Mean cycle time approximations for G/G/m queueing networks using decomposition without aggregation with application to fab datasets

Abstract

The modern semiconductor fabricator needs both accurate and fast cycle time (CT) forecasts. Due to complexity of development and computational intractability, simulation may be supplemented by queueing network methods. In this paper, we develop extensions to approximation methods for queueing networks that are suited for fab modeling using decomposition without aggregation. We conduct simulation experiments based on a semiconductor industry-inspired dataset. For sensitivity analysis, we mainly focus on the interarrival distribution, service time distributions, and bottleneck toolset loading. The results show that the approximations predict the total CT fairly well in various cases.