AbstractApproximate computing has emerged as a promising paradigm for designing energy-efficientdigital signal processing (DSP) systems, particularly in error-resilient applications such as imageprocessing, machine learning, and multimedia. This paper presents a novel rounding-basedapproximate multiplier (RBAM) that leverages controlled approximation to achieve significantimprovements in speed and energy efficiency while maintaining acceptable accuracy. The proposedmultiplier reduces computational complexity by rounding input operands to the nearest power-of-two,simplifying partial product generation. Experimental results on FPGA and ASIC platforms demonstratethat the RBAM achieves up to 35% reduction in power consumption and 40% improvement inspeed compared to exact multipliers, with a minimal mean error distance (MED) of < 2% for typicalDSP workloads. The design is evaluated in an image processing application, showing negligible visualquality degradation while significantly enhancing energy efficiency.
SHAIK SABEER HUSSAIN, YEDDULA PAWAN KUMAR, YELLANUR VARMA (Wed,) studied this question.