Maximizing yield in Near-Threshold Computing under the presence of process variation

Abstract

Near-Threshold Computing (NTC) shows potential to provide significant energy efficiency improvements as it alleviates the impact of leakage in modern deep sub-micron CMOS technology. As the gap between supply and threshold voltage shrink, however, the energy efficiency gains come at the cost of device performance variability. Thus, adopting near-threshold in modern CAD flows requires careful consideration when addressing commonly targeted objectives. We propose a process variation-aware near-threshold voltage (PV-Nvt) gate sizing framework for minimizing power subject to performance yield constraints. We evaluate our approach using an industrial-flow on a set of modern benchmarks. Our results show our method achieves significant improvement in leakage power, while meeting performance yield targets, over a state-of-the-art method that does not consider near-threshold computing.