Reconfigurable brain-like neuron circuit based on fast and slow circuit

Abstract

The invention discloses a reconfigurable brain-like neuron circuit based on a fast and slow circuit, which comprises a fast sub-circuit and a slow sub-circuit, and the fast sub-circuit and the slow sub-circuit are mutually coupled to determine a pulse distribution mode; the high-speed sub-circuit comprises a volatile threshold resistance change device CH1 and a capacitor C1, the input end of the CH1 is connected to a signal source through an input resistor, and the output end of the CH1 is connected to the input end of the low-speed sub-circuit; one end of the C1 is connected with the input end of the CH1, and the other end is grounded; the low-speed sub-circuit comprises a volatile threshold resistance change device CH2, a capacitor C2 and a non-volatile threshold resistance change device which are connected in parallel; the input end of the low-speed sub-circuit is connected with the high-speed sub-circuit and used for receiving pulse current of the high-speed sub-circuit, the output end of the low-speed sub-circuit is connected with the output resistor, and the connection point serves as a pulse output port of a neuron. The information expression ability of the neuron circuit constructed by the method is stronger, the performance of the spiking neural network is improved, and the method can be used for constructing an intelligent computing system with stronger cognitive ability and computing efficiency.

Description

Technical Field

[0001] The present invention relates to the field of semiconductor information computing, and in particular to a reconfigurable brain-like neuron circuit based on fast and slow circuits. Background Art

[0002] Computers based on traditional architecture use storage and computing separation technology, where information storage and computing are performed separately. Although the rapid development of deep learning technology has brought great progress to artificial intelligence, the traditional computing architecture will generate huge energy consumption requirements. Compared with modern computers, the human brain has stronger cognitive ability and high-level computing ability, and can complete various complex tasks with advantages such as high efficiency, high parallelism and low power consumption. This is because the information processing process of the human brain is different from the traditional computing structure. 15 synapses and 10 11 In the neural biological network formed by the connection of neurons, information is transmitted in the form of pulses, which has the characteristics of asynchrony, sparsity, spatiotemporal characteristics and event-driven, and has more powerful information processing capabilities.

[0003] The spiking neural network (SNN) that simulates biological neural computing has the characteristics of biological computing and has advantages such as low power consumption compared to traditional deep neural networks. Neurons, as the computing units of SNN, play an important role in processing and transmitting information. SNN has a variety of neuron models, such as the leaky integrate-and-fire model (LIF), the Hodgkin-Huxley model (HH), and the Izhikevich model.

[0004] For example, the Chinese patent document with publication number CN115984822A discloses a spatial attention-based SNN traffic sign recognition method and related devices, and the neuron encoding layer of the SNN network model includes soft reset LIF neurons. The Chinese patent document with publication number CN108985447A discloses a hardware pulse neural network system, and the types of neurons used include: Mokro-Pitt model, leaky integration and release model, Hodgkin-Huxley model or Izhikevich model.

[0005] At present, the commonly used SNN is based on LIF neurons, which can easily simulate neuron functions. However, the disadvantage is that it lacks the ability to simulate high-order pulse functions, resulting in weak information expression capabilities of neurons, which limits the computing performance of SNNs. In addition, some HH neuron circuits, although they can achieve richer neuron firing patterns, require large hardware overhead and need to change circuit components to complete the switching of circuit pulse firing behavior, which limits the further development of chip integration technology for pulse neurons. Summary of the invention

[0006] The present invention provides a reconfigurable brain-like neuron circuit based on fast and slow circuits, which utilizes the high-order circuit dynamic response generated by the coupling of fast and slow circuits and the design of reconfigurable circuits to achieve simulation and switching of different high-order pulse emission behaviors in biological neurons, overcoming the technical problems in the prior art such as the single function of neuron circuits and the difficulty in integration caused by non-reconfigurable circuits.

[0007] A reconfigurable brain-like neuron circuit based on a fast-slow circuit, including a fast subcircuit and a slow subcircuit, wherein the fast and slow subcircuits are coupled to each other to determine the pulse emission mode;

[0008] The fast sub-circuit includes a volatile threshold resistive switching device CH1 and a capacitor C1. The input end of the threshold resistive switching device CH1 is connected to an input resistor R in connected to a signal source, the output end of the threshold resistance switching device CH1 is connected to the input end of the slow sub-circuit; one end of the capacitor C1 is connected to the input end of the threshold resistance switching device CH1, and the other end is grounded; the threshold resistance switching device CH1 is used to simulate the voltage-gated ion channel of biological neurons to generate a pulse current;

[0009] The slow sub-circuit includes a volatile threshold resistive switching device CH2, a capacitor C2 and a non-volatile threshold resistive switching device NVTSM connected in parallel; the slow sub-circuit input end is connected to the fast sub-circuit for receiving the pulse current of the fast sub-circuit, and the slow sub-circuit output end is connected to the output resistor R load The connection point serves as the pulse output port of the neuron; the resistance of the non-volatile threshold resistive switching device NVTSM is adjusted by controlling the external voltage, and the parallel resistance of the two resistive switching devices dynamically affects the charging and discharging process of the capacitor C2, thereby controlling the capacitor charging and discharging of the fast sub-circuit and the pulse emission behavior of the neuron.

[0010] In the present invention, the signal input is connected to the input port of the fast subcircuit, and the input current generated charges the capacitor C1 of the fast subcircuit, causing the capacitor potential to increase and the membrane voltage to increase. When the membrane voltage reaches the threshold of the volatile threshold resistive variable device, the fast subcircuit is activated and generates a pulse current, which is finally converted into a pulse voltage through the output resistor and output to other neurons.

[0011] The input end of the slow subcircuit is connected to the fast subcircuit to receive the pulse current of the fast subcircuit. The output end of the slow subcircuit is connected to the output resistor, and the connection point serves as the pulse output port of the neuron. After the slow subcircuit receives the pulse current of the fast subcircuit, its internal circuit state changes, and the feedback affects the dynamic behavior of the fast subcircuit. The second-order circuits are coupled with each other to achieve high-order pulse behavior that simulates biological neurons. By adjusting the resistance value of the non-volatile threshold resistive variable device in the slow subcircuit through the control signal, the parallel resistance and potential balance state of the subcircuit will be changed, further affecting the dynamic behavior of the fast subcircuit and the pulse emission of the neuron, thereby controlling the pulse emission mode of the neuron.

[0012] Based on the above design, the firing mode of the neuron circuit is no longer a single pulse firing. Instead, according to the needs of the usage scenario, different high-order pulse firing modes are switched through control signals, which are used as computing units to build a more intelligent neural computing network.

[0013] Furthermore, in the slow sub-circuit, when the resistance of the non-volatile threshold resistive switching device NVTSM changes from low resistance to high resistance, the impact on the fast sub-circuit also gradually changes, and the pulse firing behaviors of neurons are fast peak firing, adaptive peak firing, refractory period firing and burst firing in sequence.

[0014] Furthermore, the resistance of the non-volatile threshold resistive switching device NVTSM changes continuously under the control of an external voltage (ie, a control signal), and the resistance state remains stable after the voltage is removed; according to the requirements of the pulse emission behavior, the non-volatile threshold resistive switching device NVTSM is adjusted to a certain resistance state.

[0015] Furthermore, the volatile threshold resistive switching devices CH1 and CH2 are of the same type of threshold resistive switching devices, whose resistance change is nonlinear and has a threshold switching characteristic; when the capacitor voltage reaches the threshold, the resistance of the threshold resistive switching device suddenly changes from high resistance to low resistance, that is, the device switch is quickly opened; when the capacitor voltage is lower than the threshold, the resistance of the threshold resistive switching device suddenly changes from low resistance back to high resistance, that is, the device switch is quickly closed.

[0016] Furthermore, the fast sub-circuit has a small capacitance value and a fast response speed, while the slow sub-circuit has a large capacitance value and a slow response speed.

[0017] Furthermore, when the neuron membrane potential reaches or exceeds the firing threshold, the resistance of the volatile threshold resistive variable device in the fast subcircuit decreases suddenly, the total current in the circuit increases suddenly, and the output resistance voltage to ground increases suddenly, generating a pulse voltage.

[0018] Furthermore, the pulse output of the neuron circuit is an output resistor R load The voltage divider, output resistance R loadOne end is connected to the output end of the slow subcircuit and serves as an output port, and the other end is grounded.

[0019] Furthermore, the difference between the capacitance potential of the fast sub-circuit and the capacitance potential of the slow sub-circuit is used as the membrane potential of the neuron.

[0020] Preferably, the capacitance ratio of the fast subcircuit to the slow subcircuit is in the range of 0.01 to 1, and the capacitance ratio can affect the current dynamic behavior and pulse emission behavior of the circuit.

[0021] Furthermore, the non-volatile threshold resistive switching device NVTSM includes a gate, a source and a drain, wherein the source and the drain are used to form a parallel connection with the volatile threshold resistive switching device CH2 and the capacitor C2; the gate serves as a control port, and is used to receive external voltage regulation to change the resistance of the non-volatile threshold resistive switching device NVTSM, thereby changing the capacitor potential of the slow sub-circuit by affecting the current balance state of the slow sub-circuit, and further affecting the capacitor potential of the fast sub-circuit and the pulse emission behavior of the neuron.

[0022] Furthermore, when the slow subcircuit capacitor potential is lower than the threshold of the volatile threshold resistive variable device, the device CH2 will not open, the slow subcircuit receives the output current of the fast subcircuit, and the capacitor potential of the slow subcircuit gradually increases, causing the capacitor potential of the fast subcircuit to also increase, so that the membrane potential can reach the neuron's firing threshold. As the NVSTM resistance value increases from small to large, the firing behavior of the neuron is, in order, a fast peak firing with no significant change in firing frequency, an adaptive peak firing from high to low until a stable firing frequency, and a refractory firing from high to low until a refractory period state.

[0023] Furthermore, when the capacitor potential of the slow sub-circuit is higher than the threshold of the volatile threshold resistive device, the device CH2 will be turned on, and the slow sub-circuit will produce a periodic charging and discharging process, causing the voltage across the fast sub-circuit to also change periodically. The neuron will periodically switch between the two states of firing and refractory period, and the neuron will be in the burst firing mode.

[0024] Based on the above circuit design, the change in the resistance of the non-volatile threshold resistive device will affect the current balance of the capacitor in the slow sub-circuit, causing the capacitor of the slow sub-circuit to reach different voltage values, resulting in the neuron working in different firing modes. The non-volatile threshold resistive device uses a three-terminal device as the control unit, which has the following advantages in the circuit: (1) The three-terminal device has a wide resistance adjustment range and can effectively adjust the dynamic behavior of the slow sub-circuit; (2) The operating voltage of the non-volatile threshold resistive device matches the operating voltage of the neuron; (3) The resistance of the non-volatile threshold resistive device is erased and written through an independent port, making the design of the neuron circuit simpler. Using the characteristics of the non-volatile threshold resistive device, a reconfigurable neuron circuit was constructed.

[0025] Furthermore, the resistance value of the non-volatile threshold resistive switching device is determined according to the requirements of the neuron model.

[0026] Furthermore, the resistance value of the nonvolatile threshold resistive switching device affects the leakage current of the capacitor in the slow sub-circuit, and the charging current and the leakage current work together to determine the potential change of the capacitor.

[0027] Furthermore, the capacitor potentials of the fast subcircuit and the slow subcircuit directly determine the value of the neuron membrane potential. When the membrane potential reaches the firing threshold, the neuron is activated, and when the membrane potential is lower than the firing threshold, the neuron is at rest.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The neuron circuit designed by the present invention utilizes the coupling effect of fast and slow circuits to generate high-order pulse emission behaviors that simulate biological neurons, which are specifically manifested as fast peak emission, adaptive peak emission, refractory period emission, and cluster emission. At the same time, the switching of different emission modes is achieved by changing the resistance value of the non-volatile threshold resistive variable device through a control signal, which reflects the reconfigurability of the neuron circuit. Using the neurons described in the present invention as basic computing units to construct a neural computing system that simulates the human brain overcomes the technical problems of the single function and non-reconfigurability of the neuron circuit in the existing technology, such as the difficulty in chip integration, and is expected to achieve a more realistic, more efficient, and more intelligent computing system. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a structural schematic diagram of a reconfigurable brain-like neuron circuit based on a fast and slow circuit constructed in an embodiment of the present invention;

[0031] Figure 2 It is a reconfigurable pulse emission mode of a neuron circuit constructed by an embodiment of the present invention;

[0032] Figure 3 It is a dynamic behavior analysis of the neuron circuit constructed by the embodiment of the present invention;

[0033] Figure 4 It is an example of switching and controlling different pulse emission modes of a neuron circuit constructed according to an embodiment of the present invention;

[0034] Figure 5 is an indirect training SNN constructed by an embodiment of the present invention;

[0035] Figure 6 It is a directly trained SNN constructed by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0037] like Figure 1 As shown, a reconfigurable brain-like neuron circuit based on fast and slow circuits, the neuron is composed of a fast subcircuit and a slow subcircuit, the fast subcircuit is used to generate neuron pulses, the input end of the fast subcircuit is connected to the input resistor, the output end of the fast subcircuit is connected to the output resistor after passing through the slow subcircuit, and the voltage of the output resistor to the ground end is used as the output node of the neuron; the slow subcircuit is used to adjust the firing mode of the neuron. In this embodiment, the input resistor is R in , the output resistance is R load The threshold resistive switching device of the fast sub-circuit is CH1, the capacitor is C1, the threshold resistive switching device of the slow sub-circuit is CH2, the capacitor is C2, and the non-volatile threshold resistive switching device is NVTSM.

[0038] The fast subcircuit includes CH1 and C1. The input of CH1 is connected through R in Connected to the signal source, the output of CH1 is connected to the input of the slow subcircuit, one end of C1 is connected to the input of CH1, and the other end is grounded. The slow subcircuit includes CH2, NVTSM and C2, which are connected in parallel, with the input connected to the fast subcircuit and the output connected to R load .

[0039] Furthermore, the ratio of the capacitor C1 to the capacitor C2 is 1:10, and the dynamic change of the capacitance of the slow sub-circuit is slower than the dynamic change of the capacitance of the fast sub-circuit.

[0040] Furthermore, the threshold resistive switching device is a device with volatile characteristics. After the voltage applied to both ends is removed, the change in resistance value cannot be maintained. The same type of threshold resistive switching device is used in the fast sub-circuit and the slow sub-circuit.

[0041] Furthermore, the non-volatile threshold resistive switching device in the slow sub-circuit is a three-terminal device with non-volatile characteristics, its gate is separated as a control port, receives external voltage regulation to change the resistance of the device, and the source and drain are connected in parallel with the threshold resistive switching device and the capacitor in the slow sub-circuit as two terminals. Therefore, the dynamic equilibrium state of the slow sub-circuit can be regulated in a simple way.

[0042] In the present embodiment, in the same neuron circuit, the switching of different pulse emission modes of neurons can be realized only by adjusting the resistance value of NVTSM in the slow subcircuit, such as Figure 2As shown, when the resistance of NVTSM is low and does not exceed 2KΩ, the firing mode of the neuron is fast peak firing. When the resistance gradually increases to 5KΩ, the firing mode of the neuron changes to adaptive firing. When the resistance continues to increase to 10KΩ, the neuron will fire during the refractory period. As the resistance increases to 20KΩ, the neuron will exhibit a higher-order cluster firing behavior. To regulate the firing behavior of the above neurons, it is only necessary to change the resistance of NVTSM through an independent control port.

[0043] The reconfigurable circuit characteristics of this embodiment are characterized and verified in principle, such as Figure 3 As shown in the figure, through the electrical characterization of the dynamic behavior of the neuron circuit, the process of how different NVTSM resistance values ​​affect the neuron firing behavior by adjusting the C2 potential in the slow subcircuit is explained. As the resistance value of NVTSM changes from small to large, the intrinsic dynamic behavior of the circuit changes, which is mainly manifested in the gradual increase of the saturation potential of C2. The pulse modes of the neuron circuit are fast firing, adaptive firing, refractory firing and burst firing. When the saturation potential of C2 is small, the slow subcircuit has almost no effect on the fast subcircuit, the membrane potential of the neuron can always maintain rapid charging and discharging, and the neuron firing behavior is fast firing. When the saturation potential of C2 increases, the membrane potential of the neuron is limited by the slow subcircuit, the firing frequency gradually decreases, and the neuron firing behavior is adaptive firing. When the saturation potential of C2 continues to increase, after a short firing activity, the membrane potential of the neuron cannot reach the threshold, and the neuron firing behavior is refractory firing. When the saturation potential of C2 increases to the threshold of CH2, CH2 will have a threshold transition similar to CH1, causing the potential of C2 to charge and discharge periodically, resulting in the same periodic changes in the membrane potential, and the neuron will produce higher-order burst firing. Figure 3 (b) shows a comparison of the dynamic changes in C2 potential under adaptive firing and burst firing.

[0044] Furthermore, the saturation voltage of C2 is determined by the dynamic balance of the current in the slow subcircuit. When C2 reaches the saturation voltage, the input current of the slow subcircuit is equal to the leakage current through NVTSM and CH2, and the potential of C2 remains stable.

[0045] Furthermore, when the saturation voltage of C2 reaches the threshold of CH2 in the slow subcircuit, CH2 will suddenly change from high resistance to low resistance, and the leakage current through CH2 will suddenly increase, which is greater than the input current of the slow subcircuit, so that the potential of C2 will decrease. As the potential of C2 drops below a certain value, CH2 returns to high resistance, the input current of the slow subcircuit is greater than the leakage current, and the potential of C2 increases to the saturation voltage again; this is manifested as a periodic dynamic potential change.

[0046] Furthermore, the pulse emission mode switching of the neuron circuit is mainly affected by the change of NVTSM resistance. The input voltage change has a weaker effect on the mode switching, and mainly affects the pulse frequency and other emission behaviors under the same emission mode, such as Figure 4 As shown in (a).

[0047] Furthermore, the bursting behavior is affected by the input signal strength. As the input voltage value increases, the number of pulses in each cycle of the bursting behavior of the same neuron increases, such as Figure 4 As shown in (b).

[0048] In this embodiment, the advantages of the bursting behavior of neurons in the spiking neural network are verified by software simulation. First, the network performance improvement of indirect training SNN is verified. The bursting neurons are Figure 5 The VGG16 network structure shown in (a) is connected. Compared with the traditional integrate-and-release (IF) neurons, the Burst-based neural network has higher accuracy and lower inference latency. The results are shown in Figure 5 As shown in (b).

[0049] Furthermore, Burst neurons also significantly improve the performance of directly training SNNs. Figure 6 In the self-organizing network based on STDP rule, Burst neurons can accelerate the training of the network and improve the recognition accuracy after the neural network training. Among them, (a) is the structural diagram of the self-organizing network, and (b) is the network performance of different neuron models.

[0050] Furthermore, in this embodiment, the neural network simulation of VGG16 adopts a conversion method to convert the trained deep neural network into a pulse neural network, and uses IF neurons or Burst neurons to replace traditional neurons in the deep neural network, and compares the neural network performance of IF neurons and Burst neurons. The input data uses the CIFAR-10 dataset.

[0051] Furthermore, the self-organizing spiking neural network simulation in this embodiment adopts the STDP rule to directly train the weights, using LIF neurons and Burst neurons to perform neural network training and reasoning respectively. The neural network based on Burst neurons has a more efficient training process and a higher final recognition accuracy.

[0052] Furthermore, the STDP rule updates the synaptic connection based on the pulse firing time relationship between the presynaptic neuron and the postsynaptic neuron, which has more realistic biological learning characteristics. Bursting can complete the training of neural networks more effectively due to its more precise pulse firing behavior.

[0053] The neuron circuit design provided by the present invention utilizes a high-order fast and slow subcircuit to generate behaviors that simulate more realistic biological pulses. The switch threshold device is a volatile threshold resistive device, the mode control unit is a non-volatile threshold resistive device, and the resistors, capacitors, etc. are all mature commercial devices. By adjusting the resistance of the non-volatile threshold resistive device, the switching of the neuron pulse emission mode is achieved in a reconfigurable manner. The Burst neuron model is constructed using the cluster emission behavior of neurons, which can significantly improve the performance of the neural network in both the conversion and direct training of the pulse neural network, including recognition accuracy, training speed, and inference latency.

[0054] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A reconfigurable brain-like neuron circuit based on fast and slow circuits, characterized in that: It includes a fast subcircuit and a slow subcircuit. The fast and slow subcircuits are coupled to each other to determine the pulse emission mode. The fast sub-circuit includes a volatile threshold resistive switching device CH1 and a capacitor C1. The input end of the threshold resistive switching device CH1 is connected to an input resistor R in connected to a signal source, the output end of the threshold resistance switching device CH1 is connected to the input end of the slow sub-circuit; one end of the capacitor C1 is connected to the input end of the threshold resistance switching device CH1, and the other end is grounded; the threshold resistance switching device CH1 is used to simulate the voltage-gated ion channel of biological neurons to generate a pulse current; The slow sub-circuit includes a volatile threshold resistive switching device CH2, a capacitor C2 and a non-volatile threshold resistive switching device NVTSM connected in parallel; the slow sub-circuit input end is connected to the fast sub-circuit for receiving the pulse current of the fast sub-circuit, and the slow sub-circuit output end is connected to the output resistor R load The connection point serves as the pulse output port of the neuron; the resistance of the non-volatile threshold resistive switching device NVTSM is adjusted by controlling the external voltage, and the parallel resistance of the two resistive switching devices dynamically affects the charging and discharging process of the capacitor C2, thereby controlling the capacitor charging and discharging of the fast sub-circuit and the pulse emission behavior of the neuron.

2. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 1 is characterized in that: In the slow sub-circuit, when the resistance of the non-volatile threshold resistive switching device NVTSM changes from low resistance to high resistance, the impact on the fast sub-circuit also gradually changes, and the pulse firing behavior of neurons is fast peak firing, adaptive peak firing, refractory period firing and burst firing in sequence.

3. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 2 is characterized in that: The resistance of the non-volatile threshold resistive switching device NVTSM changes continuously under external voltage control, and the resistance state remains stable after the voltage is removed; according to the requirements of the pulse emission behavior, the non-volatile threshold resistive switching device NVTSM is adjusted to a certain resistance state.

4. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 1 is characterized in that: The volatile threshold resistive switching devices CH1 and CH2 are the same type of threshold resistive switching devices, whose resistance change is nonlinear and have threshold switching characteristics; when the capacitor voltage reaches the threshold, the resistance of the threshold resistive switching device suddenly changes from high resistance to low resistance, that is, the device switch is quickly opened; when the capacitor voltage is lower than the threshold, the resistance of the threshold resistive switching device suddenly changes from low resistance back to high resistance, that is, the device switch is quickly closed.

5. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 1 is characterized in that: The pulse output of the neuron circuit is the output resistance R load The voltage divider, output resistance R load One end is connected to the output end of the slow subcircuit and serves as an output port, and the other end is grounded.

6. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 1 is characterized in that: The difference between the capacitance potential of the fast subcircuit and the capacitance potential of the slow subcircuit is used as the membrane potential of the neuron.

7. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 1 is characterized in that: The capacitance ratio of the fast subcircuit to the slow subcircuit ranges from 0.01 to 1.

8. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 1 is characterized in that: The non-volatile threshold resistive switching device NVTSM comprises a gate, a source and a drain, wherein the source and the drain are used to form a parallel connection with the volatile threshold resistive switching device CH2 and the capacitor C2; the gate is used as a control port to receive external voltage control to change the resistance of the non-volatile threshold resistive switching device NVTSM.

9. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 8, characterized in that: When the capacitor potential of the slow sub-circuit is lower than the threshold of the volatile threshold resistive device, the device CH2 is not turned on, the slow sub-circuit receives the output current of the fast sub-circuit, and the capacitor potential of the slow sub-circuit gradually increases, causing the capacitor potential of the fast sub-circuit to also increase; as the NVSTM resistance value increases from small to large, the firing behavior of neurons is, in sequence, fast peak firing with no obvious change in firing frequency, adaptive peak firing with a firing frequency from high to low until a stable state, and refractory firing with a firing frequency from high to low until a refractory period state.

10. The reconfigurable brain-like neuron circuit based on fast and slow circuits according to claim 8, characterized in that: When the capacitor potential of the slow sub-circuit is higher than the threshold of the volatile threshold resistive device, the device CH2 is turned on, and the slow sub-circuit produces a periodic charging and discharging process, causing the voltage across the fast sub-circuit to also change periodically. The neuron periodically switches between the two states of firing and refractory period, and the neuron is in a burst firing mode.

Images