Realization Of Equivalent Circuit Model Of Memristor And Its Application In Neural Network

With the rapid development of artificial intelligence,the traditional computer architecture according to von Neumann’s rule has entered the "bottleneck" period.As an important branch of artificial intelligence,brain-like intelligence has attracted more and more scientists’ attention.At present,the transition from software to hardware has become the only way for the development of brain-like intelligence.As a nonlinear passive component,memristor has the characteristics of small size,low energy consumption,non-volatile and variable resistance.All the non-volatile characteristics of memristor components make their working process very similar to that of biological synapses.Therefore,the memristor is expected to bring new hope for the development of artificial intelligence brain-like bionic circuit hardware.The researchers use these characteristics of memristor to be widely used in artificial neural network circuits to simulate biological behavior.In this paper,a general equivalent circuit model of memristor is constructed,and then the application of memristor in neural network circuits is analyzed and expanded.The main content of this paper is as follows:Aiming at the problems of low generality and low component utilization efficiency of traditional memristor equivalent model,a general hyperbolic function equivalent circuit model of memristor is constructed.The equivalent circuit model of the general hyperbolic function memristor is divided into two types: flux-controlled and charge-controlled.For the equivalent circuit model of flux-controlled memristor,the mathematical model is proposed first,and the corresponding circuit model is constructed according to the mathematical model.The equivalent circuit of flux-controlled hyperbolic function memristor is divided into magnetic signal generating and processing module,exponential circuit operation module and division circuit module,etc.Combined with practical application,the equivalent circuit model of fluxcontrolled memristor is extended to that of charge-controlled memristor,and the application of the equivalent circuit of general hyperbolic function memristor is further expanded.Either flux or charge controlled memristor equivalent circuit model can realize more and different mathematic models of memristor circuit by controlling the switch in the circuit.The proposed equivalent model circuit of general hyperbolic function memristor is helpful to better understand the characteristics of memristor,improve the efficiency of the equivalent model of memristor,and provide a reference for the application of memristor in brain-like intelligence.In view of the phenomenon that emotion plays an important role in the traditional conditioned reflex,but the existing bionic circuits involve few emotional factors,a feature recall network memristor circuit based on different emotions is constructed.The circuit includes emotion generating module,neuron module and output module.When the internal emotional signal reaches the threshold of the emotion-generating module,the emotion-generating module will generate positive or negative emotions.The generated emotional signal and the neuron module together have an effect on the output module,thus affecting the whole feature recall network.Based on the proposed feature recall network memristor circuit under different emotional conditions,further consideration of internal emotional stimuli on the associative memory feature recall process,provides a new idea for the construction of a more intelligent memristor based biomimetic brain circuit.In view of the phenomenon that different emotions have different effects on the generalization and differentiation process of conditioned reflex,a multilevel fear generalization and differentiation memristor neural network circuit with different emotional effects is constructed.The constructed circuit consists of synaptic module,emotion module,output module and voltage selection module.Emotional modules produce different emotions and have different effects on the multilevel fear generalization and differentiation process.Multistage fear generalization and differentiation memristor circuits based on different emotions are proposed.The effects of positive and negative emotions on fear generalization and differentiation are expressed by hardware circuits.It is concluded that positive emotions inhibit fear generalization and differentiation while negative emotions promote fear generalization and differentiation.The proposed circuit provides some reference for building intelligent bionic hardware circuit.In this paper,a general flux-controlled and charge-controlled memristor equivalent circuit is firstly constructed.On this basis,a recall memristor circuit with different emotional influences is constructed,and a multilevel fear generalization and differentiation memristor circuit based on different emotional influences is constructed.The study of memristor in this paper provides a reference for the development of memristor in neural network circuits in the future.