Towards LLM4Floorplan: Agents Can Do What Engineers Do in Chip Design
Keywords: floorplanning, large language model, prompting, electronic design automation, physical design
TL;DR: An LLM-powered multi-agent system designed for floorplanning.
Abstract: Open-source tools have actively propelled advancements in physical electronic design, yet the deployment still requires substantial expertise. Recent progress in large language model (LLM)-based agents offer potential for automating physical design, but challenges remain in imparting domain-specific expertise and extracting case-specific design objectives to meet complex requirements. To address these issues, we introduce LLM4Floorplan, a multi-agent Floorplanner powered by LLMs. Unlike flow-level approaches that design workflows for multiple tasks, LLM4Floorplan is the first task-level agent specifically dedicated to a single physical design task. Specifically, we propose a simple yet effective search-cluster-based retriever that extracts the most relevant and diverse solutions from prior knowledge, drawing on essential domain-specific knowledge to ensure robust design performance. Building on the retriever, LLM4Floorplan integrates a novel Dynamic Retrieval-Augmented Thought (DRAT) prompting technique in which the LLM generation interacts with the retrieval system to precisely capture case-specific design objectives. With these innovations, LLM4Floorplan simulates the workflow of human engineers by facilitating task comprehension, model selection, hyperparameter tuning, code revisions, and performance evaluation. Extensive evaluations on public circuits with seven different LLM backbones demonstrate that LLM4Floorplan exhibits strong task comprehension and decision-making capabilities. Remarkably, for the strict requirement, LLM4Floorplan boosts the success rate from 0.250 to 0.875.
Primary Area: other topics in machine learning (i.e., none of the above)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 6038
Loading