Bio-inspired Moving Target Sensing and Computing-in-Memory Photodetector System Based on Persistent Photoconductivity

Abstract

The present application discloses a biomimetic moving target sensing and computing-in-memory optoelectronic detection system based on persistent photoconductivity, which relates to the fields of biomimetic intelligent vision and brain-inspired computing, and includes an interconnected synaptic-type biomimetic optoelectronic detector array and a brain-inspired memristor array; when the synaptic-type biomimetic optoelectronic detector array detects the visual information of moving targets, a fixed bias voltage is applied to the devices in each pixel, and the spatial illumination information corresponding to each pixel is converted into a continuously decaying photocurrent signal. During the continuous decay of the photocurrent, the illumination information of multiple frames of dynamic targets is continuously received. After the input of the last frame of illumination information, the photocurrent signal of each pixel is read to determine the feature map that memorizes all the information at historical moments; the brain-inspired memristor array performs deep neural network operations on the feature map based on Ohm's law and Kirchhoff's law, and outputs a current corresponding to the task result label to identify dynamic targets, capable of meeting the requirements of deep neural networks and complex dynamic vision tasks.

Description

Technical Field

[0001] This application relates to the fields of bionic intelligent vision and brain-like computing, and particularly to a bionic moving target sense-computing integrated optoelectronic detection system based on persistent photoconductivity. Background Art

[0002] In recent years, with the rapid development of the application fields of artificial intelligence, many new application scenarios have emerged continuously, which puts higher requirements on detection systems. Traditional detection systems usually adopt a discrete architecture, and imaging computing faces high latency, making it difficult to meet the requirements of rapid detection, recognition, and judgment in fields such as intelligent driving and robot vision. Therefore, there is an urgent need to develop a lightweight and intelligent infrared optoelectronic sensing system with edge computing capabilities to achieve rapid detection and recognition of targets.

[0003] In dynamic motion recognition tasks, due to the continuous movement of objects, the detector needs to perform frequent frame-by-frame analysis to ensure that the motion trajectory and dynamic changes can be accurately captured and recognized. This frame-by-frame analysis paradigm has led to a sharp increase in the amount of data transmission, posing extremely high requirements on data transmission efficiency.

[0004] Currently, the discrete architecture of the detection and processing modules in traditional visual perception systems results in a relatively single function of bionic sense-computing integrated optoelectronic detection systems. They can only build simple single-layer perception models and perform basic image preprocessing tasks, making it difficult to meet the requirements of deep neural networks and complex dynamic vision tasks. Summary of the Invention

[0005] The purpose of this application is to provide a bionic moving target sense-computing integrated optoelectronic detection system based on persistent photoconductivity to solve the problem that existing bionic sense-computing integrated optoelectronic detection systems are difficult to meet the requirements of deep neural networks and complex dynamic vision tasks.

[0006] To achieve the above purpose, this application provides the following solutions.

[0007] In a first aspect, this application provides a bionic moving target sense-computing integrated optoelectronic detection system based on persistent photoconductivity, including: an interconnected synaptic-type bionic optoelectronic detector array and a brain-like memristor array.

[0008] When the synaptic-type bionic optoelectronic detector array is used to detect moving target visual information, a fixed bias voltage is applied to the devices in each pixel in the synaptic-type bionic optoelectronic detector array, and the spatial illumination information corresponding to each pixel is converted into a continuously decaying photocurrent signal. During the continuous decay of the photocurrent, the illumination information of multiple frames of dynamic targets is continuously received, and after the input of the last frame of illumination information, the photocurrent signal of each pixel is read to determine the feature map that memorizes all historical moment information.

[0009] The brain-like memristor array is used to perform deep neural network operations on the feature map based on Ohm's law and Kirchhoff's law, output a current corresponding to the task result label, and identify dynamic targets based on the current; the current is used to characterize the probability of the task result label.

[0010] According to the specific embodiments provided in this application, this application discloses the following technical effects.

[0011] When the present application uses a synaptic-type bionic photodetector array to detect the visual information of moving targets, a fixed bias voltage is applied to the device in each pixel to convert the spatial illumination information corresponding to each pixel into a continuously decaying photocurrent, and the illumination information of multiple frames of dynamic targets is continuously received during the continuous decay of the photocurrent, without reading the electrical signal frame by frame. The magnitude of the photocurrent continues to decay or increase over time, and the photocurrent state of each pixel in the array is read after the last frame of light information is input to obtain a feature map that memorizes all historical moment information, thereby realizing multi-frame fusion detection calculation in the intra-temporal domain; the photocurrents extracted after all pixel detections are combined into a feature map as the output of the first-layer network, which is input into the brain-like memristor array to perform deep neural network operations, and finally the current corresponding to the task result label is output to represent its probability, thereby completing the recognition and classification tasks of dynamic targets; a storage-computation integrated deep neural network operation is performed in the memristor array, the conductance value of each memristor in the memristor array is read by applying a voltage sequence, the output current of each memristor is calculated, and the output current is used as the result of the recognition task. Among them, the synaptic-type bionic photoelectric detector array utilizes the continuous photoconductivity effect of the detector when performing multi-frame fusion detection and reading in the time domain, and the photocurrent decays nonlinearly with time, while completing the linear and nonlinear calculations of the single-layer neural network; the brain-like memristor array realizes fully connected neural network calculations through Ohm's law and Kirchhoff's law, outputs current results, and completes the dynamic target classification task.

[0012] Compared with the traditional visual perception system's discrete architecture of detection and processing modules and the frame-by-frame transmission computing paradigm, this application can meet the needs of deep neural networks and complex dynamic visual tasks, greatly improve the network's recognition speed, reduce network power consumption, and achieve rapid recognition of moving targets at the full hardware level. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 Schematic structural diagram of the retina-like photodetector unit of the present application.

[0015] Figure 2 Schematic structural diagram of the positive and negative electrodes of the retina-like photodetector unit of the present application.

[0016] Figure 3 Schematic diagram of the memristor array with 1T1R structure adopted by the present application.

[0017] Figure 4 IV characteristic curve diagram of the retina-like photodetector unit of the present application.

[0018] Figure 5 Continuous photoconductance characteristic curve diagram of the retina-like photodetector unit of the present application.

[0019] Figure 6 Schematic structural diagram of the 4×4 pixel synaptic-like bionic photodetector array of the present application.

[0020] Figure 7 Schematic diagram of the bionic moving target sensing and computing integrated photodetection system of the present application.

[0021] Figure 8 Conductance value heat map of the memristor cross array of the present application.

[0022] Figure 9 Recognition accuracy curve diagram for the recognition task scenario of dynamic targets of the present application. Detailed implementation manners

[0023] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0024] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0025] To achieve large-scale integration devices for sensing, memory, and computing, which support the processing of deep neural networks and complex image recognition tasks, it is urgent to break through networked integration technologies and promote the joint design of synaptic-like bionic photodetectors and brain-inspired neural network hardware. Memristors, with their brain-synapse-like functions and excellent integration capabilities, provide new opportunities for sensing, memory, and computing integration technologies. However, there is currently no research on bionic intelligent photodetector arrays based on persistent photoconductance for multi-frame integration detection and their integration with brain-inspired memristor arrays.

[0026] To meet the high requirements for data transmission efficiency, improving the processing ability inside the detector has become the key to solving the problems of energy consumption and latency in machine vision systems. Based on research on some specific materials and device structures, the applicant found that detector materials and structures with persistent photoconductance effects can provide good ideas for solving this bottleneck. In such devices, whenever the light illumination ends, the photocurrent decays over time instead of disappearing instantaneously. By utilizing the persistent photoconductance effect of the detector, this application proposes a new paradigm for moving target recognition based on multi-frame integration imaging calculation, continuously detecting multi-frame optical signals during the continuous decay of the photocurrent, storing the continuous multi-frame image information in a single frame for reading out, which can reduce the data transmission and processing times by a multiple while not affecting the accuracy, and is very beneficial for reducing power consumption and latency.

[0027] Based on this, an embodiment of this application provides a bionic moving target sensing, memory, and computing integrated photodetection system based on persistent photoconductance, including: an interconnected synaptic-like bionic photodetector array and a brain-inspired memristor array.

[0028] When the synaptic-like bionic photodetector array is used to detect moving target visual information, a fixed bias voltage is applied to the devices in each pixel in the synaptic-like bionic photodetector array, converting the spatial illumination information corresponding to each pixel into a continuously decaying photocurrent signal, continuously receiving the illumination information of multiple frames of dynamic targets during the continuous decay of the photocurrent, and after the input of the last frame of illumination information, reading the photocurrent signal of each pixel to determine the feature map that memorizes all historical moment information.

[0029] The brain-inspired memristor array is used to perform deep neural network operations on the feature map based on Ohm's law and Kirchhoff's law, outputting a current corresponding to the task result label, and identifying the dynamic target according to the current; the current is used to characterize the probability of the task result label.

[0030] In an exemplary embodiment, the synaptic-like bionic photodetector array specifically includes: a plurality of pixels distributed in an array, and each pixel contains a retinal-like photodetector unit.

[0031] In practical applications, the synaptic-like bionic photodetector array includes 4×4 pixels, and each pixel contains 1 retinal-like photodetector unit, with a total of 16 retinal-like photodetector units.

[0032] In an exemplary embodiment, each of the retinal-like photodetector units includes a substrate, a light absorption layer, and two crossed interdigital electrodes stacked in sequence from bottom to top; among them, the light absorption layer region corresponding to the channel formed between the two crossed interdigital electrodes is the light response region; the light response region is used to provide a linear light response and non-linear attenuation of the photocurrent, and the voltage of the electrodes at both ends of each retinal-like photodetector unit is regulated to encode the light response of the retinal-like photodetector unit.

[0033] In an exemplary embodiment, the substrate is an insulating Al2O3 substrate; the light absorption layer is a MoS2 two-dimensional material; patterning design is performed on the MoS2 two-dimensional material through a photolithography etching process, and metal electrodes are deposited on the surface of the MoS2 two-dimensional material through electron beam evaporation; the metal electrodes are the interdigital electrodes.

[0034] In an exemplary embodiment, the interdigital electrodes are in a rectangular interdigital structure, forming multiple light response units; the two interdigital electrodes include a positive electrode and a negative electrode; the positive electrode and the negative electrode are arranged alternately.

[0035] In practical applications, the widths of the positive electrode and the negative electrode are 15 μm, and the spacing between adjacent electrodes is 5 μm.

[0036] In an exemplary embodiment, the light responsivity of the retinal-like photodetector unit is a randomly fixed value;

[0037] The photocurrent signal is the product of the light responsivity and the light intensity.

[0038] In an exemplary embodiment, the photocurrent signal shows non-linear continuous attenuation between frames.

[0039] In practical applications, the synaptic-like bionic photodetector array realizes single-layer non-linear recurrent neural network calculation. The light responsivity R of the synaptic-like bionic photodetector array represents the network weight value, and the light intensity P is used as the input signal of the network. Then the generated photocurrent signal is .

[0040] After the photocurrent is generated, it continuously decays. During the continuous decay of the photocurrent, it continuously receives the illumination of multiple frames of dynamic targets. The photocurrent signal at the next moment is the remaining photocurrent signal after the previous moment and the light intensity generated by the illumination at the next moment jointly determine, that is .

[0041] In an exemplary embodiment, the brain-inspired memristor array uses word line circuits, source line circuits, and bit line circuits to interweave into a mesh layout; a memristive element is arranged at each intersection in the mesh layout; these three circuits are used to input or read signals in the memristive element.

[0042] The memristive element includes a transistor and a memristor.

[0043] The word line circuit is connected to the gate of the transistor, the source line circuit is connected to the top electrode of the memristor, and the bit line circuit is connected to the bottom electrode of the memristor.

[0044] The bottom electrode of the memristor is also directly connected to the drain of the transistor; by adjusting the gate voltage of the transistor, the conductive state of the brain-inspired memristor array can be effectively controlled; the conductive state includes a conducting state and a non-conducting state. This design can effectively solve the signal crosstalk problem and improve the performance and stability of the brain-inspired memristor array.

[0045] In practical applications, each memristor in the brain-inspired memristor array can adjust its resistance value according to an applied voltage, and after removing the external voltage, it can maintain the conductance value unchanged and can stably maintain the adjusted resistance state, so as to deploy the weights of the hardware deep neural network, and the hardware deep neural network is the brain-inspired memristor array.

[0046] As the hardware basis for neural network implementation, these memristors can accurately map and write the weights of the hardware deep neural network obtained through training into their respective resistance values by applying corresponding voltages.

[0047] In an exemplary embodiment, the top layer of the memristor array is the top electrode layer, the middle layer is the resistive switching layer, and the bottom layer is the bottom electrode layer.

[0048] Input voltage is applied through row selection of the top electrode layer, and current is output through column selection of the bottom electrode layer.

[0049] The memristor array follows Ohm's law and Kirchhoff's law to complete multiplication and accumulation operations, reads currents from each column, completes the vector matrix multiplication operation in the deep neural network, and realizes a hardware deep neural network for in-memory computing.

[0050] In practical applications, the brain-inspired memristor array performs in-memory computing deep neural network operations, and the extracted feature map is input into the hardware deep neural network for recognition. The voltage values expanded for each pixel of the image information are obtained by multiplying the photocurrent obtained by the synaptic-type bionic optoelectronic detector array by a fixed coefficient. is the conductance value of each memristor. Through Ohm's law and Kirchhoff's law , perform in-memory computing of a deep neural network operation, and finally output the current represents the task recognition result.

[0051] In an exemplary embodiment, it further includes: a driving circuit; the driving circuit is directly connected to each electrode of the synaptic-type bionic photodetector array for reading the photocurrent signal.

[0052] The driving circuit is directly connected to the top electrode of the brain-like memristor array for converting the photocurrent signal into a voltage signal and inputting the voltage signal into the brain-like memristor array.

[0053] The driving circuit includes a power supply module, a bias control module, a signal reading module, and a host computer, and is used to provide necessary power supply, readout, and current-to-voltage auxiliary functions.

[0054] This application realizes in-sensor computing and readout of the synaptic-type bionic photodetector array by constructing a co-design of sensitivity regulation and signal reading.

[0055] This application uses the new computing paradigm of multi-frame integration in-memory computing for moving target perception and recognition. By simulating the biological function of the human eye retina receptor through the synaptic-type bionic photodetector array, the calculation formula of the photocurrent signal is adopted and the recurrent neural network computing paradigm , perform non-linear iterative operations at the moving target detection end, retain all feature information of the entire motion history process in one feature map, and then convert the current into a voltage and directly input it into the brain-like memristor in-memory neural network array to complete the recognition task. The in-memory neural network based on memristor, that is, the brain-like memristor array, can realize the direct calculation and storage of neural network weight information, thus eliminating the energy consumption required for data exchange between the calculation and storage units. Through Ohm's law and Kirchhoff's law , perform in-memory fusion deep neural network operations on the brain-like memristor array, and efficiently complete the infrared image recognition task.

[0056] In another exemplary embodiment, inspired by the biological visual system, this application is based on a neuro-morphological photodetector, fuses the multi-frame image information of the moving target into one frame for readout as a feature map, inputs it into the hardware deep neural network with brain-like functions for recognition, and each pixel integrates a dual detector unit to achieve multi-feature extraction and wide dynamic range detection, and finally develops a bionic moving target in-sensor computing photodetection system with fast edge computing ability.

[0057] The retinoic-like photodetector unit in this application, hereinafter referred to as the device, adopts the MSM structure to achieve linear optical response.

[0058] As Figures 1 - 2 shown, the device and the array are based on the molybdenum disulfide (MoS2) light absorption layer 2 on the sapphire substrate 1. One end of each device leads out the positive electrode 3, and the other end leads out the negative electrode 4. When detecting, a fixed bias voltage is applied between the positive electrode 3 and the negative electrode 4, and the magnitude of the current of the negative electrode 4 is recorded.

[0059] The implementation method is as follows: grow about 3 - 5 atomic layers of MoS2 light absorption layer 2 on the insulating sapphire (aluminum oxide Al2O3) substrate 1 by chemical vapor deposition, perform patterning design on the MoS2 light absorption layer 2 through photolithography and etching technology, and then deposit metal electrodes on the surface by electron beam evaporation; finally, wash the mask, and the photogenerated carriers generated by the device under light illumination are led out through the positive electrode 3 and the negative electrode 4.

[0060] In this application, the MSM structure is a metal-semiconductor-metal structure.

[0061] The synaptic-like bionic photodetector array in this application consists of 16 retinoic-like photodetector units in a 4×4 configuration. The reading of the photocurrent signal of the detector is realized through the designed signal reading module.

[0062] In this application, the synaptic-like bionic photodetector array is provided with a drive circuit system. The drive circuit system includes a power supply module, a bias voltage control module, a signal reading module, and a host computer. The drive module mainly provides necessary power supply, reading, and auxiliary functions of current-to-voltage conversion, and is directly connected to each electrode of the detector array through the circuit. Through the collaborative design of sensitivity regulation and signal reading, the in-sensor calculation and reading of the detector array are realized.

[0063] In this application, a brain-inspired memristor array is used to build a hardware-integrated deep neural network through this cross array. Each memristor can adjust its resistance value through an external voltage and continuously maintain the current resistance value after removing the applied voltage. Therefore, it is used as the hardware carrier of the neural network, and the weights of the trained neural network are mapped to resistance values and written into the memristors through an applied voltage.

[0064] Each memristor unit adopts a 1T1R structure. The 1T1R structure refers to a structure combining a transistor and a memristor. Among them, "1T" represents a transistor, which is a field-effect transistor, and "1R" represents a memristor. In the 1T1R structure, when operating on a certain unit, the corresponding transistor is turned on, while the transistors corresponding to other units are turned off, thus avoiding the problem of misoperation on surrounding units and generating reading crosstalk.

[0065] As Figure 3 shown, the line connecting the gates of a row of transistors is called the word line (WL), the line connecting the top electrodes of a column of neuromorphic memristors is called the bit line (BL), and the line connecting the sources of a column of transistors is called the source line (SL). When in use, the gating of the WL is controlled by an analog switch to turn on the transistors in the specified row, the input signal is applied to the BL end, and the calculation result of the neuromorphic memristor array is read from the SL end. The WL refers to the line connecting the gates of each row of transistors and is used to control the switching state of the transistors. The SL refers to the line connecting the sources of each row of transistors and is used to provide a path for current. The BL refers to the line connecting the upper electrodes of each column of memristors and is used to read or write data. The MUX is a multiplexer circuit used to select a specific row or column, thereby enabling access to a specific memristor cell (or a group of cells) in the memristor array.

[0066] In an exemplary embodiment, the present application uses chemical vapor deposition (CVD) to grow the MoS2 photoabsorber layer material. On an Al2O3 substrate 1 with a thickness of 800 μm, a MoS2 photoabsorber layer with a thickness of 3 nm is grown. AZ5214 photoresist is evenly coated on the material surface for lithography. After exposure, an RZX3080 developer is used for development. Subsequently, 51 nm of Cr and 35 nm of Au are sequentially electron beam evaporated as metal electrodes to form the device and array patterns. Among them, the graphic structure is drawn using L-Edit software, as Figures 1 - 2 shown, the device is a rectangular interdigital structure, and the electrode width 5 and the electrode pitch 6 are 15 μm and 5 μm respectively. The MoS2 photoabsorber layer region between the interdigital electrodes is the light response region.

[0067] Acetone stripping, ethanol cleaning, and nitrogen drying are adopted; secondly, AZ5214 photoresist is evenly coated on the material surface again for lithography. After exposure, an RZX3080 developer is used for development to make the photoresist cover only one of the devices in each pixel. Immediately, the devices not covered by the photoresist are treated with oxygen plasma at a power of 50 W for 3 seconds; finally, acetone stripping, ethanol cleaning, and nitrogen drying are performed to complete the preparation of the synapse-like bionic photodetector array.

[0068] Figure 4 This is the IV characteristic curve graph of the retina-like photodetector unit of the present application. The IV curves are respectively under dark conditions and under different powers of 520 nm laser irradiation, which reflects the good light response characteristics of this device. Figure 5This is the continuous photoconductivity characteristic curve of the retina-like photodetector unit of this application. A fixed bias voltage of 0.2 V is applied to the positive and negative electrodes of the device, and three frames of optical pulse signals are continuously irradiated with a period of 2 s and a pulse width of 100 ms. It can be seen that after three consecutive cycles, the photocurrent decays to a final value, and the photocurrent value memorizes the historical information of all past moments, demonstrating good continuous photoconductivity characteristics.

[0069] The device prepared above is spot-welded onto a 4×4 array board, as Figure 6 shown, and then connected to the signal reading module through DuPont connecting wires in sequence.

[0070] This application uses technologies such as photolithography, etching, deposition, and sputtering to fabricate a brain-like memristor array device with a sandwich structure on a silicon substrate.

[0071] First, a top-gate structure field-effect transistor is fabricated on the silicon substrate as a gating device; subsequently, through photolithography and etching processes, a memristor bottom electrode region is etched out in the drain region of the field-effect transistor; using the thermal evaporation process, a metal bottom electrode of the memristor is evaporated in this region and peeled off under heating conditions to obtain the metal bottom electrode. Then, through sputtering and deposition processes, a metal oxide thin film is fabricated on the bottom electrode as the resistive switching layer of the memristor. The selected oxide materials are hafnium dioxide HfO2, tantalum oxide TaO x and other materials compatible with the (Complementary Metal-Oxide-Semiconductor, CMOS) process; using photolithography and etching processes, a memristor top electrode region is etched out on the resistive switching layer thin film; then, through the thermal evaporation process, a metal top electrode is evaporated in this region and peeled off under heating conditions to obtain the metal top electrode. Finally, a memristor cross array with a field-effect transistor as the gater, forming a 1T1R unit structure, that is, a brain-like memristor array, is completed, as Figure 3 shown.

[0072] This application also proposes an accurate modulation method based on dynamic pulses for weight adjustment of a brain-inspired memristor array, achieving high-precision read and write operations of memristors. By an external control signal, rows and columns in the brain-inspired memristor array are selected to precisely control each memristor. Meanwhile, combined with time-division multiplexing technology, the need for digital-to-analog conversion is further reduced, effectively shrinking the circuit area and significantly reducing the overall power consumption of the circuit. In the read operation of the brain-inspired memristor array, first, the analog switch is controlled by the WL terminal signal to select the target memristor, and an analog signal is output at the BL terminal through digital-to-analog conversion. At this time, a pulse lower than the threshold voltage (0.2V) is applied to the selected memristor. After the voltage pulse is applied to the memristor, based on Ohm's law, the calculation result is directly obtained, and the output current is read through the SL terminal, and then the resistance value of the current memristor is calculated to achieve the reading of the resistance value.

[0073] During the write operation, it is divided into two processes: setting and resetting. The setting process reduces the resistance value of the memristor through a positive voltage, while the resetting process increases its resistance value through a negative voltage. In the setting operation, first, the analog switch is controlled by the WL terminal to select the specified memristor, and a pulse higher than the write voltage threshold (1V) of the memristor is output at the BL terminal through digital-to-analog conversion. After each write pulse, a read pulse is immediately performed to monitor in real time whether the resistance value of the memristor reaches the set value. If it does not reach the preset value, the voltage of the write pulse is increased with a step value of 0.2V until the target resistance value is reached; if the resistance value of the memristor is already lower than the set value after the write pulse is applied, a negative voltage pulse is applied until the target value is reached. If the resistance value of the memristor is already lower than the target value, a positive voltage pulse is applied until the resistance value returns to the set value. Through the above method, this application realizes the accurate adjustment of the memristor weight, providing an effective technical means for constructing a hardware deep neural network with in-memory computing.

[0074] Deploy a hardware visual neural network using the prepared 4×4 pixel synaptic-type bionic photodetector array and 1T1R brain-inspired memristor array to achieve multi-frame integration and fast detection and recognition of dynamic targets.

[0075] A bionic moving target sensing and in-memory computing optoelectronic detection system based on persistent photoconductance is as Figure 7 shown. The moving target light source continuously irradiates the synaptic-type bionic photodetector array. Compared with the prior art, the new computing paradigm of multi-frame integration sensing and in-memory computing in this application does not read the photocurrent in each frame. By simulating the biological function of the human eye retina photoreceptor, using the photocurrent calculation formula and the recurrent neural network computing paradigm , while the photocurrent is continuously decaying, it is exposed to multiple illuminations. Nonlinear iterative operations are performed at the moving target detection end, and all the feature information of the entire motion history process is retained in a single feature map. Then, after converting the current into voltage, it is directly input into the brain-inspired memristor computing-in-memory neural network array to complete the recognition task. The computing-in-memory neural network based on memristors can directly calculate and store the weight information of the neural network. The trainable neural network weights are mapped to the conductance values of the memristors, as Figure 8 shown Figure 8 is the conductance value heat map of the memristor cross array of this application, thus eliminating the energy consumption required for data exchange between the computing and storage units. Through Ohm's law and Kirchhoff's law , the computing-in-memory deep neural network operations are performed on the brain-inspired memristor hardware platform, and the infrared image recognition task is efficiently completed.

[0076] Figure 9 is the recognition accuracy curve graph for the recognition task scenario of dynamic targets in this application. Compared with traditional networks, this application can complete tasks that cannot be completed by traditional single-layer fully connected neural network technologies based on single-frame image information.

[0077] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as within the scope described in this specification.

[0078] Specific examples are used in this article to elaborate on the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, there will be changes in the specific implementation methods and application scopes according to the idea of this application. In summary, the content of this specification should not be construed as a limitation on this application.

Claims

1. A bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity, characterized in that: The photoelectric detection system includes: an interconnected synaptic-like bionic photoelectric detector array and a brain-like memristor array; When the synaptic-like bionic photodetector array is used to detect visual information of a moving target, a fixed bias voltage is applied to the device in each pixel within the synaptic-like bionic photodetector array, and the spatial illumination information corresponding to each pixel is converted into a continuously decaying photocurrent signal. During the process of continuous photocurrent decay, multiple frames of illumination information of the dynamic target are continuously received, and after the last frame of illumination information is input, the photocurrent signal of each pixel is read to determine a feature map that memorizes information at all historical moments. The photocurrent signal exhibits nonlinear continuous decay between frames. When the synaptic-like bionic photodetector array performs time-domain multi-frame fusion detection and reading, it utilizes the continuous photoconductivity effect of the detector, and the photocurrent decays nonlinearly over time, while simultaneously completing linear and nonlinear calculations of a single-layer neural network. The brain-like memristor array is used to perform deep neural network operations on the feature map based on Ohm's law and Kirchhoff's law, output a current corresponding to the task result label, and identify dynamic targets based on the current; the current is used to characterize the probability of the task result label.

2. The bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity according to claim 1 is characterized in that: The synapse-like bionic photoelectric detector array specifically includes: a plurality of pixels distributed in an array, each pixel including a retina-like photoelectric detector unit.

3. The bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity according to claim 2 is characterized in that: Each of the retinal-like photodetector units includes a substrate, a light absorption layer, and two crossed interdigital electrodes stacked in sequence from bottom to top; wherein, the light absorption layer area corresponding to the channel formed between the two crossed interdigital electrodes is a light response area; the light response area is used to provide linear light response and nonlinear attenuation of photocurrent, and regulate the voltage of the electrodes at both ends of each retinal-like photodetector unit to encode the light response of the retinal-like photodetector unit.

4. The bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity according to claim 3 is characterized in that: The substrate is an insulating Al2O3 substrate; the light absorption layer is a MoS2 two-dimensional material; Performing image design on the MoS2 two-dimensional material by a photolithography process, and depositing a metal electrode on the surface of the MoS2 two-dimensional material by electron beam evaporation; The metal electrodes are the interdigitated electrodes.

5. The bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity according to claim 3 is characterized in that: The interdigitated electrodes are rectangular cross-finger structures; the two interdigitated electrodes include a positive electrode and a negative electrode; the positive electrode and the negative electrode are arranged in a staggered manner.

6. The bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity according to claim 2 is characterized in that: The photoresponsivity of the retina-like photodetector unit is a random fixed value; The photocurrent signal is the product of the photoresponsivity and the light intensity.

7. The bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity according to claim 1 is characterized in that: The brain-like memristor array utilizes word line circuits, source line circuits, and bit line circuits to form a mesh layout; a memristor unit is arranged at each intersection in the mesh layout; The memristor unit includes a transistor and a memristor; The word line circuit is connected to the gate of the transistor, the source line circuit is connected to the top electrode of the memristor, and the bit line circuit is connected to the bit of the memristor; The bottom electrode of the memristor is also directly connected to the drain of the transistor; the conductive state of the brain-like memristor array is controlled by adjusting the gate voltage of the transistor; the conductive state includes an on state and an off state.

8. The bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity according to claim 7 is characterized in that: The top layer of the memristor array is a top electrode layer, the middle layer is a resistive switching layer, and the bottom layer is a bottom electrode layer; The top electrode layer is input with a voltage through a row gate, and the bottom electrode layer is output with a current through a column gate; The memristor array follows Ohm's law and Kirchhoff's law to complete multiplication and accumulation operations, reads current from each column, completes vector-matrix multiplication operations in the deep neural network, and realizes a hardware deep neural network with integrated storage and computing.

9. The bionic target sensing, storage and computing integrated photoelectric detection system based on continuous photoconductivity according to claim 1 is characterized in that: Also includes: Drive circuit; The driving circuit is directly connected to each electrode of the synaptic-type bionic photodetector array to read the photocurrent signal; The driving circuit is directly connected to the top electrode of the brain-like memristor array, and is used to convert the photocurrent signal into a voltage signal, and input the voltage signal into the brain-like memristor array.

Images