EDA-Debugger: An LLM-Based Framework for Automated EDA Runtime Issue Resolution

The increasing complexity of integrated circuits (ICs) poses significant challenges for Electronic Design Automation (EDA) tools and users. Debugging EDA runtime issues can be time-consuming and frustrating due to complex error messages, extensive documentation, and limited knowledge sharing. Although open-source EDA tools like OpenROAD and OpenLane have democratized chip design, the dependence on expert knowledge and the prolonged response times for issue resolution remain significant obstacles. We introduce EDA-Debugger, a novel multi-agent Large Language Model (LLM) based framework for automated EDA runtime issue resolution to address these challenges. The EDA- Debugger uses the power of LLMs to analyze EDA log files, identify potential issues, and provide actionable solutions. By integrating EDA domain-specific Retrieval Augmented Generation (RAG), EDA-Debugger bridges the gap between designers, LLMs, and EDA tools, enhancing efficiency and accessibility for users of all skill levels. Our evaluation of a benchmark dataset of 20 typical open-source EDA issues shows a 50% improvement in issue resolution and 20.1% improvement on average score compared to GPT-4 alone. This paper outlines the architecture, functionalities, and evaluation of EDA - Debugger, demonstrating its potential to revolutionize the EDA debugging process.