LLM Based Physical Verification Runset Generator
Luis Francisco, Srini Arikati · 2024
The complexity in design rule description and coding is drastically increasing as technology nodes advance. This complexity makes the process of implementing the physical verification (PV) rule checks more time-consuming and susceptible to human error, creating the need to explore alternate methods to improve the runset creation process. The work presented proposes a generative AI solution that uses Large Language Models (LLMs) to interpret rule descriptions and generate design rule check decks (runsets) in a language that a PV tool can interpret. The LLM is fine-tuned with existing design rule manuals and runsets. After post-processing the LLM output, the presented solution can generate rules implementation with up to 97% accuracy. The proposed solution can be used as a runset writer Co-Pilot to help develop the new physical verification runsets.