In the majority of Digital Signal Processing (DSP) applications, such as image, audio and video processing, the final result is interpreted by human senses; and, the fact of confined perception of human senses declines the strict restriction on accuracy. Thus, by adopting the emerging concept of approximate computing, we propose an approximate radix-2 hybrid redundant Multiply-and-Accumulate (ApproxMAC) unit which stems a novel Speed-Power-Accuracy-Area (SPAA) metrics. The Approx-MAC unit attains tremendous improvements in computational performance, energy efficiency and silicon area with a trivial degradation in the output quality. To inspect the effectiveness of the proposed approach in real-time DSP applications, we demonstrate an ApproxMAC unit embedded JPEG-E-X IP core architecture. The ApproxMAC unit with 40 approximate LSBs ensures 7.177× and 1.526× speedup, 1.594× and 4.163× energy efficiency, and 1.131× and 1.277× silicon area improvements over binary and hybrid redundant MAC units, respectively. Moreover, the ApproxMAC unit with 40 approximate LSBs decorates powerprecision and delay-precision metrics by 14.71% and 32.95%, respectively.

No abstract