Multi-Agent AI Rewrites a Million-Line Chip Design Tool

This episode explores "Autonomous Evolution of EDA Tools: Multi-Agent Self-Evolved ABC," a paper from NVIDIA Research and the University of Maryland examining whether a multi-agent LLM system can autonomously rewrite ABC, the 1.2-million-line, four-layer open-source logic synthesis and verification engine that underlies most open-source ASIC and FPGA toolchains. The discussion traces this work's lineage from DeepMind's AlphaEvolve, which evolved small isolated code kernels, through NVIDIA's own SATLUTION, which scaled the approach to a full SAT solver, and examines why editing a codebase as large and interdependent as ABC — where area, delay, and depth trade off across cross-module dependencies — demands a genuinely different architecture rather than just more iterations of the same loop. That architecture centers on a planning agent coordinating three specialized coding agents for flow tuning, technology mapping, and logic minimization, plus a pre-evolution stage where the system surveys the literature and selects its own scaffolding — including Cunxi Yu's prior FlowTune work and the SLAP mapper — without any heuristics hand-injected by the authors. The conversation digs into why correctness is uniquely non-negotiable here, since a synthesized circuit must be formally equivalent to spec rather than merely probabilistic, making this an unusual case of applying neural methods to edit, rather than replace, one of computing's last hand-engineered, non-learned domains. Listeners interested in AI-driven code evolution, chip design tooling, or the limits of LLM agents on large real-world codebases will find the debate over scale, architecture, and risk especially engaging. Sources: 1. Autonomous Evolution of EDA Tools: Multi-Agent Self-Evolved ABC — Cunxi Yu, Haoxing Ren, 2026 http://arxiv.org/abs/2604.15082v1 2. AlphaEvolve: A Coding Agent for Scientific and Algorithmic Discovery — Alexander Novikov, Ngân Vũ, and collaborators (Google DeepMind), 2025 https://scholar.google.com/scholar?q=AlphaEvolve%3A+A+Coding+Agent+for+Scientific+and+Algorithmic+Discovery 3. SATLUTION (LLM-agent self-evolution of a SAT solver codebase) — NVIDIA Research, 2025/2026 https://scholar.google.com/scholar?q=SATLUTION+%28LLM-agent+self-evolution+of+a+SAT+solver+codebase%29 4. DRiLLS: Deep Reinforcement Learning for Logic Synthesis — Abdelrahman Hosny, Soheil Hashemi, Mohamed Shalan, Sherief Reda, 2020 https://scholar.google.com/scholar?q=DRiLLS%3A+Deep+Reinforcement+Learning+for+Logic+Synthesis 5. A Graph Placement Methodology for Fast Chip Design — Azalia Mirhoseini, Anna Goldie, et al. (Google), 2021 https://scholar.google.com/scholar?q=A+Graph+Placement+Methodology+for+Fast+Chip+Design 6. Autonomous Code Evolution Meets NP-Completeness (SATLUTION) — Cunxi Yu, Rongjian Liang, Chia-Tung Ho, Haoxing Ren, 2025 https://scholar.google.com/scholar?q=Autonomous+Code+Evolution+Meets+NP-Completeness+%28SATLUTION%29 7. FunSearch: Making new discoveries in mathematical sciences using large language models — Bernardino Romera-Paredes, Mohammadamin Barekatain, Alexander Novikov, et al. (Google DeepMind), 2024 https://scholar.google.com/scholar?q=FunSearch%3A+Making+new+discoveries+in+mathematical+sciences+using+large+language+models 8. FlowTune: End-to-End Automatic Logic Optimization Exploration via Domain-Specific Multi-Armed Bandit — Walter L. Neto, Yingjie Li, Pierre-Emmanuel Gaillardon, Cunxi Yu, 2023 https://scholar.google.com/scholar?q=FlowTune%3A+End-to-End+Automatic+Logic+Optimization+Exploration+via+Domain-Specific+Multi-Armed+Bandit 9. SLAP: A supervised learning approach for priority cuts technology mapping — Walter Lau Neto, Matheus T. Moreira, Yingjie Li, Luca Amarù, Cunxi Yu, Pierre-Emmanuel Gaillardon, 2021 https://scholar.google.com/scholar?q=SLAP%3A+A+supervised+learning+approach+for+priority+cuts+technology+mapping 10. DAG-aware synthesis orchestration — Yingjie Li, Mingju Liu, Haoxing Ren, Alan Mishchenko, Cunxi Yu, 2024 https://scholar.google.com/scholar?q=DAG-aware+synthesis+orchestration 11. EvoPlace: Evolution of Optimization Algorithms for Global Placement via Large Language Models — Xufeng Yao, Jiaxi Jiang, Yuxuan Zhao, Peiyu Liao, Yibo Lin, Bei Yu, 2026 https://scholar.google.com/scholar?q=EvoPlace%3A+Evolution+of+Optimization+Algorithms+for+Global+Placement+via+Large+Language+Models 12. Agentic AI for Physical Design R&D: Status and Prospects — Amur Ghose, Andrew B. Kahng, Sayak Kundu, Bodhisatta Pramanik, 2026 https://scholar.google.com/scholar?q=Agentic+AI+for+Physical+Design+R%26D%3A+Status+and+Prospects 13. MapTune: Versatile ASIC Technology Mapping via Reinforcement Learning Guided Library Tuning — Mingju Liu, Daniel Robinson, Yingjie Li, Johannes Maximilian Kuehn, Rongjian Liang, Haoxing Ren, Cunxi Yu, 2026 https://scholar.google.com/scholar?q=MapTune%3A+Versatile+ASIC+Technology+Mapping+via+Reinforcement+Learning+Guided+Library+Tuning 14. Machine-Learned Algorithmic Improvement — Ivan Smirnov et al., 2023 https://scholar.google.com/scholar?q=Machine-Learned+Algorithmic+Improvement Interactive Visualization: Multi-Agent AI Rewrites a Million-Line Chip Design Tool