Back to the Future: Rethinking EDA Infrastructure for Agentic Systems in Chip Design Verification
Je Yang (Columbia, during an internship at Google) with Ivan Lobov (Google DeepMind) and Thomas Karpati (Google) present Back-to-the-Future (BTTF), which converts RTL simulation waveforms into a SQLite database so agents can answer debugging questions with SQL; it appears at the NeurIPS 2026 AI for Chip Design workshop.
Ask this paper
Gap. About 74.6% of LLM-for-hardware work targets RTL generation, while verification takes over 70% of the chip development cycle and waveform dumps are far too large for any context window.
Approach. Waveforms are normalized into a vendor-agnostic relational schema, and a multi-agent engine plans, writes queries and audits answers for root-cause analysis and formal checks.
Preprocessing. Suppressing clock insertion shrinks the store by 66.2% and cuts query latency while keeping 98.51% of functional toggle coverage.
Results. On 150 expert-annotated queries BTTF reaches 95.33% execution accuracy, and the multi-agent design scores 17.3% higher than a single agent on a five-criteria quality audit.
Abstract
The unprecedented computational scale of modern artificial intelligence depends on complex, multi-billion-transistor Systems-on-Chip, yet the workflows that verify these chips remain stubbornly manual. Although Large Language Models (LLMs) have made rapid inroads into Electronic Design Automation (EDA), approximately 74.6% of existing studies target static Register-Transfer Level (RTL) code generation, leaving post-simulation verification and interactive waveform debugging largely untouched. We introduce Back-to-the-Future (BTTF), an end-to-end agentic framework that closes this infrastructural gap. BTTF distills massive, unstructured simulation dumps into a normalized relational SQLite database and couples it with a collaborative multi-agent orchestration engine that translates natural-language verification queries into schema-aware SQL while correlating signal anomalies with versioned RTL repositories. Across a 150-query benchmark, BTTF attains 95.33% execution accuracy, charting a practical path toward autonomous EDA verification.