The Neural Network Based On Memristor Synapse And It's Application

The memristor is a novel circuit element with memory,which has favorable application in the fields of Resistive random access memory,logic operation and synapse.This thesis is based on the memristor synapse,the following research is done around the STDP study rule,CNN and the associative memory which is based on memristive neural network.First of all,the basic circuit properties of the memristor are studied,and the expression of the mathematical model of the memristor is analyzed.Secondly,the mathematical characteristics of the memristor nonlinear model are analyzed,and the nonlinear window function of several commonly used memristors are compared.Then,the author analyzes two existing circuit models of memristor bridge,and study the STDP learning rules and spiking neural networks proposed in recent years.At the same time,through the theoretical analysis,this paper validates the possibility of using memristor to realize synapse.This thesis makes a detailed analysis and study for the use of memory to achieve the feed forward memristive array,the spiking neural network based on STDP learning rule and its application.Based on the Spectre circuit simulation results,the experimental STDP learning function is obtained,which is basically consistent with the theoretical STDP learning functions.Subsequently,the thesis discuss the cellular neural network and its basic theory,and use Matlab to construct the cellular neural network and analyze its stability and transient response.In this paper,a SimScape memristor model is improved,based on the theoretical analysis of the third chapter,Simulink is used to construct the circuit of the memristor bridge.The detailed process of precisely adjusting the memristor bridge synapse with voltage pulse signal is demonstrated.Using the memristor bridge synapse to replace the weight module in the CNN,the cellular neural network based on the memristor synapse is obtained and the de-noising and edge extraction in image processing are accomplished by using this network.Finally,the paper mainly studies the application of memristor synapse in associative memory and presents an improved associative memory model.After analyzing the Hebb learning rule and associative memory,this paper analyzes two kinds of associative memory models with different structures in detail,respectively,using the digital cell and memristor synapse to achieve associative memory and using a single memristor synapse to achieve the association memory.Neither of these models achieves the fading and forgetting process of the Pavlova experimental model.Based on the second existing model,considering its shortcomings,this paper presents an improved associative memory model.The Pspice circuit simulation results show that the associative memory model proposed in this paper is not only simple in structure,but also a good way to realize the fading and forgetting process of Pavlova experiment.