D2D-GPT: Leveraging Incremental Learning GPT for Seamless Design Rule Conversion Across EDA Tools
Chao Wang, Yunxiang Zhang, Wangzilu Lu, Jiajie Huang, Qing Zhang, Minghui Yin, Yuhang Zhang, Zhiqiang Li, Yongfu Li · 2024
This paper investigates the use of fine-tuned ChatGPT, in Electronic Design Automation (EDA) for Design Rule Checking (DRC). As integrated circuits grow in complexity, so do the design rules, exacerbated by the diversity of EDA tools with unique DRC specifications. We introduce D2D-GPT, a unified tool leveraging ChatGPT’s language capabilities to translate DRC specifications across different EDA platforms. This research focuses on optimizing input datasets - native DRC rules - to improve ChatGPT’s translation accuracy and efficiency. Our objective is to streamline the EDA workflow, bridge gaps between various tools, and pave the way for more cohesive design processes in the semiconductor industry.