Research And Application Of Neural Network Circuit Based On Memristor Bridge Synapse

The memristor was put forward on mathematical logic relationship between the basic circuit variables by professor chua in 1971,it has a series of excellent properties,such as nonvolatile information,nanometer dimension,nonlinear characteristics.It is widely believed to have great application prospects in the information storage and control circuit,nonlinear circuit,artificial neural network etc.With the continuous development of the information age,more intelligent and more diminutive information processing system are expected urgently.A feasible solution is provided by simulating the nervous network of artificial neural networks,and it has been a hot area of scientific research.Memory effect of memristors is similar to the function of some synapses in neuromorphic systems.The use of memristors is expected to build a more bionic intelligent neural network system,thus accelerating the ability of information processing.In this dissertation,based on the error analysis and experimental proof,it is proved that the bipolar pulse can effectively reduce the error caused by the memristive drift.Meanwhile,a bipolar pulse circuit capable of generating equal and symmetrical pulses is designed and applied to neural synaptic and neural networks.Further,by analyzing the principles of neurons and neuronal synapses,a more flexible neural network circuit was designed and implemented.The main work of this paper includes the following:(1)First,this dissertation introduces the principle and feasibility of memristor synapses.Then,the characteristics of memristive combinational circuit are analyzed,including series structure and parallel structure.Based on the simple combinational circuit of memristor,the principle of memristive bridge synaptic circuit is analyzed,including four memristors and five memristors.Finally,the structure and characteristics of the memristive bridge neural network are introduced by combining the cellular neural network and the memristive bridge synaptic structure.(2)Then,the principle of the memristor bridge synapses is analyzed,and the memristive drifts in linear and nonlinear HP memristor model are discussed respectively.Because of the phenomenon of memristive drift,it will lead to a certain degree of simulation error,so a mechanism to reduce the simulation error by the symmetry of bipolar pulse is proposed.Based on this,a bipolar pulse circuit is designed,which is used for memristor synapses,reducing the synaptic simulation error.In addition,using the numerical analysis and simulation verify the reliability of the proposed method.(3)Further,based on the bipolar pulse generator,the synaptic structure is combined with cellular neural network(CNN)to construct an optimized memristive synaptic neural network.Due to the reduction of the error of the memristive drift,the synaptic weights are more accurate.In the CNN,some image processing functions can be realized by convolution of the binary of the template operator and the image pixel.This template operator is usually a matrix of the numerical components.Therefore,the synaptic weights corresponding to the cellular neural network template operator in the image processing can perform the image processing ability.Compared with the traditional neural network processing ability,the optimized neural network in this dissertation shows the superior image processing effect,and the Matlab simulation verify the effectiveness of the proposed scheme.(4)Finally,a new memristor bridge neuron and synaptic circuit are designed.The new synapse is more flexible to achieve synaptic weights update.Combined with the error back-propagation algorithm,the more efficient circuit of the neural network is constructed.And by the Pavlov associative memory simulation experiments,the feasibility and accuracy of the structure is verified.