Agentic Hardware Design as Repository-Level Code Evolution
2026-06-26 • Hardware Architecture
Hardware ArchitectureArtificial Intelligence
AI summaryⓘ
The authors introduce HORIZON, a system that helps evolve hardware designs automatically by treating the design files like code in a software project. It uses a special Markdown setup to create a package with knowledge and tools to test and improve the hardware design in a loop that runs by itself. They tested HORIZON on several benchmark suites and achieved complete success without human intervention. However, the authors note that while promising, this approach does not solve hardware design automation entirely and further work is needed.
hardware designself-evolving agentrepository-level code evolutionMarkdown harnessgit worktreebenchmark suiteshardware design automationEDA (Electronic Design Automation)agentic AIchip design
Authors
Cunxi Yu, Chenhui Deng, Nathaniel Pinckney, Brucek Khailany
Abstract
We present HORIZON, a self-evolving agent framework that treats hardware design as repository-level code evolution. A Markdown harness is compiled into a project pack containing domain knowledge, an executable evaluator, an acceptance predicate, and a git/runtime policy; a hands-free agent loop then evolves an isolated git worktree, using repository operations for state management, tracing, and replay. This extends prior works of repository-scale self-evolution from EDA software systems, to hardware-design artifacts themselves. We evaluate our approach on ChipBench, RTLLM, Verilog-Eval, and nine CVDP categories, achieving 100\% benchmark completion across all suites with a fully hands-free agentic loop. However, we do not claim that agentic AI for hardware design is solved: these benchmarks are controlled proxies for a much broader engineering problem in chip design. Section~\ref{sec:discuss} examines the limitations of the current study and highlights open research challenges.