Semiconductor Fab Scheduling with Self-Supervised and Reinforcement Learning

• DOI: 10.48550/arxiv.2302.07162
• Publication Date: 2023-01-01
• Authors: Pierre Tassel; Benjámin Kovács; Martin Gebser; Konstantin Scheko...; Patrick Stöckermann; Georg Seidel
• Keywords: Semiconductor device fabrication; Tardiness; Production schedule; Semiconductor Industry

Semiconductor manufacturing is a notoriously complex and costly multi-step process involving long sequence of operations on expensive quantity-limited equipment. Recent chip shortages their impacts have highlighted the importance semiconductors in global supply chains how reliant those our daily lives are. Due to investment cost, environmental impact, time scale needed build new factories, it difficult ramp up production when demand spikes. This work introduces method successfully learn schedule semiconductor facility more efficiently using deep reinforcement self-supervised learning. We propose first adaptive scheduling approach handle complex, continuous, stochastic, dynamic, modern models. Our outperforms traditional hierarchical dispatching strategies typically used plants, substantially reducing each order's tardiness until completion. As result, yields better allocation resources process.