A dynamic identification and prediction method based on photoconductive self-modulation

Abstract

The present invention discloses a dynamic recognition and prediction method based on photoconductive self-modulation, belonging to the field of intelligent recognition technology. The present invention introduces selenium vacancy defects in high-mobility layered two-dimensional semiconductor bismuth oxygen selenium to separate it from the conductive channel bismuth oxygen layer, thereby realizing a self-modulating photodetector for self-modulating channel electron transmission of ionic defects; the device array composed of the self-modulating photodetector is used as the front-end signal processing hardware to continuously sample and record the constantly changing external image information, while the back-end dynamic recognition and prediction are performed through the neural network at the software level. The present invention uses a sensing, storage and computing integrated structure to complete the preprocessing of the signal at the information input end, greatly reducing the power consumption overhead of the hardware neural network; combined with the subsequent network algorithm, small-area, low-cost hardware-level mobile recognition and prediction are realized.

Description

Technical Field

[0001] The present invention relates to a method for detecting and predicting a moving target, and in particular to a method for dynamic recognition and prediction based on self-modulated photoconductance, belonging to the field of sensing, storage and computing integration and intelligent recognition technology. Background Art

[0002] The purpose of motion detection and recognition tasks is to detect the movement of objects with time correlation and to infer and evaluate their future states. At present, the realization of motion detection and recognition tasks mainly relies on CMOS image sensor imaging technology, combining image sensing, signal conversion, storage and processing of multiple circuit component modules and network algorithms, and performing differential operations on adjacent frame image processing based on background subtraction, frame difference method, optical flow method, etc. to estimate the changes in pixels and then estimate the movement of objects. To achieve accurate and efficient dynamic recognition and prediction, it is necessary to comprehensively consider technical issues such as algorithm selection, parameter setting, real-time performance, and data storage and processing. In contrast, the sensing, storage and computing device array with relaxation characteristics regulates the current relaxation of the device through gate voltage. After image capture, it combines Ohm's law and Kirchhoff's law with the input external light power to perform in-situ cumulative multiplication and addition operations, and transmits the motion image frames of the same object at multiple times to the algorithm backend after correlation fusion, which greatly reduces redundant data transmission, reduces power consumption and improves computing efficiency. In order to identify objects with different moving speeds on multiple time scales, a wide range of adjustable relaxation speeds is required. However, existing research on devices based on relaxation dynamics often has a small range of time constant control, which greatly limits the development of existing sensing, storage and computing device systems in the field of dynamic recognition and prediction. Therefore, there is an urgent need for an electronic device unit with a wide photoconductive relaxation speed. Summary of the invention

[0003] In order to realize an electronic device unit with a wide photoconductivity relaxation rate, the present invention proposes a photodetection device based on the high-mobility bismuth oxygen selenium self-doping effect, which realizes a continuously dynamically changing photoresponse (the photoresponse variation range is 10 A / W ~ 10 5 A / W, relaxation time is 10 ms ~ 10 5 ms, and the corresponding light weight accuracy reaches 93.2%). The self-modulating photodetector of the present invention can effectively adjust the relaxation speed and be used for the recognition, detection and behavior prediction of continuously moving objects.

[0004] The present invention discloses a self-modulating photodetector based on the high-mobility bismuth oxygen selenium self-doping effect, comprising a substrate and a gate dielectric layer, and a layered two-dimensional bismuth oxygen selenium channel layer located on the gate dielectric layer, wherein both ends of the channel layer are source-drain (S / D) electrodes, wherein the lower portion of the two-dimensional bismuth oxygen selenium channel layer is an interface defect layer, and the upper portion is a channel transmission layer.

[0005] Preferably, the interface defect layer is 1-2 layers of two-dimensional bismuth oxide selenium with selenium vacancies, and its thickness is 0.6 nm-1.4 nm. The selenium vacancy defect concentration is 4.2×10 12 / cm 2 ~1.6×10 13 / cm 2 .

[0006] The thickness of the channel transmission layer is preferably 8-12 nm.

[0007] The substrate may be a silicon substrate, which is a back gate. The gate dielectric layer may be made of dielectric materials such as hafnium oxide, aluminum oxide, silicon oxide, etc. The source and drain electrodes may be metal electrodes such as titanium gold, palladium gold, titanium palladium gold, etc.

[0008] In some embodiments of the present invention, the substrate is a silicon substrate; the gate dielectric layer is made of hafnium oxide (HfO2) with a thickness in the range of 10-20 nm; the source and drain electrodes are Ti / Au metal stacks with a thickness not exceeding 60 nm; the thickness of the two-dimensional bismuth oxygen selenium channel layer is in the range of 5-20 nm, and the interface defect layer can induce selenium vacancies through interface modification engineering caused by wet etching, plasma bombardment, element doping or other methods.

[0009] The above self-modulating photodetector has gate voltage dependence: when the gate voltage is 0, the interaction between channel electrons and defects reaches equilibrium, and the photocurrent does not change with time; under negative gate voltage, the Fermi level decreases, defect electrons are released into the conductive channel, the defect sites have local positive charge, and channel electron transmission is promoted by self-modulating doping, and the photocurrent increases with time; under positive gate voltage, the Fermi level increases, electrons fill defects, and defect sites have local negative charge, and channel electron transmission is suppressed by self-modulating doping, and the photocurrent decreases with time. Based on the above characteristics, bidirectional photocurrent polarity with adjustable gate voltage can be achieved, and the gate voltage-dependent photocurrent that changes with time of the self-modulating photodetector corresponds to the information enhancement and weakening of the image frame to be identified. The output current value is used to measure the amount of effective information it contains. When a positive gate voltage is applied, the photocurrent gradually decays over time, and the photocurrent response of the photodetector caused by external information gradually weakens, that is, the detected image frame information weakens; conversely, when a negative gate voltage is applied, the photocurrent continues to rise over time, and the photocurrent of the photodetector caused by external information gradually increases, that is, the detected image frame information increases.

[0010] Based on the self-modulating photodetector of the present invention, the present invention provides a dynamic identification and prediction method. Based on interface defects, after the self-modulating photodetector applies and removes light, the channel photocurrent exhibits relaxation characteristics, and the light response time constant τ is regulated by the gate voltage. For external information, when light information enters, it is equivalent to applying a light stimulus to the channel. Due to the memory behavior of the device, if external information is applied to the device at two adjacent time nodes, the device will memorize and store the image information at the two time nodes. For fast-moving objects, the device is required to have a fast relaxation speed, corresponding to a small light response time constant in order to distinguish two images in adjacent time states; for slow-moving objects, the device is required to have a slow relaxation speed in order to match the external information well, corresponding to a large light response time constant, so as to store and display related images within a wide time window. This requires a reasonable design of the light response time constant of the device. The present invention combines a neural network and uses the images stored in the device as training data, thereby realizing the detection and identification of objects with different moving speeds. The specific process is: first, for the device array composed of the self-modulated photodetector, a fixed gate voltage bias is provided for each device so that it has relaxation characteristics with different light response time constants; then, at different times t, the image information of the external moving object is irradiated onto the device array in the form of light projection. Due to the aforementioned device-based photocurrent relaxation characteristics, after the light information at the next moment is input, the channel current at this time is equal to the algebraic sum of the residual photocurrent information of the relaxation at the previous moment and the new input at this time. Therefore, the current at the newly irradiated position in the image will reach a peak value, and the image information in the previous movement process will gradually be lost. The movement direction of the object can be judged based on the array's memory of the image information; in addition, when the object moves, the output of each device in the array is read out to obtain a two-dimensional distributed photocurrent image, which is used as the target of neural network training and recognition at the software level. First, we use software to train a neural network consisting of three convolutional layers and three fully connected layers. The object's movement speed is divided into three levels: fast, medium, and slow for training. The external information first undergoes three layers of convolution operations to complete information dimensionality reduction and feature extraction, and then is input into the three fully connected layers for output and recognition. After the network training is completed, the untrained samples are input into the trained neural network for recognition and prediction, which can effectively realize the estimation of objects with different movement speeds, and based on this design, the position of the object at a certain moment in the future can be predicted. The specific implementation method is: according to the object's movement speed evaluated by the network at the previous moment, the displacement at the future time t is equal to the product of the speed and the time t, thereby realizing the prediction and estimation of the position.

[0011] Specifically, the present invention provides a dynamic identification and prediction method based on photoconductivity self-modulation based on the self-modulation photodetector, comprising:

[0012] 1) The self-modulating photodetector array is used as the front-end signal processing structure at the hardware level. By continuously sampling the external image information that is constantly changing, based on the relaxation characteristics of the device structure, the image information collected first will degrade over time, and the feature retention of the image information collected later will be relatively high. The information with obvious light-dark contrast contains the timing information of different pixels. As the standard for judging the moment, the external image information is collected and recorded by adopting the array structure. In this way, the image information constructed by the output current obtained contains the timing information, and the detection of the moving direction can be realized by the collaborative work of software and hardware. The photodetector array at the hardware level plays the role of memorizing the time series information and preprocessing the signal. By reading the current output of each device in the array to form a two-dimensional current distribution map, these two-dimensional current distribution maps are then trained in combination with the software-level algorithm, so as to estimate the speed of the object and predict the future position based on the time difference;

[0013] 2) The software neural network is used to perform the back-end mobile orientation recognition, speed recognition and future position prediction. The software-level neural network consists of three convolutional layers and three fully connected layers. The three convolutional layers are mainly used to extract features and reduce the dimension of the collected image information. After each convolution process, the dimension of the image will be reduced, which is conducive to the subsequent processing and calculation; the three-layer fully connected output layer is used as a forward neural network to perform weighted summation on the convolution data for recognition and output. At the beginning of training, the weights of all nodes in the six-layer neural network in the network are assigned initial weights. The output value is decided by the image recognition algorithm at the software level. If the output value of the network is consistent with the label value, the next image is trained. If there is a deviation between the target value and the training value, the error of each weight is corrected by the back propagation algorithm. The training images are trained one by one, thereby completing the software training at the neural network level. After the network is trained, the previously untrained data is input into the trained neural network in the form of images, and all processes of convolution dimensionality reduction, feature extraction and back-end output are completed in sequence to complete the recognition of the target data.

[0014] Compared with the prior art, the technical advantages of the present invention are as follows:

[0015] 1. For the first time, selenium vacancy defects are introduced into the high-mobility and high-stability ionic layered two-dimensional semiconductor bismuth oxygen selenium to separate it from the conductive channel bismuth oxygen layer. The ionic defects self-modulate the channel electron transport, realizing the self-modulation effect at the level of a single device, and achieving a wide range of photocurrent amplitude modulation and photocurrent change time constant modulation.

[0016] 2. The time-dependent increase and decrease of the photocurrent that depends on the gate voltage corresponds to the enhancement and weakening of the captured image under continuous light stimulation conditions. The integrated sensing, storage and computing structure implemented at the hardware level completes the preprocessing of the signal at the information input end. The average power consumption for operating a single pixel is as low as 1 pW, which greatly reduces the power consumption overhead of the hardware neural network.

[0017] 3. By utilizing the photocurrent relaxation characteristics of self-modulating photodetectors, the time-space fusion of optical information on a wide time scale can be effectively achieved when the external optical signal is continuously input. The gate voltage-dependent wide-range adjustable time constant is applied to the detection of objects with different moving speeds. Combined with subsequent network algorithms, small-area, low-cost hardware-level motion recognition and prediction are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a structural schematic diagram of the self-modulating photoelectric detector of the present invention;

[0019] Figure 2~Figure 7 The following is a schematic diagram of the process steps for preparing a self-modulating photodetector according to an embodiment of the present invention, wherein:

[0020] Figure 2 The method comprises the steps of growing bismuth oxide selenide material on a mica sheet;

[0021] Figure 3 The step of spin coating and etching a PMMA protective layer on the surface of the bismuth oxide selenide material using a PMMA assisted transfer method;

[0022] Figure 4 The step of immersing the bismuth oxide selenium material coated with a PMMA protective layer into a dilute acid solution to introduce interfacial selenium vacancies;

[0023] Figure 5 The step of transferring bismuth oxide selenide / PMMA onto Si / HfO2 substrate;

[0024] Figure 6 A step for removing the PMMA protective layer;

[0025] Figure 7 The step of defining the source-drain electrode region and preparing the source-drain electrode by electron beam exposure;

[0026] Figures 1 to 7Middle: 1—gate dielectric layer; 2—substrate; 3—source electrode; 4—drain electrode; 5—two-dimensional bismuth oxygen selenide channel layer; 6—interface defect layer; 7—mica sheet; 8—PMMA protective layer.

[0027] Figure 8 3 is a schematic diagram of the photocurrent relaxation characteristics of the self-modulated photodetector in an embodiment of the present invention under a gate voltage bias of -2V to 2V, wherein a and b are schematic diagrams of the photocurrent obtained by applying continuous light to the device in the embodiment in the logarithmic and normalized linear coordinate systems, respectively.

[0028] Fig. 9 3 is a curve showing the change of the modulation coefficient of the self-modulating photodetector with different two-dimensional bismuth oxygen selenide channel layer thicknesses under a gate voltage of -2 V to 2 V in an embodiment of the present invention.

[0029] Fig.10 3 is a relaxation characteristic curve of the photocurrent over time of the self-modulating photodetector in an embodiment of the present invention under a gate voltage bias of -1.5 V to 1.5 V and a source-drain voltage bias of 0 to 0.2 V.

[0030] Fig.11 is the photoconductivity relaxation time constant τ of the self-modulating photodetector according to the embodiment of the present invention as a function of the gate voltage V g Among them, Figure a is the curve of the normalized channel current changing with time under different gate voltage bias conditions, and Figure b is the relaxation time constant τ extracted from Figure a as the gate voltage V g The change curve.

[0031] Fig.12 is the photoconductivity relaxation time constant τ of the self-modulating photodetector according to the embodiment of the present invention as a function of the source-drain voltage V d Among them, Figure a is the curve of the normalized channel current changing with time under different source-drain voltage bias conditions, and Figure b is the relaxation time constant τ extracted from Figure a as the source-drain voltage V d The change curve.

[0032] Fig.13 is the photoconductivity relaxation time constant τ of the self-modulating photodetector according to the embodiment of the present invention as a function of the gate voltage V g And the source-drain voltage V d Schematic diagram of the changes.

[0033] Fig.14 Schematic diagram of the working principle of the self-modulating photodetector of the present invention.

[0034] Fig.15 Schematic diagram of a device-algorithm flow for using the device's self-modulated light responsivity for motion identification and detection in an embodiment of the present invention.

[0035] Fig.16 The figure shows the processing results of the sensor array composed of self-modulating photoelectric detection devices in the embodiment of the present invention on the input images of puppies with different moving speeds.

[0036] Fig.17 This is the result of state estimation and restoration of the sensor array composed of the self-modulated doped bismuth oxygen selenium photoelectric detection devices in the embodiment of the present invention on the obtained sensor image.

[0037] Fig.18 It is a schematic diagram of a scenario for predicting the results of a real scenario in which a puppy with different moving speeds chases a rabbit in an embodiment of the present invention.

[0038] Fig.19 This is the result of predicting the result of a real scene in which a puppy with different moving speeds chases a rabbit in an embodiment of the present invention. DETAILED DESCRIPTION

[0039] To make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below through embodiments in conjunction with the accompanying drawings.

[0040] like Figure 1 As shown, the self-modulating photodetector of the present invention comprises a substrate 2 and a gate dielectric layer 1, a two-dimensional bismuth oxygen selenium channel layer 5 located on the gate dielectric layer 1, the lower part of the two-dimensional bismuth oxygen selenium channel layer 5 is an interface defect layer 6, and the upper part is a channel transmission layer; the two ends of the two-dimensional bismuth oxygen selenium channel layer 5 are a source electrode 3 and a drain electrode 4. Based on the self-modulating photodetector array, an on-chip sensing, storage and computing integrated structure is formed, and the dynamic change of light weights of different pixel points is realized by adjusting the voltage bias of a single device, which is applied to the on-chip dynamic neural network accelerator, further improving the error tolerance and environmental adaptability of the dynamic neural network.

[0041] The self-modulating photodetector is a bismuth-oxygen-selenium light-sensing and computing integrated device based on self-doping modulation of interface selenium vacancies, and its preparation steps are as follows:

[0042] 1) Using Bi2O3 and Bi2Se3 as growth sources, a layered two-dimensional bismuth oxide selenium material is grown on a mica sheet 7, such as Figure 2 As shown;

[0043] 2) Spin-coat a PMMA (polymethyl methacrylate) protective layer 8 on the surface of the bismuth oxide selenium material on the mica sheet 7. The spin-coating process conditions in this embodiment are: spin-coating at a speed of 4000 rpm for 1 min, and then curing on a heating plate at 170° C. for 3 min. Figure 3 As shown;

[0044] 3) The bismuth oxide selenium material protected by the PMMA protective layer 8 is etched at the interface of the mica sheet 7 by using HF:H2O = 1:50 (volume ratio), and the interface defect layer 6 is regulated by adjusting the etching time to control the concentration of the interface selenium vacancy defects. In this embodiment, the etching time is 4 to 12 h. Figure 4 As shown;

[0045] 4) After washing with deionized water, the bismuth oxide selenium / PMMA was transferred to the Si / HfO2 composite substrate and then cured at 80°C for 3 min. Figure 5 As shown;

[0046] 5) The bismuth oxide selenium / PMMA transferred to the Si / HfO2 substrate was immersed in an acetone solution for 30 min, and the protective layer PMMA 8 was removed. Figure 6 As shown;

[0047] 6) For the two-dimensional bismuth oxide selenide channel region (the thickness of the two-dimensional bismuth oxide selenide material in this embodiment is in the range of 5-20 nm), the source and drain (S / D) regions are defined by electron beam exposure (EBL), and the metal is evaporated by electron beam (the metal used in this embodiment is Ti / Au=5 / 40nm) and peeled off to form source and drain electrodes, such as Figure 7 shown.

[0048] The schematic diagram of the self-modulated photoconductivity of the device in this embodiment is shown in FIG. Figure 8 As shown in the figure, under negative gate voltage, the photocurrent obtained by applying continuous illumination increases with time, showing an increasing response; under positive gate voltage, the photocurrent amplitude obtained by applying continuous illumination decreases with time, showing a decreasing response. The wavelength of the applied illumination is 640 nm and the optical power is 67 μW / cm 2 , but this phenomenon is also applicable to the light power in the full spectrum range of 400-800 nm wavelength; the normalized photocurrent variation curve over time is as follows Figure 8 As shown in b.

[0049] In order to study the effect of the thickness of the two-dimensional bismuth oxygen selenide channel layer on the photoconductive self-modulation behavior of the device, a two-dimensional bismuth oxygen selenide channel layer with a thickness range of 5~15 nm was prepared. The corresponding device photoresponses have the same law. We define the modulation coefficient a =log(I final / I initial ), where I final and I initial Represent the final value and initial value of the photocurrent respectively. g <0, a>0, indicating that the photocurrent shows an enhancement effect over time; when V g >0, a<0, indicating that the photocurrent shows an inhibitory effect over time. The wavelength of the applied light is 640 nm and the light power is 67 μW / cm2 ,like Fig. 9 shown.

[0050] Combination Figure 10-12 As shown in Figure 2, we systematically studied the control methods of the photoconductivity modulation effect of the self-modulating photodetector. For the photoconductivity control behavior of the self-modulating photodetector with a typical 10 nm thick channel material, the applied light wavelength was 640 nm and the light power was 67 μW / cm 2 The polarity of the gate voltage can effectively control the trend of photoconductivity increase and decrease, and the source-drain voltage can effectively control the starting value of the photocurrent, such as Fig.10 As shown. The time constant τ of photoconductivity relaxation can be effectively controlled by the gate voltage and has a weak correlation with the source-drain voltage, such as Fig.11 and Fig.12 shown.

[0051] Electrical measurements were performed using a probe station and a semiconductor parameter analyzer 4200. A light source, a monochromator, and other devices were used to apply a 640 nm wavelength and 67 μW / cm 2 The thickness of the two-dimensional channel material of the device under study was 10 nm, and the obtained photocurrent variation curve with time was exponentially fitted, and the relaxation time constant τ was extracted to change from 10 ms to 10 8 ms, such as Fig.13 shown.

[0052] The working principle of the self-modulating photodetector of the present invention is as follows Fig.14 As shown, in the initial state, the Fermi level (E F ) between the Se vacancy defect energy level (E Se ) and the bottom of the conduction band, the selenium vacancies are partially filled, and the channel electrons and the defect interactions are balanced (e.g. Fig.14 As shown in a and d in the figure); Under positive gate voltage, the Fermi level jumps above the selenium vacancy defect level (as shown in Fig.14 As shown in (b), the electrons cross [BiO] 2+ Layer entry [Se] 2- layer and is captured by defects (e.g. Fig.14 (shown in c), [Se] 2- The local negative charge of the layer inhibits the electron transport in the channel, and the photocurrent decreases with time; under negative gate voltage, the Fermi level decreases to the band gap (such as Fig.14 As shown in e in the figure, the electrons captured by the defect level are released (as shown in Fig.14 As shown in (f), the Se vacancy acts as a local positive center, promoting the channel electron transport, and the photocurrent increases with time, thereby achieving time-varying photoconductivity self-modulation based on the self-modulation effect.

[0053] In order to study the preprocessing results of the photodetector with self-modulated photoconductivity effect on moving targets, we constructed a 3×3 self-modulated photodetector array to act as a convolution kernel, and applied different gate voltages and source-drain voltage biases to each device in the array to achieve a preprocessing array with different relaxation characteristics. The images of the puppy at different times during the movement are projected onto the device array, and the images obtained based on the relaxation characteristics of the device are represented by current values ​​and entered into the subsequent software algorithm for recognition. The specific implementation process is as follows: Fig.15 shown. Fig.15 The drain of each device in the device array shown is connected and controlled separately, the gate bias of each device is controlled independently, and the sources of the devices on each column are connected. During the test, a single device is selected by controlling the voltage input of the source and drain to read its current value, and the output current value of each device in the array is measured by fast cycle scanning, so as to obtain a two-dimensional image of the current output of the array device for external image detection. The photocurrent output corresponding to the device in the mth row and nth column in the array is:

[0054]

[0055] Among them, I mn (t) is the current output by the device in the mth row and nth column at time t, P mn (t) is the optical power input to the device in the mth row and nth column at time t; R mn (t) is the photoresponsivity of the device in the mth row and nth column at time t, and its fitting equation for changes with time is:

[0056]

[0057] Among them, R mn,0 , R mn,1 is the fitting parameter, and its value is achieved by adjusting the source-drain voltage and the gate bias; τ mn is the light response time constant of the device in the nth column and the mth row, and its value is achieved by adjusting the gate bias voltage.

[0058] The continuously changing animal image at the front end is used as the input of optical information, and is irradiated onto the channel of the photoelectric conversion device structure in the form of a light lens. Due to the different grayscale values ​​and colors of each pixel, the channel photocurrent response is also different, and the device has a relationship of continuous degradation over time for the recorded photocurrent. Based on this relaxation behavior, we can effectively record information about time. After that, we read the output current value of each device one by one and input it into the back-end neural network as a feature data. The constructed neural network structure is a three-layer convolution layer and a three-layer fully connected layer. These images are trained based on the algorithm structure encapsulated in python. Since the photocurrent information recorded by the device array contains different timestamps, the speed of the object can be estimated by the ratio of the relative distance of the image movement to the timestamp, and based on the statistics of the object speed, the position of the object at a certain moment in the future can be predicted and estimated.

[0059] Objects with different moving speed characteristics correspond to an optimal relaxation characteristic curve, so as to achieve the matching of moving speed and device function. We bias the device at V g =1 V, V d = 0.5 V, the relaxation time constant of the device is 2s. The running puppy at three moments is projected onto the device array, and the fused information has different characteristics at three moments. This is different from traditional photoelectric detection devices. Based on the long-term relaxation characteristics, we have achieved the fusion of time and space information. At the same time, we can also restore the moving state based on the pictures obtained by the self-modulated photoconductive device array, such as Fig.16 , Fig.17 shown.

[0060] In order to further demonstrate the application of the self-modulated photoelectric detection device in real-life scenarios, we designed a scenario in which a puppy chases a rabbit at different moving speeds. Within 3 seconds, the self-modulated photoelectric detection device based on this embodiment can identify the moving speeds of the rabbit and the puppy, and according to the distance d between the rabbit and the dog, dr and the relative distance d between the rabbits and their nests rh Predict the results of the pursuit in the subsequent time. When the speed of the puppy is five times that of the rabbit, the larger d rh and a smaller d dr It will increase the risk of rabbits being caught. Schematic diagram of the scene and the predicted results are as follows Fig.18 , Fig.19 shown. Fig.18 This is a schematic diagram of the scenario of the dog and rabbit chasing each other. First, at the beginning, the rabbit and the dog are separated by a certain distance d. drWe use a device array with photocurrent relaxation behavior to observe the image information from -2 to 0, and estimate the movement speed of the puppy and the rabbit, respectively, and obtain the movement speed of the puppy V dog and the rabbit's speed V rabbit , at this time the rabbit and the puppy continue to move, and the distance between the rabbit and the rabbit nest is d rh Therefore, we can estimate the time required for the rabbit to return to the rabbit nest by using the distance between the rabbit nest and the starting rabbit. Combined with the movement speed of the puppy, we can determine the movement distance of the puppy at this time, and thus determine whether the rabbit can be caught by the puppy before reaching the rabbit nest. If the movement distance of the puppy is greater than d dr With d rh , then the rabbit will be caught. Fig.19 Based on this algorithm, the following is described: when the speed of the puppy is 5 times the speed of the rabbit, for different starting distances d rh and dr Predicted capture behavior outcomes.

[0061] The embodiments described above are not intended to limit the present invention. Any person skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention is defined by the scope of the claims.

Claims

1. A dynamic identification and prediction method based on photoconductive self-modulation, characterized in that: The device array composed of self-modulating photodetectors is used as the front-end signal processing hardware to continuously sample and record the constantly changing external image information, and the back-end dynamic recognition and prediction are performed through the neural network at the software level; wherein the self-modulating photodetector includes a substrate and a gate dielectric layer, and a layered two-dimensional bismuth oxygen selenium channel layer located on the gate dielectric layer, and the two ends of the channel layer are source and drain electrodes; the lower part of the layered two-dimensional bismuth oxygen selenium channel layer is an interface defect layer with selenium vacancies, and the upper part is a channel transmission layer; the interface defect layer of the self-modulating photodetector is 1~2 layers of two-dimensional bismuth oxygen selenium with selenium vacancies, wherein the selenium vacancy defect concentration is 4.2×10 12 / cm 2 ~1.6×10 13 / cm 2 ; The thickness of the channel transmission layer is 8 nm ~12 nm.

2. The method for dynamic identification and prediction based on photoconductive self-modulation according to claim 1, characterized in that: The substrate is a silicon substrate, and the gate dielectric layer is hafnium oxide, aluminum oxide or silicon oxide.

3. The method for dynamic identification and prediction based on photoconductive self-modulation according to claim 1, characterized in that: For the device array composed of the self-modulating photodetectors, the gate voltage bias of each device is independently controlled, and the light response time constant of the device is adjusted by regulating the gate voltage. A small light response time constant is used to identify fast-moving objects, while a large light response time constant is used to identify slow-moving objects.

4. The method for dynamic identification and prediction based on photoconductive self-modulation according to claim 3, characterized in that: include: First, a fixed gate voltage bias is provided for each device in the device array so that it has relaxation characteristics with different light response time constants τ; then, at different times, the image information of the external moving object is irradiated onto the device array in the form of light projection, and the photocurrent output by each device in the array is read out to obtain a two-dimensional distributed photocurrent image, which is used as the target of neural network training and recognition; the neural network includes three convolutional layers and three fully connected layers, and the movement speed of the object is divided into three levels of fast, medium and slow for training, wherein the three convolutional layers perform feature extraction and dimensionality reduction operations on the collected image information, and then input it into the three fully connected layers to realize output and recognition; after the network training is completed, the image information of the moving object is continuously collected through the device array, and input into the trained neural network for recognition and prediction, so as to estimate the speed of the moving object and predict the future position of the object based on the time difference.

Images