Due to the useful properties of nonvolatile,memory and nanoscale,memristors have prospective promising applications in artificial networks,pattern recognition and signal processing.This paper exploits the learning rule of the memristive network based on spiking timing dependent plasticity(STDP)and uses the genetic algorithms with self-adaptation and variable topologies,which allows the number of hidden neurons,connection weights,and connectivity pattern to change self-adaptably.Three memristor models are respectively used as the synapse in the network,including HP linear memristor,non-linear memristor and threshold memristor.The comparison of the performance of the three memristive neural networks is presented,and the hybrid memristive networks' learning effects are analyzed.

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.