[paper] Multi-Agent Self-Evolved ABC

Cunxi Yu and Haoxing Ren

Autonomous Evolution of EDA Tools: Multi-Agent Self-Evolved ABC

In 63rd ACM/IEEE Design Automation Conference (DAC ’26)

July 26–29, 2026, Long Beach, CA

DOI: 10.1145/3770743.3804221

Abstract: This paper introduces the first self-evolving logic synthesis framework, which leverages Large Language Model (LLM) agents to autonomously improve the source code of ABC, the widely adopted logic synthesis system. Our framework operates on the entire integrated ABC codebase, and the output repository preserves its single-binary execution model and command interface. In the initial evolution cycle, we bootstrap the system using existing prior open-source synthesis components, covering flow tuning, logic minimization, and technology mapping, but without manually injecting new heuristics. On top of this foundation, a team of LLM-based agents iteratively rewrites and evolves specific sub-components of ABC following our “programming guidance“ prompts under a unified correctness and QoR-driven evaluation loop. Each evolution cycle proposes code modifications, compiles the integrated binary, validates correctness, and evaluates quality-of-results (QoR) on multi-suite benchmarks including ISCAS 85/89/99, VTR, EPFL, and IWLS 2005. Through continuous feedback, the system discovers optimizations beyond human-designed heuristics, effectively learning new synthesis strategies that enhance QoR. We detail the architecture of this self-improving system, its integration with ABC, and results demonstrating that the framework can autonomously and progressively improve EDA tool at full million-line scale.

Fig: Overview of the multi-agent self-evolving framework for ABC. Specialized LLM agents evolve distinct subsystems (flow optimization, core algorithms, and mapping), with each iteration undergoing compilation, formal CEC verification, and full QoR evaluation. A planning agent coordinates global decisions, a coding agent implements edits, and all agents follow a shared rulebase and unified evaluation pipeline to enable coordinated, correctness-preserving improvements.

Acknowledgment: The authors would like to thank Prof. Zhiru Zhang and his students for their valuable feedback and insightful discussions.