Researchers from Columbia University, Google DeepMind, and Google describe an agentic EDA framework aimed at the less-automated side of chip design verification. Their paper, "Back to the Future: Rethinking EDA Infrastructure for Agentic Systems in Chip Design Verification," notes that while LLMs have entered EDA, about 74.6% of existing studies focus on static RTL code generation, leaving post-simulation verification and interactive waveform debugging mostly manual.
To close that gap, the team built BTTF, which converts large unstructured simulation outputs into a normalized relational SQLite database. A multi-agent orchestration engine then translates natural-language verification queries into schema-aware SQL, and can correlate signal anomalies with versioned RTL repositories. That design lets agents execute queries over waveform data rather than just generate code.
The researchers benchmarked BTTF on 150 queries and report 95.33% execution accuracy. Because this is a single source, there are no diverging accounts to compare. The work is described as a practical step toward autonomous EDA verification.