Optimizing energy to minimize errors in dataflow graphs using approximate adders

Abstract

Approximate arithmetic is a promising, new approach to low-energy designs while tackling reliability issues. We present a method to optimally distribute a given energy budget among adders in a dataflow graph so as to minimize expected errors. The method is based on new formal mathematical models and algorithms, which quantitatively characterize the relative importance of the adders in a circuit. We demonstrate this method on a finite impulse response filter and a Fast Fourier Transform. The optimized energy distribution yields 2.05X lower error in a 16-point FFT and images with SNR 1.42X higher than those achieved by the best previous approach.