Research Of Competitive Neural Network And Associative Memory Based On Memristor

Artificial neural network is a new calculation model established by simulating the functions of biological neural network.Compared with past calculation models,artificial neural network is more adaptive and intelligent.However,as it is more and more difficult to minify the size of CMOS transistors,the scale of artificial neural networks is greatly limited and it is difficult to solve the increasingly complex problems.Memristor is a new type of resistive memory device that owns some advantages including small volume,nonvolatility,low power consumption and variable resistance controlled by charge flowing through itself.Using the memristor to replace traditional transistors as synapses in the artificial neural network circuit will greatly reduce the volume of circuit and the energy consumption and improve the ability of information processing.This paper briefly outlines the development and application prospect of the memristor and expounds the status quo of the neural network circuit based on memristors.Then this paper makes a brief introduction of the usual device model of memristor,and analysis the advantages and disadvantages of various models and applications.It lays the foundation for choosing appropriate memristor model for different neural network circuit.Afterwards the paper introduces defects of competition neural network circuit based on memristor at present,such as leak current and error of calculating similarity between input vector and prototype vector.In order to overcome these shortcomings,this paper proposes solutions that includes a training algorithm for competitive neural network and the specific circuit.In order to verify the validity of the solution,the proposed circuit that has been trained is used to identify the digital images.Finally,the paper discusses the meaning and significance of the associative memory network and designs a memristor-based associative memory network according to the original circuit.Compared with the previous circuit,the threshold value and pulse width in the proposed circuit can be adjusted,which enhances the flexibility and extensibility of the circuit.The design circuit is used to simulate the famous Pavlov's dog experiment to validate whether the proposed circuit owns associative memory function.The main work of this paper is the neural network circuit based on the memristor.The main research work in this paper mainly includes the advantages and disadvantages and the memristor and the applications of these models,the competitive neural network and associative memory network based on memristor.The achievements of this research promote the development of competitive neural network and associative memory network based on memristor,and provide references for other neural network circuits based on memristor.