Reproduce “CIll: CTI-Guided Invariant Generation via LLMs for Model Checking” 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

Yuheng Su, Tianjun Bu, Qiusong Yang, Yiwei Ci, Enyuan Tian

CIll, a framework developed at the Institute of Software, Chinese Academy of Sciences, integrates Large Language Models (LLMs) to synthesize inductive invariants for model checking, guided by counterexamples to induction. It successfully verified previously intractable properties in RISC-V NERV and PicoRV32 processor cores, demonstrating enhanced scalability for hardware formal verification.

Key insights

  • Large Language Models can effectively reason about hardware designs at a high-level abstraction, enabling the synthesis of complex invariants beyond the capabilities of purely syntactic methods.
  • Integrating LLM-based invariant generation with rigorous formal verification engines creates a robust neuro-symbolic approach that combines semantic understanding with formal guarantees.
  • Efficient LLM interaction is achieved through techniques like trace minimization and an on-demand Model Context Protocol, addressing context length limitations and focusing diagnosis on relevant information.

Summary by alphaXiv.

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