Arcane: Assertion Reduction Framework for Hardware Verification
Arcane is a novel assertion reduction framework designed to mitigate simulation overhead in assertion-based verification (ABV) by eliminating redundant assertions. It combines a two-tier semantic clustering approach for accurate classification of large assertion sets with Monte Carlo Tree Search (MCTS) to explore optimal rule-application sequences. Tested on Assertionbench, Arcane reduces assertion count by up to 76.2% while fully preserving formal coverage and mutation-detection ability, achieving a simulation speedup of 2.6x.
Key facts
- Arcane integrates two-tier semantic clustering and MCTS-guided rule exploration.
- Achieves up to 76.2% reduction in assertion count.
- Preserves formal coverage and mutation-detection ability.
- Simulation speedup of 2.6x.
- Evaluated on Assertionbench dataset.
- Addresses redundancy in LLM-based assertion generation.
- Proposed as an efficient assertion reduction framework.
- Targets hardware design verification.
Entities
—