PrefixAgent: An LLM-Powered Design Framework for Efficient Prefix Adder Optimization

PrefixAgent optimizes prefix adder arithmetic circuits by decomposing the design process into backbone synthesis and structure refinement. The LLM-driven framework uses function calls to iteratively refine local structures based on EDA feedback, leveraging e-graph rewriting and equality saturation to navigate the exponential design space of high-bit-width circuits.