Flexcompute Unveils Autonomous Agent Loop for Photonic Chip Design

Flexcompute has introduced a fully autonomous agent loop for photonic chip design, announced in a press release. The system allows AI agents to propose designs, run physics simulations, verify fabrication constraints, and iterate independently, producing tapeout-ready layouts in hours.

The platform, called PhotonForge, connects design agents directly to foundry process design kits, layout, and circuit simulation. It integrates with the Tidy3D GPU-native solver to accelerate physics simulations and ensure fabrication awareness from the first design iteration. The agents operate without human input between steps, completing tasks that previously required senior photonic engineers.

Flexcompute’s system combines several technologies, including Python APIs, the Flexagent MCP plugin, and a dedicated GPU cluster for parallel simulations. It supports multiple foundry process design kits across silicon, silicon nitride, and thin-film lithium niobate. The agents leverage reasoning models from Anthropic, OpenAI, and others, and can standardize across frameworks such as NVIDIA NemoClaw.

According to the company, recent demonstrations covered optical, electrical, and thermal components, with complete photonic circuit layouts generated autonomously. The new workflow extends Flexcompute’s physics stack beyond photonics to electromagnetics, heat transfer, charge transport, and computational fluid dynamics.