Researchers from Politecnico di Milano and STMicroelectronics published a technical paper titled “Event-Driven Reinforcement Learning Enables Long-Horizon Control in Semiconductor Fabrication.”
The paper proposes a deep reinforcement learning framework for multi-objective policy optimization in semiconductor manufacturing, where heterogeneous wafers move through hundreds of process steps across complex equipment networks. The researchers formulate fab control as a centralized-agent problem and represent system evolution as an event-driven temporal process. They develop an event-driven temporal-difference formulation that can be integrated with different policy-optimization methods and test it using high-fidelity simulations of industry-real operating scenarios. Across offline and online training settings, the agents showed consistent gains in throughput and utilization, supporting the framework’s scalability and transferability for controlling event-driven complex adaptive systems.
Find the technical paper here. June 2026.
Yeganeh, Yavar, Mahsa Shekari, Nicla Frigerio, Daniele Pagano, and Andrea Matta. “Event-Driven Reinforcement Learning Enables Long-Horizon Control in Semiconductor Fabrication.” arXiv, June 2026. https://doi.org/10.48550/arXiv.2606.10705.
Leave a Reply