Research On Filter Circuit And Neural Network Application Based On Memristor Simulation

With the discovery of memristor,the fourth kind of passive basic circuit element,the investigation of its principle and application prospect has become a research boom in the field of circuits.Memristor is a kind of non-linear resistor with memory function,showing special response ability to analog signals.Neural network architecture based on memristors could fundamentally solve the von Neumann bottleneck.However,there is no unified standard for the internal mechanism of the device and the process of preparation materials,which limits application and research of the memristor.Therefore,simulation is necessary before preparing devices and constructing circuits.As the basis of circuit simulation,the construction and simulation analysis of device models play an important role in aspects like optimizing circuit functions,improving circuit design efficiency,and reducing circuit manufacturing costs.This paper investigates the simulation model of memristor and its application from the following three aspects:1.Construction and comparative analysis of memristor models.This article constructed the memristor model for simulation in SPICE environment,comparing and analyzing model include ideal memristor model and bipolar threshold behavior model(Pershin model and Biolek model)based on boundary migration theory,compact model based on conductive filaments and the model based on MOSFET circuit.And the memristor models under different theories are compared and analyzed.Among them,the memristor model based on the boundary migration theory is more idealized,and the circuit model based on the MOS has a lower match with the actual memristor,while the compact model is closer to the actual physical device performance and provides theoretical guidance for the mechanistic research and application of the memristor.2.Building and analyzing low-pass filter circuit based on ideal memristor.Low-pass filter circuits based on traditional resistance and memristor are simulated and analyzed using the Pspice software,and the filter characteristics of first-order MC(memristor and capacitance)filter circuit and second-order MC filter circuit are compared.Simulation results show that the low-pass filter circuit based on memristor has better frequency selection characteristics.What's more,the performance of second-order MC circuit is better than that of first-order MC circuit because of its superposition,indicating that the unique ability of responding to analog signals ofmemristor has wide application prospect in traditional circuits.3.Realizing high-accuracy handwritten digit body recognition system based on the memristor compact model.Based on the compact model of conductive filament,Ohm's law is used to simulate matrix vector multiplication instead of traditional neural network convolution calculation.The neural network based on memristor array is simulated in Python environment.By adjusting the resistance of memristor to train the neural network,the handwritten digital body recognition function based on MNIST database(Mixed National Institute of Standards and Technology database)is realized,which improves the learning efficiency of neural network,also achieving a recognition accuracy of 97%.