Reproduce “AMSnet-KG: A Netlist Dataset for LLM-based AMS Circuit Auto-Design Using Knowledge Graph RAG” 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

Yichen Shi, Zhuofu Tao, Yuhao Gao, Tianjia Zhou, Cheng Chang, Yaxing Wang, Bingyu Chen, Genhao Zhang and 4 more

Researchers from UCLA and collaborators introduce AMSnet-KG, a large-scale, annotated dataset of Analog and Mixed-Signal (AMS) circuit netlists and schematics structured as a knowledge graph. They also propose AMSgen, an automated design framework that leverages this knowledge graph with Large Language Models (LLMs) and iterative optimization to generate fully-sized AMS circuits directly from high-level specifications, successfully designing OPAMPs and comparators that meet stringent performance targets.

Key insights

  • Integrating a structured knowledge graph with LLMs mitigates "hallucination" by grounding LLM-generated designs in precise circuit data and expert annotations.
  • Incorporating expert-derived "local annotations" (e.g., symmetry constraints) into Bayesian optimization significantly reduces the design parameter search space, improving sizing efficiency and solution quality.
  • An iterative topology regeneration mechanism, which feeds quantitative simulation results back to the LLM, enables the model to refine its qualitative design strategies based on actual circuit performance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.