Method and device for implementing a radial basis neuron based on a memristor

Abstract

The application discloses a radial basis neuron implementation method and equipment based on a memristor, relates to the technical field of the memristor, and comprises the following steps: receiving an input voltage of a neuron circuit; delivering the input voltage to a pre-established absolute value module circuit to obtain a voltage signal with a non-negative amplitude; inputting the voltage signal with the negative amplitude into a pre-established inverting amplifier module to obtain an inverted voltage; inputting the inverted voltage and a bias voltage provided by a bias module into a radial basis neuron module circuit through a summing module to obtain output signals with different frequencies; and obtaining a Gaussian function through superposition of the output signals with the different frequencies, so that the radial basis neuron is implemented. The application can effectively clean up input data in the form of clustering and prevent black box attacks.

Description

Technical Field

[0001] This invention relates to the field of memristor technology, specifically a method and device for implementing radial basis neurons based on memristors. Background Technology

[0002] With the rise of artificial intelligence and the rapid development of various new devices, hardware-based brain-inspired computing systems have attracted much attention due to their high energy efficiency and high integration, becoming powerful contenders to disrupt conventional approaches. The von Neumann architecture is a prominent feature. Among various novel devices, memristors stand out for their excellent computational potential and biological rationality. However, a large portion of the current mainstream brain-inspired systems' structures and algorithms draw on the experience of ANN neural networks. To address the training difficulties of Spike Neural Networks (SNNs), surrogate gradient descent and transformation-based training methods have been introduced, which are currently the two most mainstream training methods. While the inspiration from artificial neural networks has brought improvements in accuracy and performance, artificial neural networks also exhibit some problems (such as low generalization ability and weak anti-interference ability). To optimize brain-inspired systems, more and more researchers are beginning to try to design memristor synapses and neuronal circuits with more brain functions. For example, using habituated synapses to improve the system's learning ability, and introducing refractory period features into conventional mid-frequency neuronal circuits to improve system performance. Adversarial attacks exploit classification vulnerabilities in the system to poison the dataset, mixing infected data with normal data for verification, thereby achieving the effect of disrupting system performance. This type of data virus poses a significant threat to system security.

[0003] However, effective solutions are still lacking in hardware-based brain-inspired systems. In long-term research on the nervous system, researchers discovered that the visual nervous system relies on the principle of "proximal enhancement and distant inhibition" to control distance perception, and inspired by this, they designed radial basis function (RBF) neural networks. Unlike the ReLU activation function used in traditional ANN networks, the RDF function only responds to a certain range of input signals, and the closer to the center point, the stronger the response. This clustering-like property inspires us to use RDF networks to remove data viruses. Summary of the Invention

[0004] To address the shortcomings mentioned in the background art, the present invention aims to provide a method and device for implementing radial basis function neurons based on memristors. By utilizing the characteristics of threshold memristors and characterizing them with different frequencies of the output signal, a Gaussian function is obtained to realize radial basis function neurons.

[0005] The objective of this invention can be achieved through the following technical solution: a method for implementing radial basis function neurons based on memristors, the method comprising the following steps:

[0006] Receives the input voltage of the neuron circuit;

[0007] The input voltage is fed into a pre-established absolute value module circuit to obtain a voltage signal with a non-negative amplitude;

[0008] A negative voltage signal is input into a pre-established inverting amplifier module to obtain an inverting voltage;

[0009] The reverse voltage and the bias voltage provided by the bias module are input to the radial basis neuron module circuit through the summation module to obtain output signals of different frequencies. By superimposing the output signals of different frequencies, a Gaussian function is obtained to realize the radial basis neuron.

[0010] Preferably, the absolute value module circuit takes the amplitude of the input voltage Vin to obtain a voltage signal with a non-negative amplitude.

[0011] Preferably, the inverting amplifier module consists of two identical resistors and an operational amplifier, and the input of the inverting amplifier module is connected to the output of the absolute value module.

[0012] Preferably, the input port of the summing module is connected to the output of the inverting amplifier and the bias circuit, respectively, and the output port is connected to the input of the radial basis neuron module.

[0013] Preferably, the bias module provides an adjustable bias voltage and is connected to the input port of the summing module.

[0014] Preferably, the radial basis function neuron module consists of a memristor M, a capacitor C, and a resistor R1.

[0015] Preferably, the memristor is initially in a high-resistance state. By changing the voltage across the memristor, a set / reset operation is performed. When the applied voltage is higher than the threshold voltage, the memristor sets—switching from a high-resistance state to a low-resistance state. When the voltage across the memristor is lower than the holding voltage, the memristor resets—switching from a low-resistance state to a high-resistance state.

[0016] Preferably, the process of obtaining the Gaussian function involves fitting the simulated data to obtain a smooth Gaussian function:

[0017]

[0018] Preferably, the frequency of the output signal represents the output of the neuron to realize a radial basis neuron.

[0019] In another aspect of the present invention, in order to achieve the above-mentioned objective, an apparatus is disclosed, comprising:

[0020] One or more processors;

[0021] Memory, used to store one or more programs;

[0022] When one or more of the programs are executed by one or more of the processors, the one or more of the processors implement a memristor-based radial basis function neuron implementation method as described above.

[0023] The beneficial effects of this invention are:

[0024] This invention can effectively clean up input data in the form of clustering, preventing black-box attacks. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0027] Figure 2 This is a schematic diagram of the characteristics of the memristor of the present invention;

[0028] Figure 3 This is a schematic diagram of the memristor structure of the present invention;

[0029] Figure 4 This is a schematic diagram of the input signal frequency-amplitude of the present invention;

[0030] Figure 5 This is a schematic diagram of the output signal waveform of the present invention;

[0031] Figure 6 This is a schematic diagram of the overall circuit of the present invention;

[0032] Figure 7 This is a schematic diagram of the absolute value module circuit structure of the present invention;

[0033] Figure 8 This is a schematic diagram of the circuit structure of the inverting amplifier module of the present invention;

[0034] Figure 9 This is a schematic diagram of the summation module circuit structure of the present invention;

[0035] Figure 10 This is a diagram showing the waveforms and corresponding frequencies of the output signals when the input signal is from 0V to (-0.8)V and from 0V to (+0.8)V, according to an embodiment of the present invention.

[0036] Figure 11 This is a schematic diagram of a Gaussian function. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] like Figure 1 As shown, a method for implementing radial basis function neurons based on memristors includes the following steps:

[0039] Receives the input voltage of the neuron circuit;

[0040] The input voltage is fed into a pre-established absolute value module circuit to obtain a voltage signal with a non-negative amplitude;

[0041] A negative voltage signal is input into a pre-established inverting amplifier module to obtain an inverting voltage;

[0042] The reverse voltage and the bias voltage provided by the bias module are input to the radial basis neuron module circuit through the summation module to obtain output signals of different frequencies. By superimposing the output signals of different frequencies, a Gaussian function is obtained to realize the radial basis neuron.

[0043] In another aspect of this invention, to achieve the above objective, a memristor based on the PF mechanism is used, whose IV characteristics are as follows: Figure 2 The initial state of a memristor is a high-resistance state. By changing the voltage across the memristor, a set / reset operation can be performed. When the applied voltage is higher than the threshold voltage (Vth), the device sets—switching from a high-resistance state to a low-resistance state. When the voltage across the device is lower than the holding voltage (Vh), the device resets—switching from a low-resistance state to a high-resistance state.

[0044] Different relative permittivity (εr) of the conductive filaments in the model will change the threshold voltage (Vth) and holding voltage (Vh) of the memristor. With relative permittivity (εr) of 9, 13.5, 18, and 22.5, the memristor model exhibits the following IV characteristics: Figure 3 As shown.

[0045] Output signals of different frequencies are realized by using charging and discharging of a capacitor. The initial state of the memristor is a high resistance state. When the input signal Vin=0V, the obtained output Vout=0V; 1) ① When the input Vin provides a certain level higher than the threshold voltage (Vth) of the memristor, a set operation is performed on the memristor (M), which changes from a high resistance state to a low resistance state. To reach a balanced state, the capacitor starts to discharge, the voltage of the output port Vout increases, and briefly reaches a certain stable state; ② However, as Vout increases, the voltage difference across M decreases. When Vin-Vout<Vh, a reset operation is performed on M, which changes from a low resistance state to a high resistance state. To reach a new balanced state, the capacitor starts to charge, the voltage of the output port Vout decreases, and reaches a new stable state; ③ As the Vout voltage decreases, the voltage difference across M increases. When Vin-Vout>Vth, a set operation is performed on M, and the dynamic balance process described in process ① occurs subsequently. This reciprocating process results in a periodic signal of a certain frequency obtained at the output end. As Figure 4 , when the input signal reaches a certain amplitude (that is, the voltage difference across the memristor is greater than the threshold voltage), the output signal frequency increases as the input signal voltage increases; Figure 5 , shows the output signal waveforms and their corresponding frequencies when the input signals are 2V, 2.1V, 2.2V, 2.3V and 2.4V respectively.

[0046] To output a complete Gaussian function, an absolute value circuit, an inverting amplification circuit and a bias circuit are added on the basis of the basic (semi-) radial basis neuron circuit to form a Figure 6 radial basis neuron circuit shown in . The circuit consists of an absolute value module (ABS module), an inverting amplifier module (Inverting amplifier module), a bias module (Bias), a summation module (sum module), and a basic (semi-) radial basis neuron circuit module;

[0047] Composition and functions of each circuit module:

[0048] Absolute value module: The input signal of the absolute value module is Vin, and its function is to take the amplitude of the input level signal. For example: when the input signal is 1V / -0.5V, the output signal is 1V / 0.5V, as Figure 7 shown;

[0049] The inverting amplifier is composed of two identical resistors and an operational amplifier; the input of the inverting amplifier module is connected to the output of the absolute value module, and its function is to invert the signal from the ABS module. For example, the ABS module outputs a level signal of 1V / -1.5V, and after passing through the inverting amplifier module, a level signal of -1V / 1.5V is obtained, as Figure 8 shown;

[0050] Bias module: Provides an adjustable bias voltage and connects to port 2 of the summing module;

[0051] Summation Module: Ports 1 and 2 of the summation module are input ports, and port 3 is the output port. The input ports are connected to the output and bias circuit of the inverting amplifier, respectively, and the output is connected to the input of the radial basis function neuron circuit. Its function is to sum the voltage signals of the input ports 1 and 2. For example, if the voltages of ports 1 and 2 are -1V and 2.5V respectively, the output voltage of port 3 is 2.5V + (-1)V = 1.5V. Figure 9 As shown.

[0052] Basic (semi)radial basis function neuron module: This module consists of a threshold memristor (M), a capacitor (C), and a resistor (R1). Changes in the input voltage at point Q cause a change in the frequency of the output signal.

[0053] Figure 10 (a) and Figure 10 (b) shows the waveforms and corresponding frequencies of the output signals when the input signal Vin is from 0V to (-0.8)V and from 0V to (+0.8)V, respectively.

[0054] Figure 11 This demonstrates how to obtain Gaussian functions with different variances using designed radial basis function neurons to realize radial basis function neural networks (RBFNN).

[0055] In another aspect of this invention, to achieve the above objective, the process of obtaining the Gaussian function involves fitting the simulated data to obtain a smoothed Gaussian function:

[0056] Where r represents the distance to the center point, and σ represents the standard deviation.

[0057] In another aspect of the present invention, in order to achieve the above objective, the frequency of the output signal represents the output of the neuron, thereby realizing a radial basis neuron.

[0058] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0059] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0060] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0061] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

Claims

1. A method for implementing radial basis function neurons based on memristors, characterized in that, The method includes the following steps: Receives the input voltage of the neuron circuit; The input voltage is fed into a pre-established absolute value module circuit to obtain a voltage signal with a non-negative amplitude; The absolute value module circuit takes the amplitude of the input voltage Vin and obtains a voltage signal with a non-negative amplitude. A negative voltage signal is input into a pre-established inverting amplifier module to obtain an inverting voltage; The inverting amplifier module consists of two identical resistors and an operational amplifier, and the input of the inverting amplifier module is connected to the output of the absolute value module; The reverse voltage and the bias voltage provided by the bias module are input to the radial basis neuron module circuit through the summation module to obtain output signals of different frequencies; The input ports of the summation module are connected to the output and bias circuit of the inverting amplifier, respectively, and the output port is connected to the input of the radial basis neuron module. The radial basis function neuron module consists of a memristor M, a capacitor C, and a resistor R1; The bias module provides an adjustable bias voltage and connects to the input port of the summation module; The frequency of the output signal represents the output of the neuron, thus realizing a radial basis neuron; Gaussian functions are obtained by superimposing output signals of different frequencies to realize radial basis neurons.

2. The method for implementing radial basis function neurons based on memristors according to claim 1, characterized in that, The memristor is initially in a high-resistance state. By changing the voltage across the memristor, a set / reset operation is performed. When the applied voltage is higher than the threshold voltage, the memristor sets from a high-resistance state to a low-resistance state. When the voltage across the memristor is lower than the holding voltage, the memristor resets from a low-resistance state to a high-resistance state.

3. The method for implementing radial basis function neurons based on memristors according to claim 1, characterized in that, The process of obtaining the Gaussian function involves fitting the simulated data to obtain a smoothed Gaussian function. φ(x) = exp(- ), where r represents the distance to the center point and σ represents the standard deviation.

4. A device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When one or more of the programs are executed by one or more of the processors, the one or more of the processors implement a memristor-based radial basis function neuron implementation method as described in any one of claims 1-3.

Images