ChipAgents and Nvidia build Build-A-Chip-Bench for AI chip-design agents

ChipAgents, with contributions from Nvidia, is developing Build-A-Chip-Bench, which evaluates AI agents across the full chip-design workflow from workload specification to GDS layout using open-source tools and an open-source PDK. The best agent passes the hard gate on 65% of tasks and scores 16.6% overall.

• Benchmark spans the full flow from spec to GDS on open-source tools and PDK
• Hard gate checks functional correctness, routability, hold timing and DRC
• Best agent passes the gate on 65% of tasks, scoring 16.6% overall
• Agents struggle to revisit decisions and plan around slow tool runs