PICasso AI Optimizes Silicon Photonic Device Design.

Researchers have introduced PICasso, an AI-assisted framework designed for the autonomous synthesis, verification, and optimization of silicon photonic devices. This innovative system allows engineers to generate photonic integrated circuits (PICs) directly from natural language specifications. PICasso integrates a structured pipeline that converts natural language into design files (NL -> YAML -> GDS), incorporates process design kit (PDK) knowledge, automates placement and routing, and performs design rule checking (DRC), layout versus schematic (LVS) validation, and photonic simulation. To rigorously evaluate AI-driven photonic design, the team developed PIC-Set, a benchmark of 36 parameterized PIC design tasks. Benchmarking against state-of-the-art Large Language Models (LLMs), PICasso demonstrated substantial improvements in end-to-end specification satisfaction, achieving high structural and functional accuracy on complex circuits. Furthermore, the framework consistently reduced circuit insertion loss through simulation-guided optimization, showcasing its ability to produce manufacturable layouts with competitive runtimes compared to manual design workflows.