Agentic Architect Uses LLMs for Computer Architecture Design

Researchers introduced Agentic Architect, an agentic AI framework that leverages Large Language Models (LLMs) to explore and optimize computer architecture designs. The framework combines LLM-driven code evolution with cycle-accurate simulation, allowing human architects to specify optimization targets, seed designs, scoring functions, simulator interfaces, and benchmark splits. In tests covering cache replacement, data prefetching, and branch prediction, Agentic Architect matched or exceeded state-of-the-art designs. Its best evolved cache replacement design achieved a 1.062x geomean IPC speedup over LRU and 0.6% over Mockingjay (1.056x). The work was published on arXiv.