A logic gate based on neuromorphic devices

By designing logic gates based on neuromorphic devices, using membrane potential and distribution threshold to realize NAND gate and NAND gate logic, the fusion degree and power consumption problems of logic circuits in the neural network in the prior art are solved, and low power is achieved Consumable and flexible information processing.

CN114386587BActive Publication Date: 2024-11-15CETHIK GRP
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202111650068.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-30
Publication Date
2024-11-15
Estimated Expiration
2041-12-30

AI Technical Summary

Technical Problem

The existing CMOS logic circuits have low fusion, poor applicability, and high power consumption, making it difficult to meet the needs of brain-like computing.

Method used

A logic gate based on neuromorphic devices is designed, including the first neuron, the second neuron, the third neuron and the fourth neuron. The NAND gate logic is realized through membrane potential and the distribution threshold, and the NAND gate logic is obtained. Low power consumption and refractory performance, with sparse response characteristics.

Benefits of technology

It realizes efficient integration with neural networks, reduces power consumption, and realizes logical conversion by simply adjusting the distribution threshold. It has flexible and intelligent information processing and computing modes, and is good for applicability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114386587B_ABST
    Figure CN114386587B_ABST
Patent Text Reader

Abstract

The present invention discloses a logic gate based on a neuromorphic device, including a first neuron, a second neuron, a third neuron, a fourth neuron and a fifth neuron, wherein the first neuron is connected to the third neuron, the second neuron is connected to the fourth neuron, and the third neuron and the fourth neuron are both connected to the fifth neuron; the first neuron and the second neuron, when the membrane potential is greater than or equal to the first emission threshold, output a high level, otherwise, output a low level; the third neuron and the fourth neuron have a basic signal, the membrane potential is the sum of the basic signal and the input signal, and when it is greater than or equal to the second emission threshold, output a high level, otherwise, output a low level; the fifth neuron, the membrane potential is the sum of the two input signals, when it is greater than or equal to the third emission threshold, output a high level, otherwise, output a low level. The present invention can realize the conversion of different logic functions, has low power consumption and refractory period performance, and has good integration and applicability with neural networks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of neuromorphic circuits, and in particular relates to a logic gate based on a neuromorphic device. Background Art

[0002] In digital circuits, the so-called "gate" refers to a circuit that can realize basic logical relationships. Logic gates can be constructed with discrete components such as resistors, capacitors, diodes, and triodes to form discrete component gates. All devices and connecting wires of the gate circuit can also be made on the same semiconductor substrate to form an integrated logic gate circuit. In the prior art, logic gate circuits are usually implemented using CMOS logic, such as Figure 1 As shown, the logical expression of the NOR gate is Among them, Y represents the output terminal (Output), A and B are the input terminals respectively, and any input terminal is at a high level (logic 1), and the output is at a low level (logic 0). Figure 2 The following is the truth table of the NOR gate. Figure 3 As shown, the logical expression of the NAND gate is When both input terminals are at high level (logic 1), the output terminal is at low level (logic 0); if at least one of the input terminals is at low level (logic 0), the output terminal is at high level (logic 1). Figure 4 Shown is the truth table for the NAND gate.

[0003] With the development of related technologies such as artificial intelligence and neural computing, brain-like computing has attracted attention, which has given rise to spiking neural networks (SNNs). Although traditional deep neural networks are also inspired by the working mechanism of the human brain, they are far from the real situation. They transmit analog quantities, while in the human brain, information is transmitted in the form of pulses. It can be seen that spiking neural networks are closer to the working mechanism of the human brain than traditional deep neural networks. In addition, the architectural bottleneck of the separation of von Neumann computing and storage is becoming more and more significant, and the driving method of increasing density and thus improving computing performance by relying on Moore's Law physical miniaturization is becoming more and more difficult. Unlike deep learning dedicated processors that are still based on the von Neumann processor architecture, neuromorphic engineering is committed to building a more brain-like hardware computing architecture.

[0004] Similar to biological neural networks, neuromorphic primitives are composed of neuromorphic dendrites, cell bodies, and axons, and neuromorphic synaptic connections form neuromorphic networks. Information is stored by configuring the connection topology of the routing network and the neuromorphic synaptic connection weights, and information processing is achieved by combining the calculation of neuron modules with the routing transmission of signals. Neuromorphic systems mainly consist of three parts: neuron calculation, synaptic weight storage, and routing communication. Since brain neural networks use spikes for information encoding and transmission, such systems basically use SNNs models.

[0005] The pulse neural network uses spike pulses to encode information. The spike signal is a binary signal. 1 represents that the neuron is activated (generating an event) and transmitted to the neurons connected to its back end, that is, a pulse event is sent to the back-end neurons; 0 represents that the neuron is not activated and has no effect on the back-end neurons, that is, the above-mentioned pulse event is not generated. This event-driven approach (working only after receiving an event) can achieve very high energy efficiency, and it is also the starting point for many neuromorphic chips to use the SNNs model as the basic model. Therefore, in view of the fact that the COMS logic circuit in the prior art has low integration with the neural network, poor applicability, and high power consumption, the present application proposes a logic gate based on neuromorphic devices. Summary of the invention

[0006] The purpose of the present invention is to address the above-mentioned problems and propose a logic gate based on a neuromorphic device, which has low power consumption and refractory period performance, sparse response characteristics, and improved integration and applicability with neural networks.

[0007] To achieve the above object, the technical solution adopted by the present invention is:

[0008] The present invention proposes a logic gate based on a neuromorphic device, comprising a first neuron A, a second neuron B, a third neuron C, a fourth neuron D and a fifth neuron Y, wherein the first neuron A is connected to the third neuron C, the second neuron B is connected to the fourth neuron D, the third neuron C and the fourth neuron D are both connected to the fifth neuron Y, the first neuron A and the second neuron B are signal input terminals, and the fifth neuron Y is a signal output terminal, wherein:

[0009] The first neuron A and the second neuron B are both first-type neurons. When receiving a signal, when the membrane potential is greater than or equal to the first emission threshold, they output a high level, and when the membrane potential is less than the first emission threshold, they output a low level;

[0010] The third neuron C and the fourth neuron D are both second-type neurons, which have a basic signal, and the membrane potential is the sum of the basic signal and the input signal from the corresponding neuron. When the membrane potential is greater than or equal to the second emission threshold, the output is a high level, and when the membrane potential is less than the second emission threshold, the output is a low level;

[0011] The fifth neuron Y is a third type neuron, whose membrane potential is the sum of the input signals from the third neuron C and the fourth neuron D. When the membrane potential is greater than or equal to the third emission threshold, it outputs a high level, and when the membrane potential is less than the third emission threshold, it outputs a low level.

[0012] Preferably, the first neuron A and the second neuron B are both inhibitory neurons, which inhibit the activation of the neurons connected thereto by issuing inhibitory signals when outputting a high level, and the third neuron C and the fourth neuron D have a basic signal that is constantly 1.

[0013] Preferably, the logic gate based on the neuromorphic device is a NOR gate and performs the following operations:

[0014] When the first neuron A and the second neuron B both output a high level, the third neuron C and the fourth neuron D are inhibited, the membrane potential is less than the second firing threshold, and the output is a low level. The membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level.

[0015] When the first neuron A outputs a high level and the second neuron B outputs a low level, the third neuron C is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level. The fourth neuron D is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level. The membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level.

[0016] When the first neuron A outputs a low level and the second neuron B outputs a high level, the third neuron C is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level, the fourth neuron D is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level, the membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level;

[0017] When the first neuron A and the second neuron B both output a low level, the third neuron C and the fourth neuron D are not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level. The membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is a high level.

[0018] Preferably, the first firing threshold and the second firing threshold are both 1, the third firing threshold is 2, and the reset potentials of the third neuron C, the fourth neuron D and the fifth neuron Y are all 0.

[0019] Preferably, the logic gate based on the neuromorphic device is a NAND gate, and performs the following operations:

[0020] When the first neuron A and the second neuron B both output a high level, the third neuron C and the fourth neuron D are inhibited, the membrane potential is less than the second firing threshold, and the output is a low level. The membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level.

[0021] When the first neuron A outputs a high level and the second neuron B outputs a low level, the third neuron C is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level. The fourth neuron D is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level. The membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is a high level.

[0022] When the first neuron A outputs a low level, and the second neuron B outputs a high level, the third neuron C is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is high level, the fourth neuron D is inhibited, the membrane potential is less than the second firing threshold, and the output is low level, and the membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is high level;

[0023] When the first neuron A and the second neuron B both output a low level, the third neuron C and the fourth neuron D are not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level. The membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is a high level.

[0024] Preferably, the first firing threshold, the second firing threshold and the third firing threshold are all 1, and the reset potentials of the third neuron C, the fourth neuron D and the fifth neuron Y are all 0.

[0025] Preferably, after the fifth neuron Y outputs a high level or a low level, the logic gate based on the neuromorphic device does not respond within a preset time threshold, and the preset time threshold is the refractory period.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: the present application implements NOR gate or NAND gate logic based on event-driven, that is, it will not start until the first neuron and the second neuron both receive signals, which greatly reduces power consumption, and the NOR gate and NAND gate logic conversion can be achieved by simply adjusting the emission threshold; when implementing the NOR gate or NAND gate logic, it also has a refractory period, that is, within a period of time after the fifth neuron outputs, even if the first neuron and the second neuron both receive signals, they still will not start, and has a sparse response characteristic, which further reduces power consumption; and the logic gate function is realized by neuromorphic devices, which can be better integrated with various neural networks, and can realize more flexible and intelligent information processing and computing modes, and has good applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a circuit diagram of a NOR gate in the prior art;

[0028] Figure 2 It is the truth diagram of the NOR gate in the prior art;

[0029] Figure 3 It is a NAND gate circuit diagram of the prior art;

[0030] Figure 4 It is the truth diagram of NAND gate in the prior art;

[0031] Figure 5 The figure is a logic gate circuit diagram of the neuromorphic device according to the present invention;

[0032] Figure 6 It is a truth diagram of a NOR gate based on a neuromorphic device of the present invention;

[0033] Figure 7 This is the truth diagram of the NAND gate based on the neuromorphic device of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0035] It should be noted that, unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application.

[0036] like Figure 5 As shown, a logic gate based on a neuromorphic device includes a first neuron A, a second neuron B, a third neuron C, a fourth neuron D and a fifth neuron Y. The first neuron A is connected to the third neuron C, the second neuron B is connected to the fourth neuron D, the third neuron C and the fourth neuron D are both connected to the fifth neuron Y, the first neuron A and the second neuron B are signal input terminals, and the fifth neuron Y is a signal output terminal, wherein:

[0037] The first neuron A and the second neuron B are both first-type neurons. When receiving a signal, when the membrane potential is greater than or equal to the first emission threshold, they output a high level, and when the membrane potential is less than the first emission threshold, they output a low level;

[0038] The third neuron C and the fourth neuron D are both second-type neurons, which have a basic signal, and the membrane potential is the sum of the basic signal and the input signal from the corresponding neuron. When the membrane potential is greater than or equal to the second emission threshold, the output is a high level, and when the membrane potential is less than the second emission threshold, the output is a low level;

[0039] The fifth neuron Y is a third type neuron, whose membrane potential is the sum of the input signals from the third neuron C and the fourth neuron D. When the membrane potential is greater than or equal to the third emission threshold, it outputs a high level, and when the membrane potential is less than the third emission threshold, it outputs a low level.

[0040] like Figure 5 As shown, the first neuron A and the second neuron B are both signal input terminals, and the fifth neuron Y is a signal output terminal. Each terminal can be a neuron of a neuromorphic device, and the membrane potential is the current state parameter of the neuron.

[0041] Among them, the first neuron A is the predecessor neuron of the third neuron C, and the third neuron C is the successor neuron of the first neuron A. The membrane potential of each neuron changes when receiving a signal. When the membrane potential of the predecessor neuron is greater than or equal to the first emission threshold, the predecessor neuron is activated (generating a pulse event), outputs a high level, and transmits the generated pulse event to its successor neuron. When the membrane potential of the predecessor neuron is less than the first emission threshold, the predecessor neuron is not activated and outputs a low level, which has no effect on its successor neuron, and can be understood as 0 transmitted to the successor neuron.

[0042] Similarly, the second neuron B is the predecessor neuron of the fourth neuron D, and the fourth neuron D is the successor neuron of the second neuron B. The third neuron C and the fourth neuron D are the predecessor neurons of the fifth neuron Y, and the fifth neuron Y is the successor neuron of the third neuron C and the fourth neuron D. The third neuron C and the fourth neuron D have a basic signal, and the membrane potential is the sum of the basic signal and the input signal from the predecessor neuron. The input signal of the third neuron C and the fourth neuron D is the pulse event generated by the corresponding predecessor neuron. The input signal of the fifth neuron Y is the signal transmitted by the predecessor neurons (the third neuron C and the fourth neuron D).

[0043] The present application can realize logical functions based on event-driven, such as starting only when the first neuron A and the second neuron B both receive signals, that is, after receiving the signal, the membrane potential of the first neuron A and the second neuron B changes, when the membrane potential is greater than or equal to the first emission threshold, the neuron is activated and outputs a high level, when the membrane potential is less than the first emission threshold, the neuron is not activated and outputs a low level, and when at least one of the first neuron A and the second neuron B does not receive a signal, the logic gate circuit will not start the calculation, which can greatly reduce power consumption.

[0044] The present application can be set to realize the logic function of the NOR gate according to the requirements, such as by combining various types of neurons (such as inhibitory neurons, non-inhibitory neurons), setting basic signals, setting release conditions (such as release threshold, membrane potential reset potential), etc. to realize the function of the logic gate. The following is an explanation through specific embodiments.

[0045] Embodiment 1:

[0046] In this embodiment, the first neuron A and the second neuron B are both inhibitory neurons, which inhibit the activation of the neurons connected thereto by issuing inhibitory signals when outputting a high level, and the third neuron C and the fourth neuron D have a basic signal that is always 1.

[0047] The logic gates based on neuromorphic devices implement NOR gate logic and perform the following operations:

[0048] When the first neuron A and the second neuron B both output a high level, the third neuron C and the fourth neuron D are inhibited, the membrane potential is less than the second firing threshold, and the output is a low level. The membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level.

[0049] When the first neuron A outputs a high level and the second neuron B outputs a low level, the third neuron C is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level. The fourth neuron D is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level. The membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level.

[0050] When the first neuron A outputs a low level and the second neuron B outputs a high level, the third neuron C is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level, the fourth neuron D is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level, the membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level;

[0051] When the first neuron A and the second neuron B both output a low level, the third neuron C and the fourth neuron D are not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level. The membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is a high level.

[0052] Among them, the first neuron A and the second neuron B are inhibitory neurons. Inhibitory neurons can be understood as when they are fired, they will inhibit the activation of their successor neurons. If the third neuron C and the fourth neuron D have a basic signal of constant 1, in order to inhibit the activation of their successor neurons, the corresponding inhibitory neurons will transmit the generated pulse events to their successor neurons, and the pulse events can be regarded as inhibitory signals -1. The truth value diagram of the NOR gate is as follows Figure 6 shown.

[0053] The following NOR gate logic is realized by setting the firing conditions of the third neuron C, the fourth neuron D and the fifth neuron Y. The embodiment is exemplary, and those skilled in the art can adjust it according to actual needs. It will not be repeated here, and it is not used as a limitation on the firing threshold and the membrane potential reset potential. The specific details are as follows:

[0054] In one embodiment, the first firing threshold and the second firing threshold are both 1, the third firing threshold is 2, and the reset potentials of the third neuron C, the fourth neuron D, and the fifth neuron Y are all 0.

[0055] This embodiment belongs to the case where the membrane potential attenuation characteristic is not reflected. For the third neuron C, the second firing threshold can be set to 1, the reset potential can be set to 0, and the third neuron C has a basic signal that is always 1. When the predecessor neuron A of the third neuron C does not fire, the third neuron C is not inhibited, because there is a signal that is always 1, which is equal to the second firing threshold, then the third neuron C outputs a high level 1 and resets the membrane potential to 0. When the predecessor neuron A of the third neuron C fires, the third neuron C is inhibited (it can be understood that the current membrane potential of the third neuron C is determined according to the basic signal of always 1 and the inhibitory signal -1 from the predecessor neuron A), for example, the membrane potential is 0, which is less than the second firing threshold 1, and the third neuron C outputs a low level 0.

[0056] Similarly, for the fourth neuron D, the second firing threshold can be set to 1 and the reset potential can be set to 0. And the fourth neuron D has a basic signal that is always 1. When the predecessor neuron B of the fourth neuron D does not fire, the fourth neuron D is not inhibited, because there is a basic signal that is always 1, which is equal to the second firing threshold, then the fourth neuron D outputs a high level 1, and the reset potential is 0. When the predecessor neuron B of the fourth neuron D fires, the fourth neuron D is inhibited (it can be understood that the current membrane potential of the fourth neuron D is determined according to the basic signal of always 1 and the inhibitory signal -1 from the predecessor neuron B), for example, the membrane potential is 0, which is less than the firing threshold 1, and the fourth neuron D outputs a low level 0.

[0057] For the fifth neuron Y, the third firing threshold is set to 2 and the reset potential is set to 0. When the third neuron C and the fourth neuron D both fire pulses (i.e., output a high level), the current membrane potential of the fifth neuron Y is 2, which is equal to the third firing threshold, and the output is a high level 1. When at least one of the third neuron C and the fourth neuron D does not fire a pulse (i.e., output a high level), the current membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level 0.

[0058] In one embodiment, after the fifth neuron Y outputs a high level or a low level, the logic gate based on the neuromorphic device does not respond within a preset time threshold, and the preset time threshold is the refractory period to achieve the refractory period.

[0059] Wherein, the non-response of the logic gate based on the neuromorphic device within the preset time threshold can be achieved by controlling at least one neuron not to perform the membrane potential update operation. For example, when the logic gate of the neuromorphic device is in the refractory period, the fifth neuron Y can be controlled not to perform the membrane potential update operation, or the first neuron A and the second neuron B can be controlled not to perform the membrane potential update operation. It should be understood that if the first neuron A and the second neuron B do not perform the membrane potential update operation, the first neuron A and the second neuron B do not need to perform the firing judgment, and the third neuron C, the fourth neuron D and the fifth neuron Y do not need to perform the membrane potential update operation and the firing judgment. Wherein, the firing judgment may include operations such as comparing the current membrane potential with the firing threshold.

[0060] Among them, controlling the first neuron A and the second neuron B not to perform membrane potential update operation can be directly controlling the first neuron A and the second neuron B to suspend the membrane potential update operation (for example, controlling the membrane potential update module to suspend operation), or it can be disconnecting the first neuron A and the second neuron B from the input signal, such as can be achieved through a switch circuit in the prior art or designed according to actual needs, that is, during the refractory period, the first neuron A and the second neuron B do not receive the input signal, and thus do not perform the membrane potential update operation.

[0061] When the fifth neuron Y outputs a high level or a low level, the logic gate based on the neuromorphic device is controlled not to respond within a preset time threshold, thereby reducing the amount of membrane potential update calculations and emission judgment, which helps to reduce power consumption. It should be noted that the specific time of the refractory period can be set according to actual needs.

[0062] Embodiment 2:

[0063] In this embodiment, the first neuron A and the second neuron B are both inhibitory neurons, which inhibit the activation of the neurons connected thereto by issuing inhibitory signals when outputting a high level, and the third neuron C and the fourth neuron D have a basic signal that is always 1.

[0064] The logic gates based on neuromorphic devices implement NAND gate logic and perform the following operations:

[0065] When the first neuron A and the second neuron B both output a high level, the third neuron C and the fourth neuron D are inhibited, the membrane potential is less than the second firing threshold, and the output is a low level. The membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level.

[0066] When the first neuron A outputs a high level and the second neuron B outputs a low level, the third neuron C is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level. The fourth neuron D is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level. The membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is a high level.

[0067] When the first neuron A outputs a low level, and the second neuron B outputs a high level, the third neuron C is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is high level, the fourth neuron D is inhibited, the membrane potential is less than the second firing threshold, and the output is low level, and the membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is high level;

[0068] When the first neuron A and the second neuron B both output a low level, the third neuron C and the fourth neuron D are not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level. The membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is a high level.

[0069] Among them, the first neuron A and the second neuron B are inhibitory neurons. Inhibitory neurons can be understood as when they are fired, they will inhibit the activation of their successor neurons. If the third neuron C and the fourth neuron D have a basic signal of constant 1, in order to inhibit the activation of their successor neurons, the corresponding inhibitory neurons will transmit the generated pulse events to their successor neurons, and the pulse events can be regarded as inhibitory signals -1. The truth value diagram of the NAND gate is as follows Figure 7 shown.

[0070] The following NOR gate logic is realized by setting the firing conditions of the third neuron C, the fourth neuron D and the fifth neuron Y. The embodiment is exemplary, and those skilled in the art can adjust it according to actual needs. It will not be repeated here, and it is not used as a limitation on the firing threshold and the membrane potential reset potential. The specific details are as follows:

[0071] In one embodiment, the first firing threshold, the second firing threshold, and the third firing threshold are all 1, and the reset potentials of the third neuron C, the fourth neuron D, and the fifth neuron Y are all 0.

[0072] This embodiment belongs to the case where the membrane potential attenuation characteristic is not reflected. For the third neuron C, the second firing threshold can be set to 1, the reset potential can be set to 0, and the third neuron C has a basic signal that is always 1. When the predecessor neuron A of the third neuron C does not fire, the third neuron C is not inhibited, because there is a signal that is always 1, which is equal to the second firing threshold, then the third neuron C outputs a high level 1 and resets the membrane potential to 0. When the predecessor neuron A of the third neuron C fires, the third neuron C is inhibited (it can be understood that the current membrane potential of the third neuron C is determined according to the basic signal of always 1 and the inhibitory signal -1 from the predecessor neuron A), for example, the membrane potential is 0, which is less than the second firing threshold 1, and the third neuron C outputs a low level 0.

[0073] Similarly, for the fourth neuron D, the second firing threshold can be set to 1 and the reset potential can be set to 0. And the fourth neuron D has a basic signal that is always 1. When the predecessor neuron B of the fourth neuron D does not fire, the fourth neuron D is not inhibited, because there is a basic signal that is always 1, which is equal to the second firing threshold, then the fourth neuron D outputs a high level 1, and the reset potential is 0. When the predecessor neuron B of the fourth neuron D fires, the fourth neuron D is inhibited (it can be understood that the current membrane potential of the fourth neuron D is determined according to the basic signal of always 1 and the inhibitory signal -1 from the predecessor neuron B), for example, the membrane potential is 0, which is less than the firing threshold 1, and the fourth neuron D outputs a low level 0.

[0074] For the fifth neuron Y, the third firing threshold is set to 1 and the reset potential is set to 0. When at least one of the third neuron C and the fourth neuron D fires a pulse (i.e., outputs a high level), the current membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and outputs a high level 1. When neither the third neuron C nor the fourth neuron D fires a pulse (i.e., outputs a high level), the current membrane potential of the fifth neuron Y is less than the third firing threshold, and outputs a low level 0.

[0075] In one embodiment, after the fifth neuron Y outputs a high level or a low level, the logic gate based on the neuromorphic device does not respond within a preset time threshold, and the preset time threshold is the refractory period.

[0076] Wherein, the non-response of the logic gate based on the neuromorphic device within the preset time threshold can be achieved by controlling at least one neuron not to perform the membrane potential update operation. For example, when the logic gate of the neuromorphic device is in the refractory period, the fifth neuron Y can be controlled not to perform the membrane potential update operation, or the first neuron A and the second neuron B can be controlled not to perform the membrane potential update operation. It should be understood that if the first neuron A and the second neuron B do not perform the membrane potential update operation, the first neuron A and the second neuron B do not need to perform the firing judgment, and the third neuron C, the fourth neuron D and the fifth neuron Y do not need to perform the membrane potential update operation and the firing judgment. Wherein, the firing judgment may include operations such as comparing the current membrane potential with the firing threshold.

[0077] Among them, controlling the first neuron A and the second neuron B not to perform membrane potential update operation can be directly controlling the first neuron A and the second neuron B to suspend the membrane potential update operation (for example, controlling the membrane potential update module to suspend operation), or it can be disconnecting the first neuron A and the second neuron B from the input signal, such as can be achieved through a switch circuit in the prior art or designed according to actual needs, that is, during the refractory period, the first neuron A and the second neuron B do not receive the input signal, and thus do not perform the membrane potential update operation.

[0078] When the fifth neuron Y outputs a high level or a low level, the logic gate based on the neuromorphic device is controlled not to respond within a preset time threshold, thereby reducing the amount of membrane potential update calculations and emission judgment, which helps to reduce power consumption. It should be noted that the specific time of the refractory period can be set according to actual needs.

[0079] This application implements the logic of NOR gate and NAND gate based on event-driven, that is, it will start only when the first neuron A and the second neuron B receive the signal, which greatly reduces the power consumption, and the logic conversion of NOR gate and NAND gate can be realized by simply adjusting the emission threshold; and when implementing the logic of NOR gate or NAND gate, it also has a refractory period, that is, within a period of time after the output of the fifth neuron Y, even if the first neuron A and the second neuron B receive the signal, they will not start, with sparse response characteristics, further reducing power consumption; and through neuromorphic devices to realize the logic gate function, it can better integrate with various neural networks, and can realize more flexible and intelligent information processing and computing modes. Therefore, different from the gate circuit in the prior art, this application is a gate logic with sparse response characteristics, which can greatly improve the integration and applicability with neural networks and reduce power consumption.

[0080] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0081] The above-described embodiments only express the more specific and detailed embodiments described in this application, but they cannot be understood as limiting the scope of the patent application. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of this application, which all belong to the protection scope of this application. Therefore, the protection scope of the patent application shall be based on the attached claims.

Claims

1. A logic gate based on a neuromorphic device, characterized in that: The logic gate based on the neuromorphic device includes a first neuron A, a second neuron B, a third neuron C, a fourth neuron D and a fifth neuron Y, wherein the first neuron A is connected to the third neuron C, the second neuron B is connected to the fourth neuron D, the third neuron C and the fourth neuron D are both connected to the fifth neuron Y, the first neuron A and the second neuron B are signal input terminals, and the fifth neuron Y is a signal output terminal, wherein: The first neuron A and the second neuron B are both first-type neurons, which, when receiving a signal, output a high level when the membrane potential is greater than or equal to the first emission threshold, and output a low level when the membrane potential is less than the first emission threshold; The third neuron C and the fourth neuron D are both second-type neurons, which have a basic signal, and the membrane potential is the sum of the basic signal and the input signal from the corresponding neuron. When the membrane potential is greater than or equal to the second emission threshold, a high level is output, and when the membrane potential is less than the second emission threshold, a low level is output; The fifth neuron Y is a third type of neuron, whose membrane potential is the sum of the input signals from the third neuron C and the fourth neuron D. When the membrane potential is greater than or equal to the third emission threshold, it outputs a high level, and when the membrane potential is less than the third emission threshold, it outputs a low level; The first neuron A and the second neuron B are both inhibitory neurons, which inhibit the activation of the neurons connected thereto by issuing inhibitory signals when outputting a high level, and the third neuron C and the fourth neuron D have a basic signal that is always 1; The logic gate based on the neuromorphic device is a NOR gate and performs the following operations: When the first neuron A and the second neuron B both output a high level, the third neuron C and the fourth neuron D are inhibited, the membrane potential is less than the second firing threshold, and the output is a low level, and the membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level; When the first neuron A outputs a high level and the second neuron B outputs a low level, the third neuron C is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level, the fourth neuron D is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level, and the membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level; When the first neuron A outputs a low level and the second neuron B outputs a high level, the third neuron C is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level, the fourth neuron D is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level, and the membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level; When the first neuron A and the second neuron B both output a low level, the third neuron C and the fourth neuron D are not inhibited, the membrane potential is greater than or equal to the second firing threshold, and a high level is output; the membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and a high level is output; The logic gate based on the neuromorphic device is a NAND gate and performs the following operations: When the first neuron A and the second neuron B both output a high level, the third neuron C and the fourth neuron D are inhibited, the membrane potential is less than the second firing threshold, and the output is a low level, and the membrane potential of the fifth neuron Y is less than the third firing threshold, and the output is a low level; When the first neuron A outputs a high level and the second neuron B outputs a low level, the third neuron C is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level, the fourth neuron D is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level, the membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is a high level; When the first neuron A outputs a low level and the second neuron B outputs a high level, the third neuron C is not inhibited, the membrane potential is greater than or equal to the second firing threshold, and the output is a high level, the fourth neuron D is inhibited, the membrane potential is less than the second firing threshold, and the output is a low level, and the membrane potential of the fifth neuron Y is greater than or equal to the third firing threshold, and the output is a high level; When the first neuron A and the second neuron B both output a low level, the third neuron C and the fourth neuron D are not inhibited, the membrane potential is greater than or equal to the second emission threshold, and a high level is output; the membrane potential of the fifth neuron Y is greater than or equal to the third emission threshold, and a high level is output.

2. The neuromorphic device-based logic gate of claim 1, wherein: The first firing threshold and the second firing threshold are both 1, the third firing threshold is 2, and the reset potentials of the third neuron C, the fourth neuron D, and the fifth neuron Y are all 0.

3. The neuromorphic device-based logic gate of claim 1, wherein: The first firing threshold, the second firing threshold and the third firing threshold are all 1, and the reset potentials of the third neuron C, the fourth neuron D and the fifth neuron Y are all 0.

4. The neuromorphic device-based logic gate of claim 1, wherein: After the fifth neuron Y outputs a high level or a low level, the logic gate based on the neuromorphic device does not respond within a preset time threshold, and the preset time threshold is the refractory period.

Citation Information

Patent Citations

  • Recognition apparatus based on neural network and method of training neural network

    CN106997485A

  • Ultra-large-scale integrated visual processing system

    CN107545209A

  • Full-digital pulse neural network hardware system and method based on STDP rule

    CN113408714A

  • Integrate and fire electronic neurons

    US20120150781A1