Reproduce “A Generalizable Framework for Building Executable Domain-Specific LLMs under Data Scarcity: Demonstration on Semiconductor TCAD Simulation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Di Wang, Zhenhua Wu, Yu Liu, Kai Chang, Shaohua Wu
TcadGPT trains a compact model for Technology Computer-Aided Design (TCAD) script generation using documentation-derived question-answer data and preference training on structured representations of verified simulator decks. It scores 85.6% on a 264-question benchmark, versus GPT-4o at 46.6%. On a separate test, its DPO-aligned model passes SDE syntax checks for 16/20 instructions at Pass@3; this measures syntax acceptance, not completed simulation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.