Optimize the Flow Factor in a Semiconductor Factory by use of Deep Reinforcement Learning

Semiconductor manufacturing is known for being one of the most complex industrial processes. Consequently, achieving optimal efficiency in such a complicated environment poses a significant challenge. The increasing demand for computer chips, driven by the current trend of widespread digitalization, has highlighted the importance of enhancing production efficiency. This is critical for both reducing costs and meeting the significant demand for chips in the market. For this purpose, this thesis aims to evaluate state-of-the-art Deep Reinforcement Learning algorithms for optimizing semiconductor production in the context of the simulation of a real factory. In particular, the Reinforcement Learning approach consists in learning an optimal rule for lot dispatching, one of the possible optimization problems encountered in semiconductor manufacturing. A combination of heuristic rules designed by experts is used as a reference for evaluating the agent results. The experiments conducted show that the agent is capable of matching the performance of the heuristic rule despite several limitations. Moreover, the analysis of the results provides insights useful as a starting point for further research, with the aim of outperforming the heuristic approach.