The progress of modern integrated photonics is often limited by design workflows that rely on manual simulation and expert intuition. Although inverse design offers an alternative, it still requires expert supervision and lacks end‑to‑end automation. To tackle these challenges we introduce VortexChat, an agentic framework that autonomously performs inverse design of photonic devices directly from natural‑language specifications.
VortexChat couples a large language model (LLM) decision agent with topology generation, gradient‑based refinement, and full‑wave electromagnetic simulation, forming a closed‑loop architecture. With minimal human intervention the system iteratively decomposes design objectives, orchestrates computational tools, and updates strategies based on simulation feedback.
Constrained by the absolute metrics of the Vortex100 benchmark, VortexChat autonomously generates devices that strictly meet all predefined performance thresholds without any human‑in‑the‑loop. As an experimental demonstration, we fabricated a broadband terahertz perfect vortex‑beam multiplexer designed entirely by VortexChat. Measurements confirm high efficiency, high mode purity, and low inter‑channel crosstalk, in strong agreement with full‑wave simulations.
These results show that an LLM agent can assume key decision‑making roles in photonic inverse design while preserving physical fidelity and fabrication feasibility, offering a scalable route toward autonomous design of complex integrated photonic systems.
Blogger's Review: VortexChat highlights the powerful synergy between LLM reasoning and physics‑based simulation, especially for multi‑objective optimization and fully automated workflows, marking a significant step forward for photonic device engineering.