An Error Corrector for Dynamically Accuracy-Configurable Approximate Adder

Abstract

To reduce power consumption, approximate computing is an efficient approach for error-tolerant applications such as image processing. Approximate arithmetic adders can be used for the approximate computing, and can trade off accuracy for power. CMA, a dynamically accuracy-configurable approximate adder, had been proposed. CMA can sharply reduce power consumption compared with other accuracy-configurable approximate adders, while allowing it to change accuracy-setting at run-time. In this paper, we propose a tiny error corrector for CMA that needs to only two gates for each digit. By consuming 1.4% extra power, the proposed value corrector can improve accuracy for CMA by up to 29.3% on average.