Optimization of facility layout using reinforcement learning for semiconductor production

Abstract The optimization of facility layouts in constrained factory spaces significantly affects operational efficiency and performance. In semiconductor factories, the main facilities are tightly interconnected with various assistive facilities located on separate floors. This complexity requires a comprehensive consideration of the constraints and implicit rules during facility layout planning. However, most factories rely on manual planning by the production managers. To address this problem, we propose a methodology that uses reinforcement learning to optimize the layout of a semiconductor factory. The reinforcement learning method employs the double deep Q-network algorithm to optimize facility layouts while incorporating the unique constraints of semiconductor manufacturing environments. The algorithm enhances layout efficiency and supports the design process by generating multiple optimized layout solutions. Furthermore, these optimized layouts were integrated into the NVIDIA Omniverse platform, enabling users to review and modify their configurations. This study contributes to laying the groundwork for automating facility layout planning in semiconductor factories by exploring reinforcement learning-based approaches under real-world-inspired constraints.