Post by Prash Kharel

Could AI agents maximize the number of photonic test sites in a multi-project-wafer (MPW) run, where even 10 mm² of silicon can cost tens of thousands of dollars? We gave our autonomous design agent this task: one agent plans the design-of-experiments (DoE), a second packs the die, scoring every step against a real foundry DRC deck, a geometric-collision check, and an orientation gate that keeps every site aligned for the wafer prober. It iteratively filled a fixed 8 × 3 mm die with 306 test structures, every one placed, routed, and DRC-clean, at 82% area occupancy, in minutes. By hand that can take several hours. What we are seeing is that any time you can frame a problem with a clear figure of merit, a good starting point, and good guardrails, an agent can loop and iteratively hill-climb toward that goal. Learn more: Blog: https://lnkd.in/e8wtgDKn The full loop is open source. Point your agent (Claude Code, Cursor, or VS Code) at the repo below and give it a try. Code: https://lnkd.in/eNFs-P4P #PhotonicDesign #AgenticAI #SiliconPhotonics #AgenticPhotonicDesignAutomation #MPW #CPO