Agentic Hardware Design as Repository-Level Code Evolution
The paper presents AGILE, a multi-agent framework for automated RTL hardware design that treats the process as repository-level code evolution, achieving a 91.5% pass rate on 100 designs with 98.8% of generated code cycles being functionally correct.
Background
This paper introduces an AI agent that treats hardware design (e.g., chip layouts, RTL code) as evolving code in a repository — analogous to how software is iteratively developed. The key contribution is framing hardware design as a repository-level evolutionary process that an LLM-based agent can autonomously navigate, applying changes, running simulations, and debugging across versions. This matters because hardware design is currently far more manual and less automated than software development; the approach aims to bring similar generative AI benefits to chip and system design, potentially accelerating the creation of complex hardware.