脉冲神经网络的忆阻器突触联想学习电路分析  

Associative learning of memristive synapses circuits based on spiking neural networks

作　　者：李传东[1] 葛均辉[1] 田园[1] 

机构地区：[1]重庆大学计算机学院,重庆400044

出　　处：《重庆大学学报（自然科学版）》2014年第7期115-124,共10页Journal of Chongqing University

基　　金：国家自然科学基金资助项目(61374078)

摘　　要：忆阻器是具有动态特性的电阻,阻值可依赖于激励电压来变化,具有类似于生物神经突触连接强度的特性,可用来存储突触权值。在此基础上为实现忆阻器突触电路的学习功能,建立了"整合激发"型神经元SPICE仿真电路,修改了原始神经元电路结构,并对电路的脉冲信号产生过程进行了SPICE仿真。结合MOS管及忆阻器的特性重新设计了神经元突触电路结构,使突触电路更符合真实生物神经突触特征。在应用此设计的基础上,实现了2个神经元所构成神经网络之间类似于Hebbian学习的平均激发率学习规则。并且在基于多个神经元的神经网络的基础上完成了Pavlov实验,证明了此神经系统结构设计在联想学习方面的可用性。Memristor is a resistance which has dynamic characteristics,which can be changed with respect to the excitation voltage and has similar characteristics of biological synaptic. So values of the synaptic weight can be stored by memristors. An ＂integrate-and-fire＂ neurons SPICE simulation circuits are established for the realization of the memristive synapse＇s learning function, the original neuron circuits are modified, and the process of generating spiking activity is demonstrated via SPICE simulations. Combined with the characteristics of MOS and memristor the synapse circuit of neurons are redesigned,it makes the synaptic circuitry more coincident with the real biological neural which is similar to the Habbian learning and includes two experiment including multiple neurons is realized, which available in associative learning. synapses. The average excitation rate learning, neurons, has been realized. In addition, the Pavlov proves that this structure of nervous system is available in associative learning.

关 键 词：忆阻器 神经元电路 SPICE仿真 Hebbian学习 联想学习 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]