Applying Reinforcement Learning to Physical Design Routing

Gandhi, Upma

Applying Reinforcement Learning to Physical Design Routing

dc.contributor.advisor	Behjat, Laleh
dc.contributor.author	Gandhi, Upma
dc.contributor.committeemember	Bustany, Ismail S. K.
dc.contributor.committeemember	Yanushkevich, Svetlana
dc.contributor.committeemember	Taylor, Matthew E.
dc.date	2024-05
dc.date.accessioned	2024-04-30T18:09:25Z
dc.date.available	2024-04-30T18:09:25Z
dc.date.issued	2024-04-26
dc.description.abstract	Global routing is a significant step in designing an Integrated Circuit (IC). The quality of the global routing solution can affect its efficiency, functionality, and manufacturability. The Rip-up and Re-route (RRR) approach to global routing is widely used to generate solutions iteratively by ripping nets that cause violations and re-routing them. The main objective of this thesis is to model a complex problem such as global routing as an RL problem and test it on practical-sized routing benchmarks available in academia. The contributions presented in this thesis concentrate on automating the RRR approach by applying reinforcement learning (RL). The advantage of the RL over other machine learning-based models is that it can address the scarcity of data in the global routing field. All contributions model the RRR as an RL problem and present developed frameworks to generate solutions. The first contribution presented is called β Physical Design Router (β-PD-Router). Router and Ripper agents in this contribution are trained to resolve short violations on sample-sized circuits with size-independent features. β-PD-Router achieved ∼ 94 % accuracy to resolve violations on unseen netlists. An RL-based Ripper Framework has been developed as the second contribution to train a Ripper agent with the Advantage Actor-Critic RL algorithm to minimize short violations. One of the most current benchmark suits is used to test the performance of RL-Ripper. The third contribution discussed in this thesis is called the Ripper Framework 2.0, an extension to the Ripper Framework. It focuses on improving the generalizability of bigger designs by applying the Deep Q-Networks RL algorithm. After the first iteration of detailed routing, the guide generated with Ripper Framework 2.0 outperforms the state-of-the-route global router in the number of violations.
dc.identifier.citation	Gandhi, U. (2024). Applying reinforcement learning to physical design routing (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.
dc.identifier.uri	https://hdl.handle.net/1880/118559
dc.identifier.uri	https://doi.org/10.11575/PRISM/43401
dc.language.iso	en
dc.publisher.faculty	Graduate Studies
dc.publisher.institution	University of Calgary
dc.rights	University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.
dc.subject	Reinforcement Learning
dc.subject	Global Routing
dc.subject	Physical Design
dc.subject.classification	Engineering--Electronics and Electrical
dc.subject.classification	Computer Science
dc.subject.classification	Artificial Intelligence
dc.title	Applying Reinforcement Learning to Physical Design Routing
dc.type	doctoral thesis
thesis.degree.discipline	Engineering – Electrical & Computer
thesis.degree.grantor	University of Calgary
thesis.degree.name	Doctor of Philosophy (PhD)
ucalgary.thesis.accesssetbystudent	I do not require a thesis withhold – my thesis will have open access and can be viewed and downloaded publicly as soon as possible.