Design Space Optimization with Machine Learning Library Pruning

Abstract

In recent years, machine learning is a hot topic for electronic design automation (EDA). The design industries are leveraging the power of machine learning to drive the predictability of the physical design and signoff. The current machine learning application in IC design mainly focus on yield modelling, lithography hotspot, noise modeling, process variation modelling, performance modelling for analogue circuit and implementation space exploration. This paper will share the idea of library auto-pruning with machine learning. The pruned library is feed as permuton to design space optimizer AI. The pruned library permuton provides additional 7% leakage saving and potentially 15% max frequency improvement.