A method and system for predicting short-term data based on an array of spiking neurons

Abstract

The application discloses a method and system for predicting short-term data based on a pulse neuron array. A group of sensing signals are input to the pulse neuron array, so that the pulse neuron array outputs a group of spike current signals. In each pulse neuron unit, a phototransistor senses the sensing signals input to the pulse neuron unit and outputs a source-drain current signal. A volatile memristor of the pulse neuron unit performs frequency domain spike coding on the source-drain current signal and outputs a spike current signal. According to the group of spike current signals, a group of first pulse times are obtained. According to the group of first pulse times, a short-term data prediction result is obtained by using a space-time transformation equation. The short-term data prediction result is spatial information containing future time information. The scheme can realize short-term data prediction without a large amount of data training, and the pulse neuron array can complete sensing and frequency domain spike coding on the sensing signals at a very high speed and a very low power consumption.

Description

Technical Field

[0001] This invention belongs to the cutting-edge interdisciplinary field of semiconductors and information processing, and specifically relates to a method and system for predicting short-term data based on a spiking neuron array. Background Technology

[0002] In recent years, with the emergence of artificial intelligence applications in autonomous vehicles, intelligent robots, and intelligent transportation, artificial vision systems have made significant progress. The implementation of these technologies places high demands on object localization and tracking at the edges of sensing units, often requiring real-time or even forward localization. However, due to the complex nonlinearities and uncertainties inherent in artificial vision nonlinear dynamic systems, multi-step forward prediction based solely on short-term time series is a challenging and difficult task to achieve.

[0003] Since artificial vision nonlinear dynamic systems cannot perform advance localization based on short-term time series, another possible scheme for advance localization is proposed, namely LSTM (Long Short-Term Memory Recurrent Neural Network).

[0004] While current LSTMs can learn future nonlinear dynamics from large amounts of observational data, they often fall into overfitting problems due to insufficient training data. Furthermore, training a recurrent neural network requires not only a large number of samples and time but also consumes significant computational resources, hindering the further application of artificial vision systems in efficient edge computing.

[0005] Furthermore, the upper limit of the speed of traditional digital processors is largely determined by the clock frequency. The increase in clock frequency is limited by the speed of logic state transitions, and further increases in processor speed will lead to serious overheating problems. This has caused the clock frequency of advanced digital computers to stop increasing for more than a decade.

[0006] Therefore, designing a short-term data prediction method that can reconstruct recent evolution information based on short-term object movement trajectories in the continuous time domain without requiring training with a large amount of observation data is crucial for achieving efficient edge-aware computing.

[0007] How to achieve short-term data prediction without training with a large amount of observational data is a technical problem that urgently needs to be solved. Summary of the Invention

[0008] To address the aforementioned problems in the prior art, this invention provides a method and system for predicting short-term data based on a spiking neuron array.

[0009] The technical problem to be solved by this invention is achieved through the following technical solution:

[0010] In a first aspect, the present invention provides a method for predicting short-term data based on a spiking neuron array, wherein the spiking neuron array comprises: a plurality of spiking neuron units; each spiking neuron unit comprises a phototransistor and a volatile memristor; the phototransistor and the volatile memristor in each spiking neuron unit are connected in series;

[0011] The method includes:

[0012] A set of sensing signals is input to the spiking neuron array so that the spiking neuron array outputs a set of spike current signals accordingly; wherein, the phototransistor in each spiking neuron unit senses the sensing signals input to the spiking neuron unit and outputs a source-drain current signal, and the volatile memristor of the spiking neuron unit performs frequency domain spike encoding on the source-drain current signal and outputs a spike current signal.

[0013] Based on the set of peak current signals, a set of first pulse times is obtained;

[0014] Based on the set of first pulse times, short-term data prediction results are obtained using spatiotemporal transformation equations; the short-term data prediction results are spatial information containing future time information.

[0015] Optionally, obtaining short-term data prediction results using a spatiotemporal transformation equation based on the set of first pulse times includes:

[0016] Based on the time delay theory, the time of the first pulse is encoded into future time information through the time transformation matrix in the spatiotemporal transformation equation, and the future time information is decoded into spatial information containing the future time information through the spatial transformation matrix in the spatiotemporal transformation equation, thus obtaining the short-term data prediction result.

[0017] Optionally, the spatiotemporal transformation equation is:

[0018]

[0019] Where A represents the time transformation matrix; B represents the spatial transformation matrix; A×B=I, I represents the identity matrix; Φ(·) represents the nonlinear function; S t The sensor signal at time t; y t =y m+1 ,y m+2 ,…,y m+L-1 This represents the short-term data prediction result, t = 1, 2, ..., m, where m represents m known times; L represents the time length of the future time information, T t This represents the time of the first pulse (t).

[0020] Optionally, the phototransistor in each spiking neuron unit is subjected to a drain voltage and a positive gate voltage greater than its threshold voltage to sample the sensing signal input to the spiking neuron unit through the phototransistor to obtain the source-drain current signal.

[0021] Optionally, the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal. The volatile memristor of the spiking neuron unit performs frequency-domain spike encoding on the source-drain current signal and outputs a spike current signal, including:

[0022] The phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs the source leakage current signal;

[0023] When the load voltage of the volatile memristor in the spiking neuron unit is greater than the turn-on voltage of the volatile memristor, the volatile memristor switches to a low-resistance state; the load voltage is the voltage division of the source-drain current signal;

[0024] When the load voltage of the volatile memristor in the spiking neuron unit is less than the turn-on voltage of the volatile memristor, the volatile memristor switches to a high-resistance state.

[0025] The volatile memristor forms an oscillating spike code under the action of low-resistance and high-resistance states, and obtains a spike current signal through the spike code.

[0026] Secondly, the present invention provides a system for predicting short-term data based on a spiking neuron array, the system comprising:

[0027] Sensor array, spiking neuron array, first pulse time extraction module, and short-term data prediction module;

[0028] The sensor group is used to collect a set of sensing signals;

[0029] The spiking neuron array is used to output a set of spike current signals according to the set of sensing signals; wherein, the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal, and the volatile memristor of the spiking neuron unit performs frequency domain spike encoding on the source-drain current signal and outputs a spike current signal.

[0030] The first pulse time extraction module is used to obtain a set of first pulse times based on the set of peak current signals;

[0031] The short-term data prediction module is used to obtain short-term data prediction results based on the set of first pulse times using a spatiotemporal transformation equation; the short-term data prediction results are spatial information containing future time information.

[0032] Optionally, the short-term data prediction module is specifically used to encode the time of the first pulse into future time information using the time transformation matrix in the spatiotemporal transformation equation based on the time delay theory, and to decode the future time information into spatial information containing the future time information using the spatial transformation matrix in the spatiotemporal transformation equation, thereby obtaining the short-term data prediction result.

[0033] Optionally, the spatiotemporal transformation equation is:

[0034]

[0035] Where A represents the time transformation matrix; B represents the spatial transformation matrix; A×B=I, I represents the identity matrix; Φ(·) represents the nonlinear function; S t Y represents the sensing signal at time t; t =y m+1 ,y m+2 ,…,y m+L-1 This represents the short-term data prediction result, t = 1, 2, ..., m, where m represents m known times; L represents the time length of the future time information, T t This represents the time of the first pulse (t).

[0036] Optionally, the phototransistor in each spiking neuron unit is subjected to a drain voltage and a positive gate voltage greater than its threshold voltage to sample the sensing signal input to the spiking neuron unit through the phototransistor to obtain the source-drain current signal.

[0037] Optionally, in the spiking neuron array, the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal. The volatile memristor of the spiking neuron unit performs frequency domain spike encoding on the source-drain current signal and outputs a spike current signal, including:

[0038] The phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs the source leakage current signal;

[0039] When the load voltage of the volatile memristor in the spiking neuron unit is greater than the turn-on voltage of the volatile memristor, the volatile memristor switches to a low-resistance state; the load voltage is the voltage division of the source-drain current signal;

[0040] When the load voltage of the volatile memristor in the spiking neuron unit is less than the turn-on voltage of the volatile memristor, the volatile memristor switches to a high-resistance state.

[0041] The volatile memristor forms an oscillating spike code under the action of low-resistance and high-resistance states, and obtains a spike current signal through the spike code.

[0042] This invention provides a method for predicting short-term data based on a spiking neuron array. The spiking neuron array comprises multiple spiking neuron units; each spiking neuron unit includes a phototransistor and a volatile memristor; the phototransistor and volatile memristor in each spiking neuron unit are connected in series. The method includes inputting a set of sensing signals into the spiking neuron array, causing the array to output a set of peak current signals. Each spiking neuron unit's phototransistor senses the input sensing signals and outputs a source-drain current signal. The volatile memristor of the spiking neuron unit performs frequency-domain peak encoding on the source-drain current signal, outputting a peak current signal. Based on the set of peak current signals, a set of first pulse times is obtained. Based on the set of first pulse times, a short-term data prediction result is obtained using a spatiotemporal transformation equation; the short-term data prediction result is spatial information containing future time information.

[0043] After obtaining a set of first pulse times, short-term data prediction results can be obtained using the spatiotemporal transformation equation. Compared with traditional time prediction hardware, which requires a large amount of time data for training to achieve prediction, this solution can achieve short-term data prediction without a large amount of data training.

[0044] Volatile memristors possess high cycle life and high response speed. Therefore, the spiking neuron array in this scheme can complete the sensing of the perceived signal and frequency domain spike encoding at extremely high speed and extremely low power consumption, thereby quickly realizing short-term data prediction. Furthermore, due to the small size of memristors, chip area can be greatly reduced.

[0045] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating a method for predicting short-term data based on a spiking neuron array, provided in an embodiment of the present invention.

[0047] Figure 2 This is a circuit structure diagram of a spiking neuron array provided in an embodiment of the present invention;

[0048] Figure 3 This is a finished product diagram of a spiking neuron array provided in an embodiment of the present invention;

[0049] Figure 4 This is a microscope image of the internal structure of a spiking neuron array provided in an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of a scenario for short-term data prediction using a method based on a spiking neuron array, as provided in an embodiment of the present invention.

[0051] Figure 6 This is a schematic diagram of a system structure for predicting short-term data based on a spiking neuron array, provided by an embodiment of the present invention. Detailed Implementation

[0052] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0053] To address the current problem of difficulty in predicting short-term data without a large amount of observational training data, this invention provides a method and system for predicting short-term data based on a spiking neuron array.

[0054] First, a method for predicting short-term data based on a spiking neuron array, provided by an embodiment of the present invention, will be described in detail. See [link to relevant documentation]. Figure 1 , Figure 1 This is a flowchart illustrating a method for predicting short-term data based on a spiking neuron array according to an embodiment of the present invention. The spiking neuron array includes multiple spiking neuron units; each spiking neuron unit includes a phototransistor and a volatile memristor; the phototransistor and volatile memristor in each spiking neuron unit are connected in series. The method includes:

[0055] Step S101: A set of sensing signals is input to the spiking neuron array so that the spiking neuron array outputs a set of spike current signals. In this step, the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal. The volatile memristor of the spiking neuron unit performs frequency domain spike encoding on the source-drain current signal and outputs a spike current signal.

[0056] In this embodiment, the spiking neuron array is a 1PT1TS array, comprising multiple spiking neuron units, wherein each spiking neuron unit includes a phototransistor (PT) and a volatile memristor (TS). The phototransistor and volatile memristor are fabricated using photolithography, and the volatile memristor is integrated into the drain / source of the phototransistor. The phototransistor and volatile memristor in each spiking neuron unit are connected in series to form a 1PT-1TS structure.

[0057] In this embodiment, the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal. The volatile memristor of the spiking neuron unit performs frequency-domain spike encoding on the source-drain current signal and outputs a spike current signal. That is, for a set of sensing signals, a set of spike current signals will be output correspondingly under the action of the spiking neuron array, which includes multiple spiking neuron units.

[0058] In this embodiment, the phototransistor can convert the sensing signal into an electrical signal. Specifically, the phototransistor can sense the input N-dimensional sensing signal and output a source-leakage current signal. The sensing signal is an external sensing signal, such as an optical signal.

[0059] Volatile memristors have advantages such as simple structure, low power consumption, easy integration, and compatibility with CMOS (complementary metal-oxide-semiconductor, an important chip in computer systems) technology.

[0060] Specifically, the volatile memristor in this embodiment can be an NbOx Mott memristor. As a novel device based on a new principle, the NbOx Mott memristor has a simple structure, good miniaturization, and is easy to integrate on a large scale.

[0061] Step S102: Obtain a set of first pulse times based on a set of peak current signals.

[0062] In this embodiment, after obtaining a set of peak current signals, a set of first pulse times can be extracted from the set of peak current signals; there is one first pulse time for each peak current signal.

[0063] Specifically, the operation of obtaining the first pulse time from the spike current signal can be achieved through the following steps:

[0064] a) Use signal processing algorithms (such as peak detection algorithms) to detect the location of each spike.

[0065] b) Depending on the actual signal conditions, a threshold can be set to determine when a spike is detected. This threshold can typically be determined using the signal-to-noise ratio or the absolute amplitude.

[0066] c) Once the spike is detected, record the corresponding time point, which is the first pulse time.

[0067] For example, a set of spike current signals is The acquired set of first pulse times can be Where t represents t known time points; D represents the number of sensors in the spiking neuron array. From The first pulse time obtained from the data. From The first pulse time obtained from the data.

[0068] In this embodiment, the first pulse time refers to the time when the first pulse occurs, which can also be understood as the arrival time of the first spike in the signal. The basic idea of ​​obtaining the first pulse time is to encode information using the time when the neuron first generates a pulse. The time when the neuron first generates a pulse can transmit and encode information from external stimuli.

[0069] Step S103: Based on a set of first pulse times, obtain short-term data prediction results using spatiotemporal transformation equations; the short-term data prediction results are spatial information containing future time information.

[0070] In this embodiment, based on the time-varying and non-stationary nature of real-world dynamic systems, a set of first pulse times can be mapped to time dynamics information using the spatiotemporal transformation equation. That is, a set of first pulse times can be mapped to spatial information containing future time information using the spatiotemporal transformation equation.

[0071] In this embodiment, after obtaining a set of first pulse times, short-term data prediction results can be obtained using the spatiotemporal transformation equation. Compared with traditional time prediction hardware that requires a large amount of time data for training, this solution can achieve short-term data prediction without a large amount of data training.

[0072] Volatile memristors possess high cycle life and high response speed. Therefore, the spiking neuron array in this scheme can complete the sensing of the perceived signal and frequency domain spike encoding at extremely high speed and extremely low power consumption, thereby quickly realizing short-term data prediction. Furthermore, due to the small size of memristors, chip area can be greatly reduced.

[0073] Furthermore, the use of continuous-time data representation in this scheme can avoid the problem of processor processing speed being limited by the switching between different logical states, thus overcoming the processor speed bottleneck.

[0074] In this embodiment of the invention, the phototransistor in each spiking neuron unit is subjected to a drain voltage and a positive gate voltage greater than its threshold voltage, so as to sample the sensing signal input to the spiking neuron unit through the phototransistor to obtain the source-drain current signal.

[0075] In this embodiment, by applying a drain voltage greater than the threshold voltage V to the phototransistor TH Positive gate voltage V GS (V GS >V TH This turns on the phototransistor, thereby activating the sensing signal S input to the spiking neuron unit. t Sampling is performed to obtain the source-drain current signal I tAmong them, for S t =s1 t ,s2 t ,…,s D t Correspondingly, the source-drain current signals I at time t are obtained. t =I1 t I2 t ,…,I D t Where t = 1, 2, ..., m, m represents m known times.

[0076] In this embodiment of the invention, the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal. The volatile memristor of the spiking neuron unit performs frequency domain spike encoding on the source-drain current signal and outputs a spike current signal, including:

[0077] The phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs the source leakage current signal;

[0078] When the load voltage of the volatile memristor in the spiking neuron unit is greater than the turn-on voltage of the volatile memristor, the volatile memristor switches to a low-resistance state; the load voltage is the voltage division of the source-drain current signal.

[0079] When the load voltage of the volatile memristor in the spiking neuron unit is less than the turn-on voltage of the volatile memristor, the volatile memristor switches to a high-resistance state.

[0080] Under the influence of low and high resistance states, this volatile memristor generates an oscillating spike code, and through the spike code, a spike current signal is obtained.

[0081] Using a volatile memristor in low-resistance and high-resistance states, the source-drain current signal I at time t is analyzed. t =I1 t I2 t ,…,I D t Encoding is performed to form an oscillating spike code, resulting in the spike current signal X at time t. t =x1 t x2 t ,…,x D t .

[0082] See Figure 2 , Figure 3 and Figure 4 , Figure 2 This is a circuit structure diagram of a spiking neuron array provided in an embodiment of the present invention. Figure 3 This is a finished product diagram of a spiking neuron array provided in an embodiment of the present invention. Figure 4 This is a microscope image of the internal structure of a spiking neuron array provided in an embodiment of the present invention, wherein... Figure 4 The scale in the text is 1 mm.

[0083] In this embodiment, a 4×4 spiking neuron array chip is used as an example. This array includes: 4×4 phototransistors and 4×4 volatile memristors; gate voltage V GS That is, positive gate voltage, bias voltage V DS That is, the drain voltage, and I1, I2, I3, I4 represent the output spike current signals.

[0084] In this embodiment, the volatile memristor can be an NbOx Mott memristor, which has high cycle life (>10). 10 It boasts advantages such as high response speed (~10ns), self-oscillation, and CMOS compatibility. A pulse neuron array employing NbOx Mott memristors can complete signal sampling and spike encoding at extremely high speed and with extremely low power consumption, requiring no additional control unit, with a single operation power consumption of only 0.1nJ.

[0085] In this embodiment, a 1PT1TS array is used to perform motion exposure processing on external sensor signals, thereby achieving large-dimensional image prediction. It is worth noting that this solution integrates sensing and spike coding, allowing the backend signal to be processed directly in the analog domain without the need for an ADC (analog-to-digital converter).

[0086] In an embodiment of the present invention, short-term data prediction results are obtained using a spatiotemporal transformation equation based on a set of first pulse times, including:

[0087] Based on the time delay theory, a set of first pulse times is encoded into future time information through the time transformation matrix in the spatiotemporal transformation equation, and the future time information is decoded into spatial information containing future time information through the spatial transformation matrix in the spatiotemporal transformation equation, thus obtaining short-term data prediction results.

[0088] In a typical space, although the unsteady real world has many interrelationships, it is confined to a low-dimensional subspace due to dissipation. Now we need to reconstruct the low-dimensional attractors, and according to the time delay theory, we can use high-dimensional spatial information to exchange for transient information of time dynamics.

[0089] In this embodiment of the invention, the spatiotemporal transformation equation is:

[0090]

[0091] Where A represents the time transformation matrix; B represents the spatial transformation matrix; A×B=I, I represents the identity matrix; Φ(·) represents the nonlinear function; S t Y represents the sensing signal at time t; t =y m+1 ,y m+2 ,…,y m+L-1 This represents the short-term data prediction result, t = 1, 2, ..., m, where m represents m known time points; L represents the time length of the future time information, T t This represents the time of the first pulse (t).

[0092] In this embodiment, the time transformation matrix is ​​an L×D matrix, the space transformation matrix is ​​a D×L matrix, and I represents the L×L identity matrix, where D>L.

[0093] In this embodiment, Y t =(Φ(t1) t ),Φ(t2 t )…Φ(t D t )), T t =ΨY t , Φ°Ψ=id, where ° represents the function combination operation, id represents the identity function, and Ψ represents the inverse transformation.

[0094] For the known T t Solving the spacetime transformation equation, i.e., solving the time transformation matrix and the space transformation matrix, allows us to obtain Y. t This refers to short-term data forecast results.

[0095] Specifically, for a given spatiotemporal transformation equation, the spatial transformation matrix and the temporal transformation matrix are obtained by using time delay theory and the dropout method, and then the spatial information Y containing future time information is obtained. t Specifically, the dropout method can be used to update the spatial transformation matrix, obtaining the updated spatial transformation matrix at each step, and thus obtaining the iterative solution Y. t The L2 norm is calculated until convergence, and the final output is Y. t .

[0096] See Figure 5 , Figure 5 This is a schematic diagram illustrating a scenario where a method based on a spiking neuron array is used to predict short-term data, as provided in an embodiment of the present invention. Specifically, it shows the effect of applying this method to predict and reconstruct the future trajectory of a moving object. Here, t=1 to t=4 are known times, and t=5, t=6…t=9 are unknown times, i.e., the prediction times.

[0097] exist Figure 5The image shows the original feature map and the prediction results of this scheme. 0μs to 4μs represents the first pulse time. Figure 5 The graphs corresponding to t=5, t=6…t=9 in this work unit show the prediction results obtained using this scheme. From Figure 5 As can be seen, the method for predicting short-term data based on a spiking neuron array provided by the embodiments of the present invention can achieve fast and accurate prediction of short-term data.

[0098] Based on the same inventive concept, embodiments of the present invention also provide a system for predicting short-term data based on a spiking neuron array, see [link to relevant documentation]. Figure 6 , Figure 6 This is a schematic diagram of a system architecture for predicting short-term data based on a spiking neuron array, provided by an embodiment of the present invention, including:

[0099] Sensor group 601, spiking neuron array 602, first pulse time extraction module 603, and short-term data prediction module 604;

[0100] Sensor group 601 is used to collect a set of sensing signals;

[0101] The spiking neuron array 602 is used to output a set of spike current signals according to a set of sensing signals; wherein, the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal, and the volatile memristor of the spiking neuron unit performs frequency domain spike encoding on the source-drain current signal and outputs a spike current signal.

[0102] The first pulse time extraction module 603 is used to obtain a set of first pulse times based on a set of peak current signals;

[0103] The short-term data prediction module 604 is used to obtain short-term data prediction results based on a set of first pulse times using a spatiotemporal transformation equation; the short-term data prediction results are spatial information containing future time information.

[0104] In this embodiment, after obtaining a set of first pulse times, a short-term data prediction structure can be obtained using the spatiotemporal transformation equation. Compared with traditional time prediction hardware that requires a large amount of time data for training, this solution can achieve short-term data prediction without a large amount of data training.

[0105] Volatile memristors possess high cycle life and high response speed. Therefore, the spiking neuron array in this scheme can complete the sensing of the perceived signal and frequency domain spike encoding at extremely high speed and extremely low power consumption, thereby quickly realizing short-term data prediction. Furthermore, due to the small size of memristors, chip area can be greatly reduced.

[0106] Furthermore, the use of continuous-time data representation in this scheme can avoid the problem of processor processing speed being limited by the switching between different logical states, thus overcoming the processor speed bottleneck.

[0107] Optionally, the short-term data prediction module is specifically used to encode the time of the first pulse into future time information using the time transformation matrix in the spatiotemporal transformation equation based on the time delay theory, and to decode the future time information into spatial information containing the future time information using the spatial transformation matrix in the spatiotemporal transformation equation, thereby obtaining the short-term data prediction result.

[0108] Optionally, the spatiotemporal transformation equation is:

[0109]

[0110] Where A represents the time transformation matrix; B represents the spatial transformation matrix; A×B=I, I represents the identity matrix; Φ(·) represents the nonlinear function; S t Y represents the sensing signal at time t; t =y m+1 ,y m+2 ,…,y m+L-1 This represents the short-term data prediction result, t = 1, 2, ..., m, where m represents m known times; L represents the time length of the future time information, T t This represents the time of the first pulse (t).

[0111] Optionally, the phototransistor in each spiking neuron unit is subjected to a drain voltage and a positive gate voltage greater than its threshold voltage to sample the sensing signal input to the spiking neuron unit through the phototransistor to obtain the source-drain current signal.

[0112] Optionally, a spiking neuron array is provided, wherein the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal, and the volatile memristor of the spiking neuron unit performs frequency-domain spike encoding on the source-drain current signal to output a spike current signal, including:

[0113] The phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs the source leakage current signal;

[0114] When the load voltage of the volatile memristor in the spiking neuron unit is greater than the turn-on voltage of the volatile memristor, the volatile memristor switches to a low-resistance state; the load voltage is the voltage division of the source-drain current signal.

[0115] When the load voltage of the volatile memristor in the spiking neuron unit is less than the turn-on voltage of the volatile memristor, the volatile memristor switches to a high-resistance state.

[0116] Under the influence of low and high resistance states, this volatile memristor generates an oscillating spike code, and through the spike code, a spike current signal is obtained.

[0117] It should be noted that, for the system implementation, since it is basically similar to the method implementation, the description is relatively simple, and relevant parts can be referred to in the description of the method implementation.

[0118] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of systems and methods consistent with some aspects of the invention.

[0119] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0120] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0121] It should be noted that the system in this embodiment of the invention is a system that applies the above-described method for predicting short-term data based on a spiking neuron array. Therefore, all embodiments of the above-described method for predicting short-term data based on a spiking neuron array are applicable to this system and can achieve the same or similar beneficial effects.

[0122] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for predicting short-term data based on a spiking neuron array, characterized in that, The spiking neuron array includes: multiple spiking neuron units; each spiking neuron unit includes a phototransistor and a volatile memristor; the phototransistor and the volatile memristor in each spiking neuron unit are connected in series; The method includes: A set of sensing signals is input to the spiking neuron array so that the spiking neuron array outputs a set of spike current signals accordingly; wherein, the phototransistor in each spiking neuron unit senses the sensing signals input to the spiking neuron unit and outputs a source-drain current signal, and the volatile memristor of the spiking neuron unit performs frequency domain spike encoding on the source-drain current signal and outputs a spike current signal. Based on the set of peak current signals, a set of first pulse times is obtained; Based on the set of first pulse times, short-term data prediction results are obtained using spatiotemporal transformation equations; the short-term data prediction results are spatial information containing future time information. The step of obtaining short-term data prediction results based on the set of first pulse times using a spatiotemporal transformation equation includes: Based on the time delay theory, the time of the first pulse is encoded into future time information through the time transformation matrix in the spatiotemporal transformation equation, and the future time information is decoded into spatial information containing the future time information through the spatial transformation matrix in the spatiotemporal transformation equation, so as to obtain the short-term data prediction result. The spatiotemporal transformation equation is: ； in, This represents the time transformation matrix; Represents the spatial transformation matrix; , Represents the identity matrix; Represents a nonlinear function; express The sensing signal at any given moment; This indicates the short-term data prediction result. express A known time; This indicates the duration of the future time information. express The duration of each first pulse.

2. The method according to claim 1, characterized in that, In each spiking neuron unit, a phototransistor is subjected to a drain voltage and a positive gate voltage greater than its threshold voltage to sample the sensing signal input to the spiking neuron unit and obtain the source-drain current signal.

3. The method according to claim 1, characterized in that, The phototransistor in each spiking neuron unit senses the input sensing signal and outputs a source-drain current signal. The volatile memristor of the spiking neuron unit performs frequency-domain spike encoding on the source-drain current signal and outputs a spike current signal, including: The phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs the source leakage current signal; When the load voltage of the volatile memristor in the spiking neuron unit is greater than the turn-on voltage of the volatile memristor, the volatile memristor switches to a low-resistance state; the load voltage is the voltage division of the source-drain current signal; When the load voltage of the volatile memristor in the spiking neuron unit is less than the turn-on voltage of the volatile memristor, the volatile memristor switches to a high-resistance state. The volatile memristor forms an oscillating spike code under the action of low-resistance and high-resistance states, and obtains a spike current signal through the spike code.

4. A system for predicting short-term data based on a spiking neuron array, characterized in that, The system includes: Sensor array, spiking neuron array, first pulse time extraction module, and short-term data prediction module; The sensor group is used to collect a set of sensing signals; The spiking neuron array is used to output a set of spike current signals according to the set of sensing signals; wherein, the phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs a source-drain current signal, and the volatile memristor of the spiking neuron unit performs frequency domain spike encoding on the source-drain current signal and outputs a spike current signal. The first pulse time extraction module is used to obtain a set of first pulse times based on the set of peak current signals; The short-term data prediction module is used to obtain short-term data prediction results based on the set of first pulse times using a spatiotemporal transformation equation; the short-term data prediction results are spatial information containing future time information. The short-term data prediction module is specifically used to encode the time of the first pulse into future time information using the time transformation matrix in the spatiotemporal transformation equation based on the time delay theory, and to decode the future time information into spatial information containing the future time information using the spatial transformation matrix in the spatiotemporal transformation equation, thereby obtaining the short-term data prediction result. The spatiotemporal transformation equation is: ； in, This represents the time transformation matrix; Represents the spatial transformation matrix; , Represents the identity matrix; Represents a nonlinear function; express The sensing signal at any given moment; This indicates the short-term data prediction result. express A known time; This indicates the duration of the future time information. express The duration of each first pulse.

5. The system according to claim 4, characterized in that, In each spiking neuron unit, a phototransistor is subjected to a drain voltage and a positive gate voltage greater than its threshold voltage to sample the sensing signal input to the spiking neuron unit and obtain the source-drain current signal.

6. The system according to claim 4, characterized in that, The spiking neuron array, in which each spiking neuron unit's phototransistor senses the sensing signal input to that spiking neuron unit and outputs a source-drain current signal, and the volatile memristor of that spiking neuron unit performs frequency-domain spike encoding on the source-drain current signal, outputting a spike current signal, including: The phototransistor in each spiking neuron unit senses the sensing signal input to the spiking neuron unit and outputs the source leakage current signal; When the load voltage of the volatile memristor in the spiking neuron unit is greater than the turn-on voltage of the volatile memristor, the volatile memristor switches to a low-resistance state; the load voltage is the voltage division of the source-drain current signal; When the load voltage of the volatile memristor in the spiking neuron unit is less than the turn-on voltage of the volatile memristor, the volatile memristor switches to a high-resistance state. The volatile memristor forms an oscillating spike code under the action of low-resistance and high-resistance states, and obtains a spike current signal through the spike code.

Images