Analysis of wafer fabrication facilities using four variations of the open queueing network decomposition model

We analyze three wafer fabrication facilities with four open queueing network decomposition models - Whitt (1983, 1985), and Bitran and Tirupati (1988), Suri et al. (1993). In the first facility, which is located in North America, the values of cycle time and WIP inventory predicted by each model are about 10% higher than the actual values observed in the facility. The estimates are on the high side because managers and operators in the facility take special action to expedite production whenever queues become too large or small. These special actions cannot be incorporated into the open queueing network models. In the second facility, which is taken from the literature, the values of cycle time and WIP predicted by each model are about 5% higher than the values reported in the literature. An investigation at a third facility located in South-east Asia identified a common manufacturing practice that caused a model assumption to be violated. New insights into the models and directions for future research are identified. We find that two of the four open queueing network models are easy to apply and give good results for facilities like the ones studied in this paper. The other two models are complex and are likely to be better-suited for more complex facilities.