Image Processing Optimization Research Based On Memristive Crossbar Arrays

Memristive crossbar arrays have huge application potential in the field of image processing,due to its ability to support Vector-Matrix Multiplication(VMM)operation.However,the current research on the memristor crossbar arrays in the field of image processing is still in the preliminary stage,and there is a problem that the calculation of the array is interfered by non-ideal factors,which decreases the reliability of the calculation results.Some image processing operations based on memristive crossbar arrays,such as image convolution,still have room for further optimization.When VMM operations are performed in the large-scale memristive crossbar arrays,there exists a serious IR drop problem,resulting in poor output image quality.In order to alleviate the IR drop problem and improve the reliability of the output image,the Complementary Transpose(CT)scheme is proposed.Specially,the analysis found that the sum of the wire resistance on the gating circuit of two memristor cells,which are symmetrical about the center point in the array,is a constant.Thus,a symmetrical circuitry design,which needs a spare array with the same size as the original array to store the same image data,is adopted to suppress the influence of interconnect resistance.In order to improve the image convolution speed based on the memristive crossbar arrays,the Image Strorage Convolution(ISC)scheme is proposed.Specially,analyzing the image convolution formula,it is found that the image convolution can be split into several VMM processes,and the corresponding circuitry design scheme is proposed,and the CT scheme is used to alleviate the IR drop problem in the large-scale array to improve the output image quality.Compared with the existing Kernal Strorage Convolution(KSC)scheme,the optimization scheme can reduce the time complexity from O(n~2)to O(n),and accelerate the image convolution speed.Experimental results show that,compared with other existing IR drop optimization schemes,the CT scheme maintains the best image quality under different scales of images and wire resistance conditions,but increases the area and energy consumption overhead.When the image is size 256×256,and the wire resistance is 0.72?,the CT scheme is selected to achieve an optimal balance between image quality and peripheral circuit overhead.Compared with the KSC scheme,the ISC scheme can greatly improve computational efficiency without reducing the quality of the output image.When the Canny image edge detection algorithm experiment is implemented based on the image of size 256×256,the image convolution speed can be increased by 85 times.