FVDebug: An LLM-Driven Debugging Assistant for
Automated Root Cause Analysis of Formal Verification Failures · 2025
Read FVDebug: An LLM-Driven Debugging Assistant for by Automated Root Cause Analysis of Formal Verification Failures (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-15906.txt.
Description
Yunsheng Bai leads four co-authors in automating the hunt for why a hardware property failed formal checks. Such debugging is one of the slowest steps in chip design. Engineers trace counterexamples over many cycles. They read waveforms and cross-check design specs by hand. One bug can take hours or days to find.
Their system, FVDebug, joins waveforms, design code, and specs in one analysis. A causal graph first structures the failure trace. A scanner then uses batched language model checks, with for-and-against prompts, to flag suspect nodes. A rover agent explores the graph and writes a plain causal story. A fix tool proposes concrete code repairs. On open benchmarks, the system scores high on hypothesis quality and fix rates. The team also reports results on two production-scale cases from industry.
Catalog record
| CREATOR | Automated Root Cause Analysis of Formal Verification Failures |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 70 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-15906