基于Temporal rule的忆阻神经网络电路

计算机工程与科学

基于Temporal rule的忆阻神经网络电路

黄成龙，郝栋栋，方粮

（国防科技大学计算机学院，湖南长沙 410073）

收稿日期:2018-08-15 修回日期:2018-10-25 出版日期:2019-03-25 发布日期:2019-03-25
基金资助:
国家自然科学基金（61332003）

A memristor neural network circuit based on temporal rule

HUANG Chenglong，HAO Dongdong,FANG Liang

（School of Computer,National University of Defense Technology,Changsha 410073,China）

Received:2018-08-15 Revised:2018-10-25 Online:2019-03-25 Published:2019-03-25

摘要/Abstract

摘要：

忆阻器是一种动态特性的电阻，其阻值可以根据外场的变化而变化，并且在外场撤掉后能够保持原来的阻值，具有类似于生物神经突触连接强度的特性，可以用来存储突触权值。在此基础上，为了实现基于Temporal rule对IRIS数据集识别学习的功能，建立了以桥式忆阻器为突触的神经网络SPICE仿真电路。采用单个脉冲的编码方式，脉冲的时刻代表着数据信息，该神经网络电路由48个脉冲输入端口、144个突触、3个输出端口组成。基于Temporal rule学习规则对突触的权值修改，通过仿真该神经网络电路对IRIS数据集的分类正确率最高能达到93.33%,表明了此神经系统结构设计在类脑脉冲神经网络中的可用性。

关键词: 忆阻器, Temporal rule, 神经网络电路, 桥式忆阻器突触

Abstract:

The memristor is the resistor with dynamic characteristics. Its resistance value can be changed according to the variation of the external field, and it can maintain the original resistance value after the external field is removed. Memristors can be used to store synaptic weights thanks to their similar characteristics to the connection strength of biological synapses. In order to realize the function of recognition and learning of the IRIS dataset based on the temporal rule, we design a SPICE simulation circuit of neural networks which takes the bridge memristor as synapses. The circuit uses the single pulse encoding method, and the time of the pulse represents data information. The neural network circuit consists of 48 pulse input ports, 144 synapses and three output ports. The synaptic weights are modified based on the learning rules of the temporal rule. Simulation results show that the classification accuracy on the IRIS dataset can reach 93.33%, which proves that the proposed neural network circuit can be used in brain-like pulse neural networks.

Key words: memristor, temporal rule, neural network circuit, memristor-based synapse

黄成龙，郝栋栋，方粮. 基于Temporal rule的忆阻神经网络电路[J]. 计算机工程与科学.

HUANG Chenglong，HAO Dongdong,FANG Liang. A memristor neural network circuit based on temporal rule[J]. Computer Engineering & Science.

[1]	胡钇宏, 马德胜, 许诺, 王文清, 黄成龙, 方粮. 面向三维忆阻阵列的状态逻辑计算[J]. 计算机工程与科学, 2023, 45(03): 381-389.
[2]	邓亚峰, 魏子健, 王栋. 忆阻器阻变机制及其材料研究进展[J]. 计算机工程与科学, 2022, 44(01): 36-47.
[3]	摆玉龙, 潘星宇, 段济开, 杨阳. 基于双曲正弦函数的四翼忆阻混沌系统及其FPGA实现[J]. 计算机工程与科学, 2021, 43(10): 1744-1749.
[4]	方旭东, 吴俊杰. 基于忆阻器的计算存储融合体系结构研究进展[J]. 计算机工程与科学, 2020, 42(11): 1929-1940.
[5]	张荣芬,邓朝勇,傅兴华. 可记忆电阻器及其应用展望综述[J]. J4, 2011, 33(2): 80-85.