A multi-mode monolithic integrated image sensor based on event frame fusion

Abstract

This invention discloses a multi-mode monolithically integrated image sensor based on event frame fusion, belonging to the field of photodetector technology. A single pixel is composed of a thin-film transistor, a capacitor, and a photoelectric conversion device connected in sequence. The overall structure, from bottom to top, includes a substrate layer, a bottom electrode layer, a dielectric layer, a photosensitive layer, a hole transport layer, a buffer layer, and a top electrode layer. This invention utilizes the charging and discharging effect of photogenerated charges on the capacitor, enabling the pixel to output a positive pulse when light intensity increases, a plateau current in steady state, and a negative pulse when light intensity decreases, achieving self-driven event sensing without requiring a continuous bias circuit within the pixel, resulting in extremely low static power consumption. By adjusting the bias voltage, the sensor can switch between dynamic event mode and static intensity mode, balancing high-speed motion capture and background information acquisition. The sensor can be fabricated on a transparent or flexible substrate using thin-film transistor technology, facilitating large-area integration and three-dimensional heterogeneous integration. This invention has advantages such as simplified structure, low power consumption, fast response speed, and rich information dimensions, and can be widely applied in autonomous driving, robot vision, augmented reality, and intelligent edge sensing devices.

Description

Technical Field

[0001] This invention belongs to the field of photodetector and image sensor technology, specifically relating to a multi-mode monolithic integrated image sensor based on event frame fusion. Background Technology

[0002] An event-based camera, also known as a dynamic vision sensor (DVS) or dynamic and active pixel vision sensor (DAVIS), is a biomimetic vision sensor. Unlike traditional cameras that output images at a fixed frame rate, the pixel structure of an event camera consists of photodiodes, current comparators, memory, and triggers. Pixels continuously compare the current light intensity with a reference value from the most recent trigger. When the logarithmic intensity difference exceeds a positive or negative threshold, the sensor immediately outputs an "event," including pixel coordinates, the change in polarity (brightening or darkening), and a timestamp. Because it only responds to changes in light intensity, event cameras offer millisecond-level or even microsecond-level low latency, high temporal resolution, and high dynamic range. Simultaneously, event cameras avoid transmitting redundant data from static backgrounds, significantly reducing data volume and power consumption. Their sensing power consumption can be as low as milliwatts or even lower. Compared to the motion blur and overexposure / underexposure problems that traditional cameras are prone to in high-speed or high-contrast scenes, event cameras can capture motion information more sensitively, and are therefore widely used in high-speed perception scenarios such as autonomous driving, robot vision, and AR / VR.

[0003] With the continuous advancement of event camera technology, its commercial applications in various fields are gradually increasing, demonstrating its advantages in high-dynamic scenarios. For example, the IMX636HD sensor, co-developed by Prophesee and Sony, has been adopted by IDS's uEyeEVS industrial camera series and is widely used in industrial automation scenarios such as high-speed motion analysis and vibration monitoring. Furthermore, Prophesee's GenX320 sensor has been integrated into the AMD Kria Vision AI platform, supporting eye tracking and gesture recognition functions in AR / VR devices. These commercial applications showcase the potential and advantages of event cameras in practical applications.

[0004] Existing commercial event camera products and experimental products still have some shortcomings in terms of structure and performance, which limit their further large-scale promotion. These mainly include the following aspects:

[0005] 1. Complex pixel circuitry and demanding manufacturing processes: To detect brightness changes, traditional DVS / DAVIS pixel circuits typically include several transistors (such as logarithmic photoamplifiers, current comparators, reset / hold amplifiers, charge feedback amplifiers, etc.) and photodiodes. Traditional event camera sensor pixels integrate photodiodes, transimpedance amplifiers, source followers, capacitive feedback amplifiers, comparators, and digital storage units. Their implementation even requires 3D stacking processes to transfer some circuitry to the silicon substrate beneath the pixel. Large pixels with up to 5-6 active devices are not only complex to design and difficult to miniaturize, but also place high demands on semiconductor processes, increasing manufacturing costs and integration difficulty.

[0006] 2. Power Constraint: Although event cameras only output data when there is change, resulting in lower overall power consumption than frame-based cameras of the same resolution, their power consumption is not entirely negligible as the number of pixels increases and the event frequency rises. On the one hand, many pixel circuits require continuous bias current (such as comparator references and current sources) to maintain operational sensitivity, resulting in static power consumption. On the other hand, in high-speed scenarios, the instantaneous event rate increases sharply, increasing the load on data readout and processing circuits. Therefore, reducing pixel static power consumption and optimizing event readout efficiency remain important issues in ultra-large-scale high-resolution arrays or long-term continuous operation.

[0007] 3. System Integration and Application Limitations: While some simplified designs have emerged for existing experimental event cameras aimed at reducing circuit complexity and manufacturing costs, they still face challenges in integrating with large-scale array readout systems. Furthermore, existing event camera technologies exhibit significant limitations in achieving transparency, preventing them from fully realizing their potential in certain specialized applications. Transparency is crucial for some applications (such as automotive driving and transparent displays), but current technologies have not yet effectively integrated and applied semi-transparent event cameras. This limitation severely hinders the practical application and widespread adoption of event cameras across various fields.

[0008] 4. Limited Functional Dimensions and Insufficient Information Extraction Capabilities: Existing experimental event cameras with simple structures primarily output basic event stream data, namely pixel coordinates, polarity (bright / dark), and timestamps. While this data contains dynamic information about the scene, it is low-dimensional and unstructured, making it difficult to directly use in complex autonomous driving applications. Further acquisition of information such as the precise direction and velocity of moving objects (i.e., motion vectors) and even the depth (3D) of the scene is crucial for intelligent vision systems. Commercial CMOS image sensors rely entirely on complex and energy-intensive backend algorithms for extensive computation and inference. This "sensor only collects data, processor then calculates" approach not only increases system latency but also fails to fully leverage the high timeliness advantage of the event sensing front-end.

[0009] In summary, there is an urgent need for a new event camera solution with a simpler structure, lower power consumption, easier backend integration, and a certain degree of transparency, in order to break through existing technological bottlenecks and meet the needs of future intelligent vision systems in terms of large scale, high spatiotemporal resolution, and low power consumption. Summary of the Invention

[0010] In view of this, the purpose of this invention is to provide a multi-mode monolithic integrated image sensor based on event frame fusion. This integrated image sensor has a simplified structure and powerful functions. It not only simplifies the pixel structure, reduces power consumption and improves process compatibility, but also enables multi-dimensional native perception of dynamic information.

[0011] An image sensor is disclosed, wherein a single pixel comprises a thin-film transistor, a capacitor, and a photoelectric conversion device connected in sequence; the structure of the image sensor, from bottom to top, comprises a substrate layer, a bottom electrode layer, a dielectric layer, a photosensitive layer, a hole transport layer, a buffer layer, and a top electrode layer; wherein the bottom electrode layer is used to form the thin-film transistor, the dielectric layer is used to form the capacitor, the photosensitive layer is used to form the photoelectric conversion device, and the top electrode layer is used to form the electrode of a photodiode.

[0012] Preferably, the thin-film transistor is one of an oxide semiconductor thin-film transistor, a polycrystalline silicon thin-film transistor, an organic semiconductor thin-film transistor, a MOSFET transistor, a junction field-effect transistor, or a bipolar transistor.

[0013] Preferably, the photosensitive layer is made of one of the following materials: inorganic semiconductor material, organic optoelectronic semiconductor material, perovskite material, or quantum dot material.

[0014] Preferably, the substrate is a transparent substrate, an opaque substrate, or a flexible plastic substrate.

[0015] Preferably, the photoelectric conversion device is a phototransistor or a resistive photosensitive device; or, the photoelectric conversion device is a photodiode, the structure of which is a planar heterojunction, a vertically stacked PN junction, or an MSM diode.

[0016] Preferably, the dielectric layer is a single layer or a combination of multiple layers; the material of the dielectric layer is selected from one or more of hafnium oxide, aluminum fluoride, silicon nitride, parylene or polymethyl methacrylate; or, the dielectric layer is a double dielectric layer or a composite dielectric layer, including a double-gate or multi-gate structure of a phototransistor, or a capacitive coupling element introduced between a photodiode and a thin-film transistor.

[0017] Preferably, the bottom electrode layer is an indium tin oxide electrode or a CMOS addressing electrode; the dielectric layer is an aluminum oxide thin film formed by atomic layer deposition; the photosensitive layer is a silver bismuth sulfur quantum dot thin film formed by spin coating; the hole transport layer is a polytriarylamine thin film formed by spin coating; the buffer layer is a molybdenum trioxide thin film; and the top electrode is a silver or indium tin oxide transparent electrode.

[0018] Preferably, by adjusting the bias voltage applied to the top electrode layer, the image sensor can operate in either a dynamic event mode or a static intensity mode; in dynamic event mode, the sensor outputs an event signal characterizing changes in light intensity; in static intensity mode, the sensor outputs a static signal characterizing ambient light intensity.

[0019] Preferably, the pixels of the image sensor are configured such that: when the incident light intensity increases, the photoelectric conversion device generates transient photogenerated charge, and the pixel outputs a positive pulse current through capacitive coupling; when the incident light intensity stabilizes, it outputs a plateau current; and when the incident light intensity decreases, the capacitor discharges, causing the pixel to output a negative pulse current.

[0020] Preferably, the image sensor further includes a readout circuit, which is a CMOS chip or an external high-speed analog front-end circuit; the pixel array of the image sensor is electrically connected to the CMOS chip via flip-chip bonding or three-dimensional integration.

[0021] The present invention has the following beneficial effects: 1. Significantly simplified pixel structure and reduced manufacturing costs: Each basic pixel unit requires only one TFT and one photodiode (which also functions as a capacitor), replacing multiple transistors and amplification / comparison devices in traditional architectures. This makes the pixel circuitry more compact, the process flow simpler, and the yield higher. The array can be mass-produced directly using spin-coating technology, and large-area fabrication is inexpensive, which is beneficial for the economic realization of high-resolution / large-size sensors.

[0022] 2. Self-driven event sensing with ultra-low static power consumption: Utilizing light signals to trigger pixel output eliminates the power overhead of continuous biasing in traditional pixel circuits, resulting in low array power consumption when no event occurs. Relevant pixels only output pulses instantaneously when the scene changes. This on-demand triggering characteristic makes the average power consumption significantly lower than that of a comparable frame camera. Especially in static scenes, this sensor can monitor the environment with low power consumption, immediately sending an event signal to alert subsequent circuitry upon the appearance of a moving target, greatly improving system energy efficiency.

[0023] 3. Diverse Modes, Complete Information Acquisition: By employing bias voltage control, this invention achieves dual modes: dynamic events and static images. It combines the advantages of traditional cameras and event cameras within the same hardware. Dynamic mode captures high-speed motion details and sparsely changing data; static mode acquires environmental background and slowly changing information. The data from both modes complement each other, improving the completeness and robustness of environmental perception. For example, in autonomous driving, event mode can be used to detect suddenly appearing pedestrians or fast-moving objects, while static mode can be used to acquire the positions of road signs and stationary obstacles, thus achieving more comprehensive perception.

[0024] 4. Easy Large-Area Integration and System Integration: Due to the use of thin-film processes such as TFT, the sensor of this invention can be fabricated on large-area substrates, which is crucial for ultra-wide field-of-view cameras and panoramic monitoring sensors. The sensor can also be directly integrated onto curved surfaces or wearable carriers such as robot surfaces, smart clothing, and augmented reality glasses, providing biomimetic vision capabilities—something difficult to achieve with silicon-based chip cameras. Furthermore, this sensor is easily integrated with other planar electronic components: for example, integration with a display screen enables self-sensing display (the display senses the viewer's movements in real time); integration with a memory computing unit can build an edge intelligent vision SoC. In terms of semiconductor heterogeneous integration, the pixels of this invention, acting as photosensitive layers, can be bonded to CMOS neuromorphic processing chips to form a "three-dimensional integrated" event camera, improving system integration and response speed.

[0025] 5. Wide Range of Applicable Scenarios: The low power consumption, self-driving, and high dynamic response characteristics of this invention make it ideal for edge computing devices and unmanned systems. In edge AI, the sensor can be continuously on standby with low power consumption. Once an event is captured, it immediately triggers the backend AI accelerator for processing, enabling event-driven on-demand computing that is energy-efficient and highly effective. In unmanned systems such as drones and unmanned vehicles, the sensor of this invention is lightweight and has low power consumption, acting as a continuous "visual touch" for vigilance. It can promptly detect environmental changes (such as the sudden appearance of obstacles) and notify the main vision system or control system to intervene, improving the safety and real-time performance of autonomous decision-making. Furthermore, in fields such as biomedical imaging, security monitoring, and industrial inspection, this sensor can leverage its advantages of high speed, wide dynamic range, and low data volume to provide new dimensions of information that are difficult for traditional cameras to obtain, expanding the boundaries of applications.

[0026] 6. Empowering Efficient Backend Algorithms and Enhancing Overall System Performance: The greatest advantage of this invention lies in its ability to empower more efficient backend processing algorithms with the generated data, thereby achieving a leap in performance at the system level. By supporting the "single-frame optical flow" algorithm, this invention shortens the time window required for motion perception from the traditional multi-frame interval to a very small time slice, greatly reducing perception latency, which is crucial for time-sensitive applications such as autonomous driving and high-speed tracking. Simultaneously, the high sparsity and high information density data provided by this invention simplifies the processing flow of backend algorithms (such as optical flow and depth estimation). This means that, while achieving the same performance indicators, the backend processor requires less computation and consumes less power, contributing to improved energy efficiency of the entire system. Furthermore, the improved source data quality reduces the difficulty of developing complex dynamic vision algorithms. Developers can leverage the data characteristics of this sensor to design simpler and more robust algorithms, thereby accelerating product development cycles and reducing the overall complexity and cost of the system.

[0027] In summary, this invention innovatively combines TFTs, photodiodes, and dielectric layers to achieve a simple, self-driven, low-power, easily integrated event camera sensor with dual-mode functionality. It is expected to simplify the hardware of intelligent vision systems, enabling more efficient event perception, and has broad application prospects and significant technological value in bionic vision, IoT sensing, and intelligent edge devices. It not only solves the bottlenecks of existing event cameras in terms of pixel complexity and power consumption, but also provides a new approach and pathway for future large-area integration and multifunctional visual sensing. Attached Figure Description

[0028] Figure 1 In the diagram, 'a' represents a circuit diagram of a single pixel based on a TFT chip image sensor, and 'b' represents a circuit diagram of a TFT imaging chip with 256×256 pixels. Figure 2 A schematic diagram of a single pixel structure (left) and a schematic diagram of the corresponding cross-sectional structure of a pixel in a TFT chip-based imaging sensor (right). Figure 3 The transmittance curve of a semi-transparent device; Figure 4 The it curve of the sensor during one switching on and off cycle; Figure 5 This is a real-time image captured by the event camera. Image a is a screenshot from the original video of vehicles moving at the intersection; image b is a real-time event image of the moving vehicles captured by the event sensor. Figure 6 In the diagram, 'a' represents the single-frame event information simulated by the device, and 'b' represents the direction and intensity of the corresponding optical flow graph. Figure 7This represents the depth information extraction from the multimodal sensor. 'a' represents the platform current of the device under different bias voltages and light intensities. 'b' represents the event image acquired in real-time by the event sensor. 'c' represents the 3D depth estimate output by the depth estimation network, combining the event signal and the static signal.

[0029] Among them, 1-substrate layer, 2-bottom electrode layer, 3-dielectric layer, 4-photosensitive layer, 5-hole transport layer, 6-buffer layer, 7-top electrode layer Detailed Implementation The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0030] This invention mainly solves the following key technical problems: 1. Simplified pixel structure: By using only a single thin-film transistor (TFT), a photodiode, and a dielectric layer (capacitor) to form the basic pixel, direct sensing of light intensity changes is achieved, avoiding complex analog circuits and significantly reducing the complexity and manufacturing difficulty of the pixel circuitry. Figure 1 As shown.

[0031] 2. Self-driven low power consumption: The event signal output is directly driven by the charge / current generated by photoelectric conversion, eliminating the need for a continuous bias voltage within the pixel, resulting in lower static power consumption per pixel. Transient current only occurs during light change events, significantly reducing overall energy consumption.

[0032] 3. Easy integration and expansion: Employing mature thin-film processes (such as oxide TFT technology) to fabricate sensor arrays on large-area substrates enables the manufacturing of sensors on flexible and transparent substrates, and allows for planar integration with TFT-based devices such as displays and memory. It also supports hybrid integration with CMOS readout circuits to meet diverse application requirements.

[0033] 4. Dynamic-Static Dual-Mode Sensing: Provides a bias control scheme for pixel working modes, which can switch between dynamic event response mode (outputting only changes) and static intensity acquisition mode (outputting ambient light intensity), enabling the sensor to capture both rapid changes and static scene information, making it a dual-purpose device and improving environmental adaptability.

[0034] 5. High-dimensional information perception: This invention aims to output event data with high temporal resolution and high information density through its unique transient impulse response characteristics, so as to analyze the directional information of moving targets and further support the backend to realize efficient motion analysis algorithms such as optical flow from single-frame images, thereby fundamentally reducing perception latency and breaking through the limitations of traditional motion calculation relying on multiple frames of images.

[0035] The material selection and integration method technical solution uses indium gallium zinc oxide (IGZO) thin-film transistor (TFT) chips or CMOS chips as the readout circuit. A schematic diagram of the device structure and cross-section of a single pixel is shown below. Figure 2 As shown, the specific preparation process is as follows: 1. On the surface of the drain electrode of a thin-film transistor (TFT) (i.e., indium tin oxide (ITO) electrode or CMOS addressing electrode), an aluminum oxide (Al2O3) thin film deposited by an atomic layer deposition (ALD) system is used as a dielectric layer. 2. Spin-coated silver bismuth sulfur quantum dot (AgBiS2) thin film is used as a photosensitive layer to convert optical signals into electrical signals; 3. Spin-coating polytriarylamine (PTAA) as a hole transport layer; 4. Deposit a buffer layer of MoO3 and a top electrode (Ag or ITO). Using a transparent ITO electrode as the top electrode can achieve semi-transparency of the event image sensor. The transmittance curve of the device is shown in the figure. Figure 3 As shown, the top electrode (Ag / ITO) of the photodiode is connected to an external bias voltage or reference potential.

[0036] In the pixel of this invention, when the light intensity suddenly increases, the photodiode generates a momentary photogenerated charge, charging the capacitor. This process manifests as a positive pulse current output at the pixel output terminal (corresponding to a bright event). Subsequently, if the light intensity remains steady, the photodiode current rapidly decreases, the dielectric layer capacitor charging tends to balance, and the output current enters a plateau current stage (i.e., maintaining a small steady-state current or near-zero current). When the light weakens or disappears, the aluminum oxide, acting as the dielectric layer, undergoes capacitor discharge, and the photodiode generates a momentary reverse photogenerated current, outputting a negative turn-off pulse current (corresponding to a dark event). Afterward, the pixel returns to its initial equilibrium state. In one cycle of light intensity increase / decrease, a three-stage transient response current characteristic of "positive pulse – plateau – negative pulse" is generated, precisely corresponding to the increase, duration, and decrease of light intensity, as shown below. Figure 4 As shown. This self-driven transient response mechanism does not require complex amplification and comparison circuits; it can achieve a sensitive response to events similar to that of the biological retina solely based on the physical characteristics of the device.

[0037] The high sparsity and low power consumption of the output data in this invention stem directly from its "positive pulse-plateau-negative pulse" physical response mechanism. Under 0 V bias, the output recovers to a very low level under continuous constant illumination, thus capturing only light change events. Finally, by modulating the TFT conduction state, the event signal is output in real time, thereby consuming instantaneous power, such as... Figure 5 As shown, ( Figure 5 'a' is a screenshot from the original video of vehicles moving at the intersection. Figure 5(b shows the real-time imaging of event signals from moving vehicles by the event sensor). This operating mode, determined by the physical characteristics of the device, fundamentally avoids redundant acquisition and transmission of static background information, achieves high data sparsity, and keeps overall power consumption at an extremely low level.

[0038] The event data output by this invention, due to its unique transient response characteristics, contains multi-dimensional event information between two frames of a traditional CMOS sensor, making it possible to develop an efficient back-end optical flow extraction algorithm. Traditional frame-based algorithms for CMOS image sensors require at least two frames (times t and t+1) to calculate optical flow through spatiotemporal gradient comparison. This invention, however, can generate dense event information in an extremely short time. The microsecond-level precise timestamps and pulse characteristics of each event provide the data foundation for "single-frame" optical flow extraction. The back-end algorithm can directly calculate the motion field (optical flow) of an object by analyzing the spatial distribution and high-precision temporal relationships of single-frame events, thereby obtaining the motion direction and velocity of each pixel. This ability to "output optical flow in a single frame" is a fundamental improvement over traditional two-frame or multi-frame methods, significantly reducing motion perception latency, and is one of the core advantages of this invention compared to existing technologies. Figure 6 As shown in a and b, the present invention can simulate single-frame event information output by the sensor ( Figure 6 a) The corresponding optical flow field is directly generated through a motion fitting algorithm. Figure 6 (b) thus enabling real-time perception of the direction and intensity of object motion with low latency.

[0039] When static information needs to be extracted, a bias voltage can be applied to switch the device to a mode that can simultaneously extract event and static information. The applied bias voltage increases the probability of steady-state current quantum tunneling, increases the intensity of the steady-state current, and allows for the acquisition of static scene information, such as… Figure 7 As shown in a. Through event signals ( Figure 7 b) Combining with static information can extract depth information of the current scene, such as... Figure 7 As shown in c.

[0040] To cover a broader scope of protection, the present invention also includes the following variations and alternative embodiments, which are consistent with or equivalent to the core ideas described above: 1. Different transistor types and diverse readout circuits: In alternative implementations, the pixel switching device can employ different types of transistors. This solution uses oxide semiconductor TFTs, such as IGZO (indium gallium zinc oxide) thin-film transistors. Alternatively, polycrystalline silicon (LTPS) TFTs can be used to improve driving capability, or organic semiconductor TFTs can be used to realize flexible, bendable devices. Besides thin-film field-effect transistors (TFTs), MOSFET transistors, junction field-effect transistors (JFETs), bipolar transistors, etc., can also be selected. The selected transistor must be able to adapt to different application scenarios by modulating its conduction state and ensure the desired photoelectric conversion effect is achieved.

[0041] This invention is not limited to TFT technology; it also realizes an event camera using a 640 × 512 CMOS integrated readout circuit, and even directly integrates neuromorphic processing circuitry into the system. In a pure TFT implementation, column line signals can be read through an external high-speed analog front-end, or the timing of pulse occurrences can be recorded using time coding. In a hybrid CMOS implementation, the pixel array can be directly integrated onto the CMOS signal processing chip via flip-chip bonding, enabling direct digital output of each pixel signal. Furthermore, simple pulse counting circuits or programmable threshold circuits can be integrated near each pixel (e.g., by fabricating simple circuits on the TFT layer or combining them with the CMOS side) to implement different event decision logics. These variations in readout circuitry all fall within the scope of this invention.

[0042] 2. Different photosensitizers, substrate materials, and sensor architectures: The photosensitive materials for photoelectric sensors can also be diversified, allowing for the selection of different photosensitive materials according to requirements. For example, organic optoelectronic semiconductor materials (such as P3HT:PCBM, etc.), perovskite materials (such as halide perovskite thin films), and quantum dot materials (such as silver bismuth sulfur quantum dots (AgBiS2), lead sulfide (PbS), etc.) can be used. These materials can effectively absorb light energy in different spectral ranges and convert it into electric current.

[0043] This invention can be fabricated on transparent substrates (such as glass, quartz, sapphire) to achieve front-side incident light illumination; it can also be fabricated on opaque substrates (silicon, silicon dioxide, metal foil, etc.) but uses a transparent top electrode / encapsulation layer to introduce light signals. In flexible applications, flexible plastic substrates such as polyimide (PI) and paeylene can be used. The device can be encapsulated in a thin film to ensure the stability of the photosensitive layer while retaining the light incident window.

[0044] This invention is not limited to traditional PIN photodiodes. Various alternatives to photoelectric conversion elements can be used to meet different application requirements. For example, a phototransistor can be used, where the TFT itself is made into a photosensitive TFT, and the conduction state is adjusted by the change in its threshold voltage under illumination. This approach can improve the integration and response speed of the device. Another alternative is to use a resistive photosensitive device (such as a photoresistor or photoconductor connected in series with a capacitor to form an RC circuit). In this structure, the resistance of the photosensitive device changes in response to changes in illumination, thus affecting the circuit response; this design can also achieve photoelectric conversion. Furthermore, the structure of the photodiode can be designed as planar (lateral heterojunction) or vertically stacked (vertical PN junction or MSM diode) to optimize the efficiency and stability of photoelectric conversion and meet different performance requirements.

[0045] 3. Selection of dielectric layer structure: In an alternative embodiment of the invention, the dielectric layer design between the photodiode and the TFT array can employ a single-layer or multi-layer material combination to achieve optimal capacitive coupling and charge storage effects. In a single-layer dielectric layer design, inorganic materials with high dielectric constants, such as hafnium oxide (HfO2), aluminum fluoride (AlF3), or silicon nitride (Si3N4), and organic high-dielectric polymer materials, such as Parylene or PMMA, are typically chosen. These materials possess excellent dielectric properties and low leakage characteristics, which contribute to stable charge storage and regulation of capacitive coupling between the photodiode and the TFT.

[0046] Double dielectric layers or composite dielectrics can be used, such as depositing multiple layers of photosensitive high-dielectric materials in the readout circuit and photosensitive element, including dual-gate or multi-gate structures of phototransistors, or introducing capacitive coupling elements (such as exposed capacitors) between the photodiode and the TFT. This can enhance or adjust the effect of photogenerated charge on the capacitor, thereby adjusting the response sensitivity and time constant. For example, increasing the capacitance value can prolong the pulse width (increase the time constant), and adding a dielectric buffer layer can filter noise.

[0047] 4. Spectrum and photosensitivity adjustment: Different materials can be replaced to achieve event detection in specific spectral bands. For example, narrow-bandgap semiconductors (such as InGaAs) can be used to sense infrared light, or a light filter layer / color filter can be added in front of a photodiode to achieve a color event camera. Even multi-layer optoelectronic structures can be stacked to achieve multispectral event detection (different layers generate events for different wavelengths of light). These are all extended applications of this invention. In short, this invention covers any sensor structure and system that utilizes photosensitive elements coupled with capacitive coupling to generate a self-driven event response and outputs it through a field-effect switch. The various modifications described above can be combined according to specific applications to expand upon the core ideas of this invention without departing from its scope.

[0048] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An image sensor, characterized in that, A single pixel includes a thin-film transistor, a capacitor, and a photoelectric conversion device connected in sequence; the structure of the image sensor, from bottom to top, includes a substrate layer, a bottom electrode layer, a dielectric layer, a photosensitive layer, a hole transport layer, a buffer layer, and a top electrode layer; wherein, the bottom electrode layer is used to form the thin-film transistor, the dielectric layer is used to form the capacitor, the photosensitive layer is used to form the photoelectric conversion device, and the top electrode layer is used to form the electrode of the photodiode.

2. An image sensor according to claim 1, characterized in that, The thin-film transistor is one of oxide semiconductor thin-film transistors, polycrystalline silicon thin-film transistors, organic semiconductor thin-film transistors, MOSFET transistors, junction field-effect transistors, or bipolar transistors.

3. An image sensor according to claim 1, characterized in that, The photosensitive layer is made of one of the following materials: inorganic semiconductor material, organic optoelectronic semiconductor material, perovskite material, or quantum dot material.

4. An image sensor according to claim 1, characterized in that, The substrate layer is a transparent substrate, an opaque substrate, or a flexible plastic substrate.

5. An image sensor according to claim 1, characterized in that, The photoelectric conversion device is a phototransistor or a resistive photosensitive device; or, the photoelectric conversion device is a photodiode, the structure of which is a planar heterojunction, a vertically stacked PN junction, or an MSM diode.

6. An image sensor according to claim 1, characterized in that, The dielectric layer is a single layer or a combination of multiple layers; the material of the dielectric layer is selected from one or more of hafnium oxide, aluminum fluoride, silicon nitride, parylene or polymethyl methacrylate; or, the dielectric layer is a double dielectric layer or a composite dielectric layer, including a double-gate or multi-gate structure of a phototransistor, or a capacitive coupling element introduced between a photodiode and a thin-film transistor.

7. An image sensor according to claim 1, characterized in that, The bottom electrode layer is an indium tin oxide electrode or a CMOS addressing electrode; the dielectric layer is an aluminum oxide thin film formed by atomic layer deposition; the photosensitive layer is a silver bismuth sulfur quantum dot thin film formed by spin coating; the hole transport layer is a polytriarylamine thin film formed by spin coating; the buffer layer is a molybdenum trioxide thin film; and the top electrode is a silver or indium tin oxide transparent electrode.

8. An image sensor according to claim 1, characterized in that, By adjusting the bias voltage applied to the top electrode layer, the image sensor can operate in either a dynamic event mode or a static intensity mode; in dynamic event mode, the sensor outputs an event signal characterizing changes in light intensity. In static intensity mode, the sensor outputs a static signal characterizing the ambient light intensity.

9. An image sensor according to claim 1, characterized in that, The pixels of the image sensor are configured such that when the incident light intensity increases, the photoelectric conversion device generates transient photogenerated charges, and the pixels output a positive pulse current through the capacitive coupling. When the incident light intensity is stable, the output plateau current is reached. When the incident light intensity decreases, the capacitor discharges, causing the pixel to output a negative pulse current.

10. An image sensor according to claim 1, characterized in that, The image sensor also includes a readout circuit, which is a CMOS chip or an external high-speed analog front-end circuit; the pixel array of the image sensor is electrically connected to the CMOS chip through flip-chip bonding or three-dimensional integration.

Images