicating a robust correspondence between stimulus variation and event strength (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003ed). As \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e= -10 V and under 800 nm illumination, the device exhibits a dynamic range of over 82 dB (Supplementary Fig.\u0026nbsp;2). It exceeds the range of the human retina, indicating the superior ability to detect both weak and strong stimuli without saturation. The device exhibits a rapid photoelectric pulse response, with a response time of 1.2 \u0026micro;s for positive pulses and 1.0 \u0026micro;s for negative pulses (Supplementary Fig.\u0026nbsp;3). In addition, owing to its narrow bandgap, the wide spectral response of MoTe\u003csub\u003e2\u003c/sub\u003e, spanning from 360 to 1064 nm, is particularly advantageous for event-driven vision sensors as it allows for reliable operation under diverse lighting conditions, from visible light to infrared, thereby enhancing the sensor\u0026rsquo;s versatility and robustness in real-world environments. (Supplementary Fig.\u0026nbsp;4).\u003c/p\u003e \u003cp\u003eBeyond event-spiking generation, the lateral heterojunction architecture enables active electrical modulation of the ETS via \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e, a capability that is fundamentally absent in conventional vertical designs. By extracting the minimum detectable illumination change as a function of the \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e from \u0026minus;\u0026thinsp;10 to 0 V, the device achieves a DETR exceeding 60 dB, corresponding to illumination variations from 10 \u0026micro;W/cm\u003csup\u003e2\u003c/sup\u003e to 11.7 mW/cm\u003csup\u003e2\u003c/sup\u003e. This range covers the vast majority of lighting conditions encountered in natural environments (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ee). This enhanced dynamic range is crucial for applications in real-world environments where illumination varies drastically. This gate-modulated behavior arises from the tunable electric field of the heterojunction. At negative gate biases (\u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e=-15 V), reduced carrier density in In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e modifies the interfacial band alignment while leaving the MoTe\u003csub\u003e2\u003c/sub\u003e electrostatic potential largely unchanged. This enhances the effective junction field, promoting more efficient charge separation. As \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e approaches 0 V, the junction field strength diminishes (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003ef). The mechanism analysis of gate-modulated heterojunction electric field strength is shown in Supplementary Fig.\u0026nbsp;6. By modulating the strength of the heterojunction electric field, we achieve in-sensor event information processing (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e2\u003c/span\u003eg). The tunable response intensity allows for a functional switch between signal enhancement and noise suppression. From an event-stream perspective, threshold modulation regulates event density and temporal sparsity. Lower thresholds enhance sensitivity to weak features, making it ideal for low-contrast scenes, while higher thresholds suppress spurious events caused by noise or rapid background fluctuations. Electrically controlled modulation of the heterojunction field enables flexible regulation of event-spiking generation within a single transistor. By overcoming the fixed-threshold limitations of conventional designs, this approach provides a robust physical foundation for environmentally adaptive, threshold-tunable event-driven vision sensors.\u003c/p\u003e\n\u003ch3\u003eHardware implementation and system-level validation of threshold-tunable event-driven vision sensor array\u003c/h3\u003e\n\u003cp\u003eTo validate the practical deployability of the gate-tunable event mechanism at the system level, we integrated a 6\u0026times;6 MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e threshold-tunable event-driven vision sensor (TEVS) array with a designed printed circuit board (PCB) platform for real-time acquisition and processing (Fig.\u0026nbsp;4a). The system comprises a photosensor array, a dedicated gate-bias PCB for global threshold control, and an acquisition PCB for signal conditioning and digitization.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eFigure\u0026nbsp;4. | Hardware implementation and system-level validation of the TEVS array. a\u003c/b\u003e, Schematic of the 6\u0026times;6 MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e TEVS array integrated with a custom-designed PCB platform for real-time acquisition and processing of event data. \u003cb\u003eb\u003c/b\u003e, Diagram of the system-level signal flow from the sensor array to host computer imaging. \u003cb\u003ec\u003c/b\u003e, Representative signals recorded through the acquisition chain, illustrating the process of photocurrent conversion, amplification, threshold detection and pulse shaping. \u003cb\u003ed\u003c/b\u003e, Single-pixel event imaging experiments demonstrate the motion of a car model captured by a single phototransistor. \u003cb\u003ee\u003c/b\u003e, Event imaging results of the car model. The system generates images from 3 different positions, with event signatures encoded solely through the temporal sequence of ON and OFF events.\u003c/p\u003e \u003cp\u003eFigure 4b illustrates the system-level signal flow from on-board PCB sampling to host-computer imaging. The photocurrent signals are first converted to voltage and amplified using trans-impedance amplifiers (TIAs), followed by polarity-selective comparators for threshold detection and pulse shaping, after which event-based images are displayed on the host computer. The layout and detailed circuit design of the PCB platform are provided in Supplementary Figs.\u0026nbsp;12 and 13. This TEVS-PCB design thereby provides a compact and scalable system-level solution, eliminating the need for per-pixel differential amplifiers and comparators. And it partially addresses the long-standing challenge of requiring independent circuit units per pixel in conventional event cameras.\u003c/p\u003e \u003cp\u003eRepresentative signals recorded along the acquisition chain are presented in Fig.\u0026nbsp;4c. The MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistor generates bipolar, spike current transients in response to illumination changes. The transient signals are captured directly using an oscilloscope. The original signals, being relatively small in amplitude, are susceptible to interference such as lead-injected noise, which could distort the waveform. To preserve the intrinsic event dynamics, the current pulses are processed with an unfiltered transimpedance amplifier that provides gain without temporal smoothing. The resulting voltage pulses exhibit a high intrinsic signal-to-noise ratio for both ON and OFF events. This indicates that the observed noise predominantly originates from the measurement setup (e.g., the noise of the oscilloscope leads), whereas the MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistor array itself generates events with high fidelity and low intrinsic noise. A subsequent comparator stage converts these supra-threshold events into clean digital pulses, providing reliable polarity discrimination.\u003c/p\u003e \u003cp\u003eSingle-pixel event imaging of a moving object provided a stringent system-level validation of our device-circuit design (Fig.\u0026nbsp;4d, e). In the measurement, a moving car model traverses the active area of a single phototransistor, and its motion is encoded exclusively through the temporal sequence of ON and OFF events generated by illumination changes. Despite the absence of spatial multiplexing, distinct object geometries and trajectories produce reproducible and characteristic temporal event signatures. Variations in object length, shape, and traversal speed are reflected in the event density, polarity sequence, and inter-event timing, enabling discrimination based purely on temporal information. Real-time imaging of the vehicle model is presented in Supplementary Video 1. These results demonstrate that meaningful visual information can be extracted without frame reconstruction or multi-pixel correlation. Instead, perception emerges from the precise timing of sparse events at a single sensing node, directly reflecting the core operating principle of event-driven vision.\u003c/p\u003e\n\u003ch3\u003eIn-sensor event-stream processing via threshold modulation\u003c/h3\u003e\n\u003cp\u003eWe next demonstrate array-level in-sensor event-stream processing enabled by gate-controlled threshold modulation using a 6\u0026times;6 MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e TEVS array integrated with a custom PCB platform. The experimental configuration and representative real-time outputs are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003ea. A 100 Hz modulated laser illuminates the array through a spatial mask made of optical films with different transmittances. This setup simultaneously emulates both the spatially varying attenuation (e.g., fog) and the stochastic noise (e.g., rain) found in challenging environmental conditions. The gate voltage \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e is adjusted in real time to globally tune the ETS across the array, without modifying the readout circuitry.\u003c/p\u003e \u003cp\u003eEvent entropy was computed by defining a block of pixels (2\u0026times;2 for the 6\u0026times;6 array) and calculating the probability of event occurrence within each block over time. This block size balances assessment accuracy with computational load. In the rain-mimicking scenario (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003eb), the mask generates abundant noise-induced events when the threshold is set low (\u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e = -12 V). Under this condition, both noise and object-related signals are indiscriminately captured, resulting in a high event entropy of 2.5 bits and a reduced imaging accuracy of 73.9% for the car model. Raising the threshold by adjusting \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e to -6 V effectively suppresses rain-induced noise, reducing event entropy to 2.0 bits and restoring imaging accuracy to 100%. This result illustrates how threshold elevation selectively filters noise while preserving salient object features. Conversely, in the fog-mimicking scenario (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003ec), a high threshold (\u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e = -5 V) renders many pixels unresponsive to weak intensity variations, leading to incomplete object representation, a low event entropy of 1.3 bits, and an imaging accuracy of 76.5%. Lowering the threshold to \u003cem\u003eV\u003c/em\u003e\u003csub\u003eGS\u003c/sub\u003e = -12 V increases sensitivity to these subtle variations, which raises the event entropy to 2.0 bits and recovers the full structure of the car model with 100% accuracy. This illustrates how threshold reduction enhances weak feature detection. The dynamic threshold modulation process is demonstrated in the Supplementary Video 2 and 3, showing the dynamic adjustments in real-time. Together, these results demonstrate that gate-controlled threshold modulation can dynamically balance noise suppression and feature enhancement based on environmental conditions, offering task-oriented event-stream processing.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eTo further assess the scalability and generality of threshold modulation and event entropy, we performed event-stream simulations on a 1280\u0026times;720 pixel array derived from real-world driving scenes under rain and fog, respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e5\u003c/span\u003ed, e). Given the high pixel density, a 16\u0026times;16 pixel region is adopted as the block size to ensure computational efficiency. The ETS is defined as the ratio of the variation in grayscale intensity to the total grayscale intensity. In the rain scenario, a low ETS (3%) produces dense noise-dominated event streams, whereas increasing the threshold to 5% suppresses rain-induced interference and reduces event entropy from 10.8 to 10.0 bits, substantially improving perceptual clarity. In contrast, under foggy conditions, an overly high threshold (12%) leads to severe loss of structural information, while lowering the threshold to 4% recovers fine vehicle contours and enhances scene interpretability, with the event entropy increasing from 6.7 to 7.6 bits. Although obtained from large-scale simulations, the threshold values are directly mapped from experimentally validated gate modulation, highlighting the scalability of the proposed physical mechanism. Collectively, these results establish threshold modulation as a powerful in-sensor mechanism for event-stream processing. By dynamically regulating event density and information content in response to environmental conditions, the proposed sensor enables robust perception without relying on external computation or complex circuitry. Event entropy serves as a quantitative bridge linking physical threshold control to perceptual performance, providing a valuable metric and tool for evaluating and optimizing event-based vision in complex real-world environments.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eIn conclusion, we demonstrate a tunable-threshold event-driven vision sensor based on a single MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction phototransistor. The device achieves microsecond-scale response (1.2/1.0 \u0026micro;s) and a DETR exceeding 60 dB, which enables in-sensor event processing for noise suppression and feature enhancement without external capacitors or threshold-control circuits. System-level validation using a 6\u0026times;6 sensor array confirms real-time imaging performance in complex dynamic environments. This work establishes single-transistor tunable-threshold photodetection as a minimal functional building block that collapses circuit complexity into device-level functionality, while introducing event entropy as a quantitative framework linking threshold modulation to information efficiency, forming a basis for the design of event-driven vision architectures and the characterization of information density in asynchronous event streams.\u003c/p\u003e \u003c/div\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003eMaterials\u003c/h2\u003e \u003cp\u003eIndium nitrate hydrate (In(NO\u003csub\u003e3\u003c/sub\u003e)\u003csub\u003e3\u003c/sub\u003e\u0026middot;xH\u003csub\u003e2\u003c/sub\u003eO, 99.999%) powder was purchased from Sigma-Aldrich. 2-methoxyethanol (2-ME, 99.3%), acetylacetone (AcAc, 99%) and ammonium hydroxide (NH\u003csub\u003e3\u003c/sub\u003e\u0026middot;H\u003csub\u003e2\u003c/sub\u003eO, 28%) were purchased from Alfa Aesar. 2H-MoTe\u003csub\u003e2\u003c/sub\u003e, grown via chemical vapor deposition (CVD) was obtained from SixCarbon Technology Shenzhen.\u003c/p\u003e \u003c/div\u003e \n\u003ch3\u003ePreparation of InO precursor solution\u003c/h3\u003e\n\u003cp\u003e0.3 g In(NO\u003csub\u003e3\u003c/sub\u003e)\u003csub\u003e3\u003c/sub\u003e\u0026middot;xH\u003csub\u003e2\u003c/sub\u003eO was dissolved in 10 mL 2-ME. Then, add 100 \u0026micro;L AcAc and 35 \u0026micro;L NH\u003csub\u003e3\u003c/sub\u003e\u0026middot;H\u003csub\u003e2\u003c/sub\u003eO in the solution as additives. After that, stir the In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e precursor solution at room temperature for 6 hours.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eMoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistor fabrication\u003c/h2\u003e \u003cp\u003eThe In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e precursor solution was spin coated at 3000 rpm for 40 s on SiO\u003csub\u003e2\u003c/sub\u003e/p-Si (100 nm/500 \u0026micro;m) substrate after O\u003csub\u003e2\u003c/sub\u003e plasma cleaning. It was then annealed at 200\u0026deg;C for 10 mins. And use the maskless lithography machine to design the In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e pattern and etch the film with a solution of HCl: H\u003csub\u003e2\u003c/sub\u003eO (1:15 volume ratio). After that, put the film on the heating stage to anneal at 300\u0026deg;C for 1 hour in air. 2H-MoTe\u003csub\u003e2\u003c/sub\u003e film was transferred onto an In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e/SiO\u003csub\u003e2\u003c/sub\u003e/Si substrate using a wet transfer method and was annealed at 200\u0026deg;C for 2 hours to improve the interface contact. Next, use the maskless lithography machine to design the In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e pattern and etch the MoTe\u003csub\u003e2\u003c/sub\u003e film by Ar and O\u003csub\u003e2\u003c/sub\u003e inductively coupled plasma. Finally, the source-drain Ni/Au (8/50 nm) electrodes were deposited through EBE. The W/L ratio is 400/20 \u0026micro;m.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eMaterial and device characterizations\u003c/h2\u003e \u003cp\u003eField emission high resolution images were obtained by transmission electron microscopy (Talos F200X G2). The elemental composition analysis of MoTe\u003csub\u003e2\u003c/sub\u003e and In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e thin films was performed using micro-focus X-ray photoelectron spectroscopy (Thermo Scientific, Nexsa G2). Ultrafast transient absorption fluorescence microspectroscopy system (PH-Tuning, Light Conversion and TA100, Time Tech Spectra) was used to measure the corresponding defects within MoTe\u003csub\u003e2\u003c/sub\u003e. The transmittance of MoTe\u003csub\u003e2\u003c/sub\u003e and In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e thin films was measured using a UV-VIS-NIR microspectrophotometer (CRAIC Technologies Inc, CRAIC 20/30PV).\u003c/p\u003e \u003cp\u003eThe electrical characteristics of MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistor was measure by the Agilent B1500 semiconductor parameter analyzer. The incident light at multiple wavelengths is generated by lasers with wavelengths of 360/450/520/640/800/1064 nm (Changchun New Industries Optoelectronics Tech Co., Ltd, China) and modulated by an arbitrary waveform generator (Tektronix AFG3152C). The light intensity was calibrated by Thorlabs S120VC standard silicon photodiode.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eFormula for calculating the dynamic range of MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistors\u003c/h2\u003e \u003cp\u003e \u003cdiv id=\"Equb\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:DR\\left(dB\\right)=20{log}_{10}\\left(\\frac{{S}_{max}}{{S}_{min}}\\right)\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(2\\right)\\:$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{max}\\:\\)\u003c/span\u003e \u003c/span\u003eis the maximum signal intensity that the detector can process without saturation, while \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{S}_{min}\\:\\)\u003c/span\u003e\u003c/span\u003eis the minimum signal intensity that the detector can detect. Here, the signal intensity refers to the change in light intensity. The dynamic range (DR) physically represents the span of input signal intensities over which a system can reliably detect and process signals, from the minimum discernible level above noise to the maximum level before saturation.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eFormula for calculating the DETR of MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistors\u003c/h2\u003e \u003cp\u003e \u003cdiv id=\"Equc\" class=\"Equation\"\u003e \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equc\" name=\"EquationSource\"\u003e\n$$\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:DETR\\left(dB\\right)=20{log}_{10}\\left(\\frac{{ETS}_{max}}{{ETS}_{min}}\\right)\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\left(3\\right)$$\u003c/div\u003e \u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:{ETS}_{max}\\)\u003c/span\u003e \u003c/span\u003e and \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{ETS}_{min}\\)\u003c/span\u003e\u003c/span\u003e represent the minimum detectable light intensity change under different gate biases. DETR represents the gate-controlled dynamic range of ETS which the detector can reliably modulate its minimum detectable illumination change.\u003c/p\u003e \u003c/div\u003e "},{"header":"Declarations","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eData Availability\u003c/h2\u003e \u003cp\u003eThe main data supporting the findings of this study are available within the Article. Source data are provided with this paper. Additional data and information are available from the corresponding author upon reasonable request.\u003c/p\u003e \u003c/div\u003e\u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interest.\u003c/p\u003e\u003ch2\u003eAuthor contributions\u003c/h2\u003e \u003cp\u003eB.Z., H.W., G.Z. conceived and designed the experiments. G.Z., Z.Z. designed and fabricated the MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e phototransistor array and conduct material mechanism characterization. S.Z. designed the PCB acquisition circuit and an upper computer display system for event imaging. All authors contributed to the preparation of the manuscript. B.Z., H.W. supervised the project.\u003c/p\u003e\u003ch2\u003eAcknowledgments\u003c/h2\u003e \u003cp\u003eThis work is supported by the National Key Research and Development Program of China (2024YFA1208800) funded by MOST, the National Natural Science Foundation of China (Grant Number. 62574169, 62504208), and Scientific Research Project of Westlake University Westlake (No. WU20248032). We thank the Westlake Centre for Micro/Nano Fabrication, the Instrumentation and Service Centre for Physical Sciences (ISCPS), and the Instrumentation and Service Centre for Molecular Sciences (ISCMS) at Westlake University for the facility support and technical assistance. And we thank Dr. Xue LOU from Instrumentation and Service Center for Molecular Sciences at Westlake University for the assistance of transient absorption spectroscopy measurement.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMennel L et al (2020) Ultrafast machine vision with 2D material neural network image sensors. Nature 579:62\u0026ndash;66\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTan H, van Dijken S (2023) Dynamic machine vision with retinomorphic photomemristor-reservoir computing. Nat Commun 14:2169\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen R et al (2026) Integrated bionic LiDAR for adaptive 4D machine vision. Nat Commun 17:24\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGehrig D, Scaramuzza D (2024) Low-latency automotive vision with event cameras. Nature 629:1034\u0026ndash;1040\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang Z et al (2024) A vision chip with complementary pathways for open-world sensing. Nature 629:1027\u0026ndash;1033\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePan J et al (2025) Adaptive spatial-temporal information processing based on in-memory attention-inspired devices. Nat Commun 16:7449\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang H et al (2025) In-sensor compressing via programmable optoelectronic sensors based on van der Waals heterostructures for intelligent machine vision. Nat Commun 16:3836\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eParedes-Valles F et al (2024) Fully neuromorphic vision and control for autonomous drone flight. Sci Robot 9:eadi0591\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHirayama T (2013) \u0026amp;IEEE. The Evolution of CMOS Image Sensors. In \u003cem\u003eAsian Solid-State Circuits Conference\u003c/em\u003e 5\u0026ndash;8IEEE\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang M et al (2025) Dynamic range and precision of hybrid vision sensors. Nature 642:E7\u0026ndash;E8\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin Q et al (2025) Event-driven retinomorphic photodiode with bio-plausible temporal dynamics. Nat Nanotechnol 20:1213\u0026ndash;1220\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou Y et al (2023) Computational event-driven vision sensors for in-sensor spiking neural networks. Nat Electron 6:870\u0026ndash;878\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEuler T, Haverkamp S, Schubert T, \u0026amp;Baden T (2014) Retinal bipolar cells: elementary building blocks of vision. Nat Rev Neurosci 15:507\u0026ndash;519\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGollisch T, Meister M (2010) Eve Smarter than Scientists Believed: Neural Computations in Circuits of the Retina. Neuron 65:150\u0026ndash;164\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMasland R (2012) The Neuronal Organization of the Retina. Neuron 76:266\u0026ndash;280\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu J et al (2025) High-order dynamics in an ultra-adaptive neuromorphic vision device. Nat Nanotechnol 20:1419\u0026ndash;1430\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTiw PJ et al (2026) An end-to-end memristive hardware system based on single-spike coding for human\u0026ndash;machine interfaces. Nat Electron 9:225\u0026ndash;236\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Z et al (2025) A memristor-based adaptive neuromorphic decoder for brain\u0026ndash;computer interfaces. Nat Electron 8:362\u0026ndash;372\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLichtsteiner P, Posch C, Delbruck T (2008) A 128\u0026times;128 120 dB 15 \u0026micro;s latency asynchronous temporal contrast vision sensor. IEEE J Solid-State Circuits 43:566\u0026ndash;576\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJang H et al (2022) In-sensor optoelectronic computing using electrostatically doped silicon. Nat Electron 5:519\u0026ndash;525\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou Y et al A 2T2R1C vision cell with 140 dB dynamic range and event-driven characteristics for in-sensor spiking neural network. In Int Electron Devices Meeting 31\u0026ndash;34 (IEEE 2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKumar M, Park H, Seo HA (2025) Single-Pixel Event Photoactive Device for Real-Time, In-Sensor Spatiotemporal Optical Information Processing. Adv Mater 37:2406607\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFeng S et al (2025) A Charge-Coupled Phototransistor Enabling Synchronous Dynamic and Static Image Detection. Adv Mater 37:2417675\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi K et al (2025) Thin-film event-based vision sensors for enhanced multispectral perception beyond human vision. Infomat 7:e70007\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDang B et al (2026) Spiking neural networks with fatigue spike-timing-dependent plasticity learning using hybrid memristor arrays. Nat Electron 9:213\u0026ndash;224\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi X et al (2025) Federated learning using a memristor compute-in-memory chip with in situ physical unclonable function and true random number generator. Nat Electron 8:518\u0026ndash;528\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong W et al (2024) Programming memristor arrays with arbitrarily high precision for analog computing. Science 383:903\u0026ndash;910\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHahn J et al (2023) Evolution of neuronal cell classes and types in the vertebrate retina. Nature 624:415\u0026ndash;424\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDosenbach N, Raichle M, \u0026amp;Gordon E (2025) The brain's action-mode network. Nat Rev Neurosci 26:158\u0026ndash;168\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStowell R et al (2019) Noradrenergic signaling in the wakeful state inhibits microglial surveillance and synaptic plasticity in the mouse visual cortex. Nat Neurosci 22:1782\u0026ndash;1792\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRavichandran V et al (2026) Memristive cellular neural networks for fast in-pixel computing. Nat Electron 9:189\u0026ndash;199\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePannone A et al (2024) Robust chemical analysis with graphene chemosensors and machine learning. Nature 634:572\u0026ndash;578\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng Y, Ghosh S, Das S (2024) A butterfly-inspired multisensory neuromorphic platform for integration of visual and chemical cues. Adv Mater 36:2307380\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNoudoost B, Chang M, Steinmetz N, Moore T (2010) Top-down control of visual attention. Curr Opin Neurobiol 20:183\u0026ndash;190\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eReimer J et al (2016) Pupil fluctuations track rapid changes in adrenergic and cholinergic activity in cortex. Nat Commun 7:13289\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng Y et al (2021) MoTe\u003csub\u003e2\u003c/sub\u003e: Semiconductor or Semimetal? ACS Nano 15:12465\u0026ndash;12474\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi D et al (2023) Bidirectionally Photoresponsive Optoelectronic Transistors with Dual Photogates for All-Optical-Configured Neuromorphic Vision. Adv Funct Mater 33:2303198\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi D et al (2024) An Active-Matrix Synaptic Phototransistor Array for In-Sensor Spectral Processing. Adv Sci 11\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCelik T (2014) Spatial Entropy-Based Global and Local Image Contrast Enhancement. IEEE Trans Image Process 23:5298\u0026ndash;5308\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhalid S, Medasani B, Lyons J, Wickramaratne D, Janotti A (2024) The deep-acceptor nature of the chalcogen vacancies in 2D transition-metal dichalcogenides. 2d Mater 11:021001\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheng Z et al (2024) P/N-type conversion of 2D MoTe\u003csub\u003e2\u003c/sub\u003e controlled by top gate engineering for logic circuits. ACS Appl Mater Interfaces 16(28):36539\u0026ndash;36546\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcDonough J et al (2008) Thermodynamic, kinetic, and computational study of heavier chalcogen (S, Se, and Te) terminal multiple bonds to molybdenum, carbon, and phosphorus. Inorg Chem 47:2133\u0026ndash;2141\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"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":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-9334649/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9334649/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eEvent-driven vision sensors provide high-temporal-resolution perception with minimal data redundancy, offering significant advantages for machine vision applications. However, conventional designs suffer from high circuit complexity or limited adaptability in complex dynamic scenes. Herein, we present a neuromorphic vision sensor based on MoTe\u003csub\u003e2\u003c/sub\u003e/In\u003csub\u003e2\u003c/sub\u003eO\u003csub\u003e3\u003c/sub\u003e heterojunction phototransistor, enabling retina-like adaptive visual processing. The device achieves microsecond-scale response (1.2/1.0 \u0026micro;s) and a dynamic event threshold range (DETR) exceeding 60 dB. This tunability allows in-sensor processing, such as noise suppression and feature enhancement, eliminating the need for external circuitry. System-level validation is achieved using a 6\u0026times;6 sensor array, which demonstrates denoising effect for raindrop noise in rainy environments and the enhancement of weak signals in foggy conditions. Additionally, we introduce an event-entropy metric to evaluate the effects of threshold modulation on the event stream and establish its quantitative connection to information efficiency. By co-designing materials, devices, and circuits with the novel entropy metric, this work provides a scalable framework for building adaptive, robust, and highly integrated event-based vision systems.\u003c/p\u003e","manuscriptTitle":"Threshold-tunable event-driven vision sensor for adaptive visual processing","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-30 03:08:23","doi":"10.21203/rs.3.rs-9334649/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"43beedec-ceee-44f8-9e45-8e0ff5986c81","owner":[],"postedDate":"April 30th, 2026","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"This content is not available.","date":"2026-05-13T08:03:41+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2026-05-06T08:36:40+00:00","index":2,"fulltext":"This content is not available."},{"type":"reviewerAgreed","content":"This content is not available.","date":"2026-04-29T13:52:50+00:00","index":1,"fulltext":"This content is not available."},{"type":"reviewersInvited","content":"3","date":"2026-04-29T01:19:41+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":67201012,"name":"Physical sciences/Engineering/Electrical and electronic engineering"},{"id":67201013,"name":"Physical sciences/Nanoscience and technology/Nanoscale devices/Electronic devices"}],"tags":[],"updatedAt":"2026-04-30T03:08:24+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-30 03:08:23","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9334649","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9334649","identity":"rs-9334649","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}