ACM: An Energy-Efficient Accuracy Configurable Multiplier for Error-Resilient Applications

The multimedia applications such as image, audio and video processing allow approximation in computations, provided that, errors are of definite types and have austerities within confined limits, thus exhibiting error-resiliency. An approximate arithmetic circuit can be exploited to avail this error-resiliency for improving energy-efficiency. This paper presents an approximate multiplier that provides higher energy-efficiency at the cost of minor loss of accuracy. The proposed multiplier offers twofold improved performance because of reduced level of gates and curtailed inherent switched capacitances. Further, to achieve variable accuracy, an Accuracy Configurable Multiplier (ACM) algorithm is proposed that provides improved Speed-Power-Area-Accuracy (SPAA) metrics. The proposed ACM enables dynamic accuracy configurability via small error correction logic. Simulation results over accurate 8-bit multiplier indicate 57.37% and 25.17% reduced power and area, respectively. Moreover, accuracy configurability is achieved with only 10.5% and 12.32%, area and power overhead, respectively. Moreover, the proposed multiplier in real applications such as Gaussian smoothing filter attains better SPAA tradeoff over the existing approximate multipliers.