An LLM-based Agentic Framework For Photonic IC Design Automation (U. of Toronto, Max Planck, MIT Et Al.)

Researchers from the University of Toronto, Max Planck Institute, MIT, and others have published a technical paper on AI agents for automating photonic integrated circuit (PIC) design. They introduce the Photonics Intelligent Design and Optimization (PhIDO) framework, which uses multiple agents to convert natural-language PIC design requests into layout mask files. The study compared 7 large language models for PIC design automation, achieving success rates of up to 91% for single-device designs and approximately 57% for designs with up to 15 components. Future steps include standardized knowledge representations, expanded datasets, extended verification, and robotic automation for autonomous PIC development. The paper is available as an arXiv preprint for August 2025.