IDrAx: A tool-chain for designing efficient approximate adders

Abstract

As transistors shrink down, improving the energy-efficiency of nanometer ICs has gained a great deal of attention. Recently, approximate computing has been introduced to address energy-efficiency problems of error-tolerant systems. Adders are the most widely used arithmetic modules that consume significant amounts of system's total energy and area. The error-resiliency of these systems has made it possible to innovate energy-efficiency approximate adders to obtain low power and high-speed circuits. This paper presents IDrAx, an approach to improve the accuracy of approximate adders. The goal is to find the most significant errors in the error distribution for the given circuit and correct them through low-cost additional circuits. The proposed method is applied to several state-of-the-art approximate adders for evaluation by considering two different input distributions. Our simulation results indicate that the proposed design achieves by 75% and 78.01% accuracy regarding mean error distance metric for uniform and normal input distributions, respectively.