Modeling And Verification Of Logic Network Based On Memristor

With the development of information age,neural network technology has been widely used in more and more fields.At present,with the increasing amount of information to be processed and the degree of calculation becoming more and more complex,how to improve the speed of network training and deal with massive data efficiently and accurately has become the key problem of neural network algorithm breakthrough.Memristor is a very popular new material in recent years,which has the characteristics of nonvolatile and high density.It is found that logic operation can be realized on the cross array of memristor.Based on memristor operation and storage function,this paper mainly takes Echo State Network(ESN)as the object to model and analyze the feasibility of logical network.It is proved that the memory operation fusion module based on memristor can not only improve the training speed of neural network,but also has a high application prospect.This paper mainly studies from the following three aspects:Firstly,this thesis studies the model design based on memristor's nonvolatile and operation function.From the perspective of image data processing and the fusion of operation function and storage module.The purpose is to greatly reduce the memory access operation of training data in image preprocessing,and improve the efficiency of the whole neural network.Secondly,the model design of basic operation unit suitable for image data preprocessing is proposed,and the circuit is simulated on the platform of Simulink.In the simulation,the circuit functions of 3-8 decoder,counter,shifter,adder and multiplier based on memristor are verified,and the expected results are achieved.Finally,in order to verify the correctness of the proposed structure and design,the handwritten digital image recognition is taken as the object of simulation experiment.Through the convolution operation of the target image by the designed basic operation unit,the characteristic value is trained and tested by ESN algorithm,and compared with the traditional BP neural network algorithm,the efficiency of the design is verified.