Study On The Bionic Characteristics And Application Of Synapse Based On Memristor Devices

As a new type of electronic device with the advantages of nonlinearity,nanoscale size,synaptic-like properties,and logical operation,memristors are regarded as one of the most promising bionic devices for realizing neural network hardware.At present,there are many memristor models to realize synaptic bionic function,and the construction of memristor models with multi-functional synaptic bionic is the necessary way to realize the development of neural network hardware.In this paper,we realize the simulation of neural synapses and complete the basic research application of logic calculation and picture imaging based on memristor devices.The main body of the content is developed in four parts:Firstly,the physical and mathematical models of the HP memristor and the WO_xmemristor model are examined.Then the mathematical model is used to build the simulation model to analyze the basic characteristics of the memristors,such as the I-V characteristic.The synaptic plasticity and nonlinear transmission characteristics of the two memristors are further analyzed,and the WO_xmemristor model,which fits better into the neural synapse,is selected by comparison as the basic device model for the simulation research work later on.Secondly,an improved tungsten oxide memristor model with temperature characteristics is constructed,and variables controlling the temperature are introduced on the basis of the original model to make it closer to the actual measurement device.On this basis,a single memristor device is used to achieve more comprehensive synaptic learning functions,including learning forgetting properties,short-term synaptic plasticity,long-term synaptic plasticity and experiential learning,thus showing that the model has the function of simulating biological synaptic properties.Thirdly,the frequency response characteristics of the tungsten oxide model were analyzed.The weights decreased under the influence of low-frequency signals,while high-frequency signals could cause the weights to increase.The simulation of the neural firing pattern of a neural network is realized by using the influence of different pulse frequencies on the weights,and the feasibility of applying the model to the pulsed neural network is confirmed.To further realize the application of the memristor,we mainly use the synaptic plasticity of the memristor model to build a circuit to realize logical"AND"and"OR"gates and use the nonlinear transmission characteristics of the memristor to build a memristor array to realize the Chinese character recognition process.The implementation of these applications shows that the model can provide a research basis for neural network hardware implementation.Finally,based on the memcapacitor model of the memristor-derived device,the state variables controlling the forgetting time and memory retention value are added to the memcapacitor model with forgetting characteristics.On this basis,the learning function of the memcapacitor as a neural synapse,including learning and forgetting characteristics and experiential learning,is investigated.And the effect of pulse frequency on the memcapacitor and its application in neural networks are further studied.These findings validate the similarity of the properties of memristors and memcapacitors and also provide ideas for memcapacitors as devices for implementing neural networks.By comparing the similarities and differences of memristors and memcapacitors as neural synapses to realize synaptic bionic functions,the differences and connections between memristors and memcapacitors are clarified.