IEEE CASS RS Talks 2025: AI for Physical Design: Overcoming EDA Bottlenecks in Optimization, Analysis, and Automation
Upcoming Talk:
Join us for the 2025 IEEE CASS RS Talk, hosted by the Rio Grande do Sul Chapter.
This live-streamed event will take place on 1:30 PM (Brasilia Time, GMT-3), 6 June 2025 and features a talk by Prof. Vidya A. Chhabria titled "AI for Physical Design: Overcoming EDA Bottlenecks in Optimization, Analysis, and Automation."
Attendance is entirely free and the talk will be live streamed via Youtube on the IEEE CASS Rio Grande do Sul Chapter Youtube page.
Biography
Vidya A. Chhabria is an assistant professor in the School of Electrical, Computer, and Energy Engineering at Arizona State University. She graduated with her Ph.D. and M.S. degrees from the University of Minnesota in 2022 and 2018. Her research interests lie in computer-aided design (CAD) for VLSI systems and primarily revolve around physical design, optimization, and analysis algorithms. She has received the ICCAD Best Paper Award (2021), the University of Minnesota Graduate School's Best Dissertation Award in the Physical Sciences, Mathematics, and Engineering Category for 2024. She is currently the IEEE CEDA DATC Chair and has served on several conference TPCs and organizations committees. Her current research explores the application of machine learning algorithms for electronic design automation and environmentally sustainable computing.
Abstract
"AI for Physical Design: Overcoming EDA Bottlenecks in Optimization, Analysis, and Automation"
Physical design and electronic design automation (EDA) involve three fundamental tasks: optimization, analysis, and automation. Optimization tasks often rely on domain-specific heuristics with limited generalizability, analysis tasks are time-consuming and computationally expensive, and automation remains challenging due to the ad hoc nature of EDA flows. These limitations have often left significant power, performance, and area (PPA) improvements on the table—resources that are increasingly critical for today’s compute-intensive machine learning (ML) applications. EDA and physical design workflows encompass diverse types of data, including layout images, netlist graphs, and textual design flows. Our work explores how different AI models—tailored to each of these data modalities—can enhance EDA. Convolutional neural networks (CNNs) are applied to image-like layout data for fast and accurate analysis. Graph neural networks (GNNs) model netlist structures to support timing prediction and circuit optimization. Large language models (LLMs) automate tool usage by generating scripts and enabling conversational interfaces and chatbots. Through these targeted applications, we demonstrate how AI can accelerate and improve the effectiveness of physical design workflows.