Approximation for waiting time and its application to fab-level preventive maintenance plan optimization

Abstract

In semiconductor manufacturing, preventive maintenance (PM) planning is key to operate the system without unanticipated breakdown. Due to the complexity and diversity of the semiconductor manufacturing process, PM planning is complex. So far, PM planning research has focused on separately optimizing the PM plan for each toolset. However, there are many toolsets in a fab whose performance is correlated with each other, so individual PM planning may yield a sub-optimal solution. By jointly optimizing the toolsets at the fab-level, we can obtain the optimum PM plan of the system. In this study, we propose two main findings. First, we modify existing approximation for the cycle time in queueing network consisting of G/G/1 non-preemptive priority (NPPR) queues. The approximation captures key features of semiconductor manufacturing facilities. We propose the simulation result using the MIMAC datasets which are based on industrial data. By this, we demonstrate that the new approximation equation lead to more accurate total cycle time prediction than the other method in general. Second, based on this approximation, we develop a nonlinear optimization model to determine PM plans at the fab-level. Simulation with the MIMAC datasets are also conducted to demonstrate that total cycle time improvement is possible via PM planning optimization. This study could be expanded to using G/G/m NPPR queues.