MappingEvolve: LLM-Driven Code Evolution for Technology Mapping

MappingEvolve is an open-source framework that uses large language models (LLMs) to directly evolve technology mapping code in logic synthesis. The method abstracts the mapping process into optimization operators and employs a hierarchical agent-based architecture with a Planner, Evolver, and Evaluator. Experiments show a 10.04% area reduction over ABC and 7.93% over mockturtle, with 46.6%–96.0% S_overall improvement on EPFL benchmarks, while navigating the area–delay trade-off. The code and data are available on arXiv.