SASHA: A Shift-Add Segmented Hybrid Approximated Multiplier for Image Processing

Abstract

In this paper, a new shift-add segmented hybrid approximated (SASHA) multiplier for image processing applications is proposed. The new multiplier uses segmentation to achieve high performance in power reduction and accuracy of results. It segments the operands and most significant bits. Three hardware implementations of the 2-bit, 4-bit and 6-bit SASHA approximate multiplier are presented in this paper. The power consumption and accuracy of the proposed multipliers are evaluated by comparing performance with non-approximated multipliers using different design parameters. Experimental results show that the accuracy in terms of mean relative error percentage of 2-bit, 4-bit and 6-bit SASHA have minimum impact on the performance of image processing applications such as edge detection of an image. Additionally, a reduction of signal and logic power consumption is observed.