Dynamic Lookahead Generation for Optimized Scheduling and Dispatching in Front-end Semiconductor Manufacturing

Semiconductor manufacturing requires timely and informed decisions to manage high-mix re-entrant flows under tight constraints, thereby fostering efficient production. This thesis introduces a machine learning framework for step-level cycle time prediction to forecast lot arrivals at downstream resources, providing views of future fab states. The resulting lookaheads support proactive prioritization, capacity alignment, and bottleneck mitigation, and can be integrated with heuristic schedulers or reinforcement learning dispatching agents to systematically optimize policies under the extreme stochasticity that this industrial context entails.