SEALS: Sensitivity-driven Efficient Approximate Logic Synthesis
TimeTuesday, July 12th4:50pm - 5:10pm PDT
Location3007, Level 3
RTL/Logic Level and High-level Synthesis
DescriptionApproximate computing is an emerging computing paradigm to design energy-efficient systems. In this paper, we propose ALSPDS, an approximate logic synthesis (ALS) method driven by partial difference (PD) and sensitivity to speed up ALS flows. First, an efficient local approximate change (LAC) filtering method is proposed to filter out some unpromising LACs based on sensitivity measure. Then we propose a fast and accurate error estimation method using the local change propagation of PD. ALSPDS can be applied to any statistical error measure. We apply it to two state-of-the-art ALS approaches and demonstrate its effectiveness to synthesize better approximate circuits.