NVIDIA 与 ChipAgents 发布芯片设计基准NVIDIA, ChipAgents Launch Chip Design Benchmark
半导体行业消息：NVIDIA 与 ChipAgents 推出 Build-A-Chip-Bench 基准，用于检验 AI 能否完成从规格到实体验证的完整芯片设计流程。ChipAgents 表示，初测中表现最好的 AI 代理由 Anthropic Fable 5.1 驱动，仅得 16.6%，显示端到端自动化仍有明显落差。该基准考察 AI Agent 能否在多个设计阶段持续推论与决策，同时满足功能、时序以及功耗、效能与面积（PPA）要求，范围涵盖架构与微架构、RTL、功能验证、逻辑合成、布局与绕线、时钟树合成、时序与功耗签核及实体验证。Semiconductor news: NVIDIA and ChipAgents launched Build-A-Chip-Bench to test whether AI can complete a full chip flow from specification to physical verification. ChipAgents said the best-performing AI agent in initial tests, powered by Anthropic Fable 5.1, scored only 16.6%, showing a clear gap in end-to-end automation. The benchmark checks whether AI agents can keep reasoning and deciding across design stages while meeting functional, timing and power-performance-area (PPA) requirements, covering architecture and microarchitecture, RTL, functional verification, logic synthesis, place-and-route, clock tree synthesis, timing and power signoff, and physical verification.
