A Reflection on Machine Learning for EDA: what are possible future directions?

The increasing power of deep learning-based machine learning (ML) techniques have started to erode many of the traditional equation-based physical modeling techniques, some of which are corner stone technologies for the development of electronic design automation (EDA) as an active area in the past decades. Since the inception of the new area of ML for EDA, there have always been controversies or doubts on the sensibility of such a formulation. Based on the speaker’s years of research experience in both EDA and ML as two separate domains, he will reflect upon his personal learning in these two seemingly different areas, how the controversies may be reconciled, and how a cohesive view on the subject of ML for EDA can be formed. In doing so, the speaker would like to shed some lights on the future of ML for EDA, and what research challenges are ahead of the ML for the EDA research community.