SIMULATIONS FOR OPTIMIZING DISPATCHING STRATEGIES IN SEMICONDUCTOR FABS USING MACHINE LEARNING TECHNIQUES

Optimizing operations of semiconductor manufacturing plants is a tremendous challenge due to the complexity and scale of real-world problem instances. Simulations are widely used for prototyping, evaluating, comparing, and verifying new control strategies, reducing the costs, risks, and time required for the development. We present an open-source, customizable discrete-event simulator tool for the semiconductor industry, simulating real-world-scale problems based on open datasets incorporating the challenging characteristics and constraints of the process. The simulator provides a general, customizable interface to allow benchmarking of various methods. A reinforcement learning environment is also bundled with the toolbox. Using our software, we develop evolutionary algorithms and reinforcement learning-based dispatching strategies and compare them to heuristics widely adapted in the industry.