Skip to main content Skip to main navigation

Publication

PolyEMAC: Polynomial Error Metrics Analysis in Approximate Computing

Mohamed Nadeem; Chandan Jha; Rolf Drechsler
In: 39th International Conference On VLSI Design. International Conference on VLSI Design (VLSID-2026), January 3-4, Pune, India, 2026.

Abstract

Approximate Computing (AC) is a design paradigm widely used in error-resilient applications. The goal of AC is to gain benefits over exact designs by trading some accuracy for a reduction in power, delay, and area. Varying error metrics are used to assess the degree of approximation. Several formal methods exist for computing these error metrics. However, these methods often fail to provide theoretical bounds on computational resources and are mostly limited to single error metrics. In this work, we address the limitations of the prior methods by introducing PolyEMAC, a novel approach that combines Cutwidth Partitioning and Answer Set Programming (ASP) to compute several error metrics for approximate circuits with constant cutwidth in polynomial time w.r.t. circuit size and the number of approximations. We evaluate several approximate adders in terms of cutwidth, memory consumption, and computation time to highlight the efficacy of the proposed methodology.