High performance 8-bit approximate multiplier using novel 4:2 approximate compressors for fast image processing

Abstract

In this paper, a novel 8-bit approximate multiplier is proposed based on three novel 4:2 approximate compressors which its delay and error is less than those of the multipliers constructed by traditional 4:2 approximate compressors, and its delay is also less than that of an 8-bit multiplier constructed by using 3:2 precise compressors. To do so, each novel compressor is designed such that its output carry is independent of the output carry of its previous compressor in the multiplier. Therefore, the problem of carry propagation delay is eliminated and a fast multiplier is constructed. To obtain the most accurate multiplier, the best compressor of the three proposed compressors for each multiplier’s column is determined using the genetic algorithm. Moreover, one can use the approximate compressors only at the k least significant multiplier’s columns for more error reduction. The proposed multiplier is used for image blending and image compression. Our simulations show that for example the error and the delay of the proposed method for k=9 is at-least 32.52% and 33.10% less than those of traditional 4:2 approximate compressor based multipliers, respectively.

Abstract: In this paper, a novel 8-bit approximate multiplier is proposed based on three novel 4:2 approximate compressors which its delay and error is less than those of the multipliers constructed by traditional 4:2 approximate compressors, and its delay is also less than that of an 8-bit multiplier constructed by using 3:2 precise compressors. To do so, each novel compressor is designed such that its output carry is independent of the output carry of its previous compressor in the multiplier. Therefore, the problem of carry propagation delay is eliminated and a fast multiplier is constructed. To obtain the most accurate multiplier, the best compressor of the three proposed compressors for each multiplier’s column is determined using the genetic algorithm. Moreover, one can use the approximate compressors only at the k least significant multiplier’s columns for more error reduction. The proposed multiplier is used for image blending and image compression. Our simulations show that for example the error and the delay of the proposed method for k=9 is at-least 32.52% and 33.10% less than those of traditional 4:2 approximate compressor based multipliers, respectively.