IC3-Evolve: Proof-/Witness-Gated Offline LLM-Driven Heuristic Evolution for IC3 Hardware Model Checking
arXiv cs.AI / 4/7/2026
📰 NewsIdeas & Deep AnalysisTools & Practical UsageModels & Research
Key Points
- The paper introduces IC3-Evolve, an automated offline code-evolution framework that uses an LLM to propose small, auditable, slot-restricted patches to an IC3 (PDR) hardware model checker implementation.
- It enforces correctness via proof-/witness-gated validation, requiring independently checkable certificates for SAFE results and replayable counterexample traces for UNSAFE results to prevent unsound changes.
- Because LLM inference is used only during offline patch search, the final deployed artifact is a standalone evolved checker with no runtime ML/LLM dependency or inference overhead.
- Experiments on the HWMCC benchmark show that the approach can reliably discover practical heuristic improvements and generalize to additional unseen public and industrial benchmarks under strict correctness gates.



