Deep Learning-Based Dynamic Scheduling for Semiconductor Manufacturing With High Uncertainty of Automated Material Handling System Capability

Recently, the transportation capability of the automated material handling system (AMHS) has emerged as a major barrier to the semiconductor fabrication facility (FAB), because it can limit the FAB production capacity. In this study, we propose a prediction method for a machine allocation problem of production scheduling in consideration with the AMHS's constraints. The proposed method dynamically targets a machine for the next process by identifying diverse production conditions. We use a deep neural network-based dynamic scheduling method considering the overall production environment, which includes the remaining processing time, facility states, transportation time and traffic congestion, work-in-process distribution, and intermediate buffer states. To demonstrate the superiority and efficiency of the proposed method, we conducted experimental studies to compare the proposed model with the existing priority rule-based and analytic models under the static and dynamic environments. From the results, we verify that the proposed dynamic scheduling system can enhance the performance of existing AMHS and reduce machine starvation and production losses.