Abstract
Custom routing of power and ground nets in an integrated circuit remains a manual task, in that a layout engineer must balance irregular blockages, electromigration (EM) limits, and voltage drop (IR drop) budgets at the same time. To address these competing pressures, a two-stage routing method divides that work between a learned global search and an exact local refinement. In a first stage of the method, a reinforcement learning (RL) agent explores a coarse routing grid and proposes a trial route together with an approximate wire width for each segment. In a second stage of the method, an integer linear programming (ILP) solver refines the trial route on a fine routing grid restricted to a narrow corridor around that route. That refinement returns a precise path and a final width for each wire segment. The method can shorten routing turnaround from days or weeks to minutes, and the returned routes respect EM, voltage drop, and design rule limits.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Pendyala, Prateek, "Two-Stage Routing of Power and Ground Nets in an Integrated Circuit Using Reinforcement Learning and Integer Linear Programming", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/11689