Applications of Machine Learning in VLSI Design
Sneh Saurabh, Pranav Jain, Madhvi Agarwal, OVS Shashank Ram · 2021
Traditionally, VLSI design consists of several distinct steps carried out sequentially using electronic design automation (EDA) tools. Though EDA tools have matured over decades, with the advancement of technology, new problems emerge. Additionally, it requires a great deal of human intervention in making EDA tools work on new designs, especially at advanced process nodes. With the recent advancement in machine learning (ML) and a plethora of freely available tools, we can solve many challenging problems of VLSI design effectively using ML. The ML techniques enable deriving complex dependencies, especially for voluminous data that VLSI design flows routinely generate. In this chapter, we review applications of ML techniques for various VLSI design tasks. Specifically, we review their applications in system-level design, logic synthesis, physical design, verification, test, diagnosis, and validation. We illustrate the power and suitability of various ML techniques in tackling different types of design problems. We examine the potential of ML in streamlining and reducing the intervention of designers in the VLSI design flow. We also highlight the challenges encountered in deploying ML-based solutions in VLSI design and possible solutions.