Machine Learning based Circuit Routing

Routing is a problem in Electronics Design Automation (EDA), which sits as the intersection of Electrical Engineering and Computer Science. It is more challenging than path finding in other domains, such as robotics, in that paths cannot intersect in circuit routing – so a greedy, optiomal path can make a solution globally suboptimal or even invalid. An example is given below (from Kahng et al., VLSI Physical Design, 2nd edition, 2022)

Conventionally, circuit routing is done using heuristics. Such algorithms need to be manually designed and maintained. This is laborious and hard, especially after the (corner) cases explode. As believers of Machine Learning, we wanna teach computers to route circuits by learning. This great dream of science has great practical and commercial potentials. We are “dumping” different machine learning techniques onto it: reinforcement learning, computer vision, and even NLP.

Our efforts

Recent, important papers

On Using ML for routing (unfortunately, none of them made to top EDA/AI conferences.)

On more conventional approaches (from top EDA confs.)





Open source projects