D2D-LLM+: Unified Translation Between Design Rules/Manuals and DRC - Bridging Inconsistencies for Accurate Implementation

The accurate and consistent translation of design rules/manuals into machine-readable Design Rule Check (DRC) formats is critical for ensuring the manufacturability and reliability of integrated circuits. However, the process is often hindered by inconsistencies in the language and structure of design rules/manuals, which lead to errors in DRC implementation across various Electronic Design Automation (EDA) tools. This paper presents a new tool D2D-LLM+, that leverages large language models to automate the translation of DRC codes into design rules/manuals compatible with leading EDA tools such as Mentor, Cadence, Synopsys, and open-source platforms like KLayout and Magic. Key challenges addressed include the hallucination and stability issues caused by inconsistent terminology and ambiguous language, as well as the visualization challenges of design rules. Additionally, we explore strategies to enhance large language models (LLM) learning through dataset augmentation, improving result accuracy, and increasing the stability and efficiency of LLM models through various prompt engineering techniques. Through extensive validation, we demonstrate the effectiveness of our approach in reducing errors and streamlining the DRC workflow, paving the way for more reliable IC design processes.