EDAML 2022 Invited Speaker 1: Application of Machine Learning in High Level Synthesis

Abstract

Traditionally high-level optimizations like CSA and sharing have been done technology independent. Doing word level optimizations PPA aware require accurate models for power, timing and area and that needs mapping to technology library and iterations which are runtime intensive and not suitable for multimillion instance designs. In this talk, we discuss how machine learning could help make the right power/area/delay tradeoffs early in the synthesis flow not sacrificing on turnaround time.