Convolution Operation Based On Spin Memristor Crossbar

In recent years,image preprocessing tends to consume a lot of time and energy in the emerging data-intensive application-Deep Convolutional Neural Network(DCNN).Convolution operation which is often used to remove noise and extract the edges of an image is a widely used and far-reaching algorithm in the field of image processing.When performing convolution operations on images involves a lot of multiplication and addition operations,the operation overhead is very large,real-time requirements are high.Insufficient memory processing performance will lead to ”memory walls effect”,frequent use of the above algorithms will bring great memory/communication burden to the computing system under data-intensive conditions.In view of the limit of technology development of CMOS process,people's attention has shifted from seeking faster processors to alleviating the ”memory bottleneck”.Memristor has the storage characteristics of storage and calculation fusion,which make it have a wide range of application prospects in the field of CIM(Computation in Memory).CIM is a brain-like architecture with a relatively simple structure and flexible application.In addition,the information density of the multi-value memristor is high,whose size has reached the nanometer level.The memristor circuit has a very high degree of integration,so it has a wide range of application prospects in the fields of large-scale memory,logic operation,neural network and other fields.This paper has conducted an in-depth study on the cross array of memristor and storage applications.The mathematical and physical models of four common memristors are studied,a logic switch based on the spin memristor is proposed,and cross array based on spin memristor is designed to realize the multiplication and addition operation,and designs the peripheral circuit,which realize the control of the image convolution algorithm on the cross array,greatly alleviates the memory bottleneck of the calculation.The specific content of this paper is as follows:(1)Firstly,this paper summarizes the common memristor models,i.e.HP model,Simmons channel barrier model,TEAM model,and spin memristor model,introduces the physical mechanism and characteristics of these models,and conducts mathematical simulations,analyzes their characteristic curves under different external excitation conditions,and compares the advantages and disadvantages of these models.This paper focuses on the resistive switching mechanism of the spin memristor,and deduces the switching time of magnetically controlled spin memristor.A memristor switch(MS)based on a magnetic flux controlled spin memristor is designed to realize logic operations of AND,OR,and NOT,the circuit is analyzed in detail,and the circuit is analyzed in detail and simulated at the circuit level.(2)Secondly,this paper proposes an image processing circuit based on cross array of spin memristor,which includes cross array of spin memristor,row switch selector,column switch selector,address decoder,weighted mean filter,read circuit,read/write control circuit,and the write,read and recovery operation of charge controlled spin memristor is simulated at circuit level.The circuit architecture alleviates the memory bottleneck of the calculation,and improves the efficiency of the circuit through the parallel processing architecture.Two application-level simulations of color image filtering and color image edge extraction are realized on the circuit.SRMC operator,median filter operator,Gaussian filter operator,and image sharpening operator are applied in color image filtering.Prewitt operator,Soble operator,Kirsch operator,Robert operator,Laplacian operator are used in color image edge extraction,and the Prewitt operator and Soble operator are improved to make the edge extraction information more abundant.By calculating the peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)of the image results,the processing effects of different operators are compared.Finally,it summarizes the work of this paper and prospects the work that needs to be done in the future.