Analysis Of Dynamic Characteristics Of Memristor Synaptic Bridge Circuit And Its Application In Neural Networks

The memristor is a novel circuit element with memory,which has broad application prospects in Resistive random access memory,logic operation and neural network.Based on memristor,this paper made the following research work on memristive neural network and associative memory model:The basic principle and mathematical model of memristor were studied.The characteristics of the linear HP memristor,nonlinear HP memristor,matching memristor and Schottky-membrane memristor are studied,and the common window functions of the nonlinear memristor are compared and analyzed.The principle mechanism of each memristor model is introduced.Mathematical modeling and simulation analysis of each model is performed with matlb software.The simulation results verify that the memristor is a memory element,and its volt-ampere characteristic curve will show obvious hysteresis loop characteristics,and will gradually degenerate into a linear resistance as the frequency of the applied AC signal source rises.In addition,the simulation results also verify the resistance switching characteristics of the memristor and the forgetting characteristics of the Schottky forgetting model.The dynamic characteristics of four classic synaptic bridge circuits based on HP memristor are researched.The main types of M1,M2,M4 and M5 type memristance synaptic bridge circuits are studied.The structure and principle of each synaptic bridge circuit are introduced,and the weight and resistance of each circuit are simulated and analyzed by PSPICE software.The simulation results show that M1 and M2 memristance synaptic bridge circuits can not achieve synaptic negative weights and zero weights;M5 memristive synaptic bridges can achieve synaptic positive and negative weights but can not achieve zero weights;M4 type memristy synapses The bridge can realize the positive and negative weights of the synapses as well as the zero weights,and due to the complementary symmetry characteristics of the M4 type synaptic bridge,the weights approximately linearly change,and linearly adjustable memristive synapses can be realized.The training learning method of memristive neural network based on M4 synaptic bridge circuit is studied.Memristive neural network is implemented based on M4 type synaptic bridge circuit.The traditional gradient descent method and RWC(random weight changing)method suitable for hardware circuit training are introduced.The advantages and disadvantages of the two methods when applied to hardware memristive neural network are analyzed.The advantages of the two weight update methods suggest a hybrid training method.The training method is applied to the pattern recognition field by using MATLAB,and the experimental results of the XOR problem and the handwriting recognition problem are simulated.The simulation results show that the hybrid training method not only can greatly reduce the number of training iterations,but also can obtain good recognition accuracy.Therefore,it is a training method suitable for hardware memristive neural network.A forgetting threshold memristor model was proposed and its application in associative memory was studied.The Hebb learning rule and the Pavlova association experiment are introduced.Based on the HP memristor model and the matching model,a threshold memristor model with exponential forgetting rule is proposed and the PSPICE macro model is given.Based on the forgotten threshold model,an associative memory model is given.Using the PSPICE simulation,the Pavlova experiment with forgetting process is reproduced.The associative memory model is further extended to an associative memory network,and two kinds of Pavlova expansion experiments under multi-input signals are implemented by PSPICE.The simulation results confirm that the forgetting threshold memristor model can effectively construct an associative memory network.The network can realize the positive and negative settings of synaptic weights and reproduce the associative memory and forgetting process of the brain.