Study On The Memory And Neural Network Application Based On Resistive Switching Memristor

However,with the continuous reduction of the size of semiconductor components,the size and performance of integrated circuits will approach the physical limit,which makes Moore’s law a great challenge.The memristor with nano-sized structure,nonvolatile,CMOS compatible and fast storage speed is considered to be a strong candidate to replace the existing transistors and continue Moore’s law.Therefore,the researchers built the memory-computing integrated systems and neural network systems based on memristor.However,the complexity of circuit is high and has large hardware resource consumption.Therefore,this paper aims to reduce the consumption of hardware resources,simplify the circuit structure,and construct a new memory circuit and neural network application circuit based on memristor,the specific research work is as follows:(1)Aiming at the problems of boundary locking in HP memristor model,by analyzing the ion drift phenomenon of memristor and comparing the related performance of different window functions,a memristor model based on Biolek window function is designed.Setting a small threshold voltage for the memristor model can improve the resistance change rate of the device and reduce the power consumption of the device.The simulation results show that the v-i characteristic and variable resistance characteristic of the model accord with the basic theory of memristor.(2)Designing a binary memristor storage read-write circuit.On this basis,using the idea of dividing memristor values at equal intervals,the binary storage circuit can be converted into a quaternary memristor storage circuit.The simulation test shows that the circuit can accurately write or read the set value by adding different pulse voltages,and has the advantage of simple operation.(3)An artificial neural synaptic circuit based on the improved memristor model is designed and applied to the perceptron circuit.By adding different pulse voltages to the synaptic circuit and changing the weight of the perceptron,the Boolean logic operations of "AND" and "OR" are realized.The memristor cellular neural network(CNN)circuit based on the synaptic circuit is designed and implemented.According to the design rules of classical general CNN and multiple nested general CNN discriminant function,the template parameters for realizing two-input nonlinear Boolean logic and multi-input nonlinear Boolean logic are calculated respectively,and the two input reconfigurable “XOR” circuit and four input 2-bit multiplier circuit are constructed.The idea and method of applying memristor CNN to image denoise processing are studied.The experiments show that the reconfigurable logic circuit and image denoise circuit based on memristor CNN have the advantages of single structure,simple operation,easy control and low power consumption.