Abstract
As semiconductor foundry is built-to-order, the variation of demands of various customers affect the total quantity of customer order directly, and thus create variability on inputted wafer start on wafer fabrication facility (fab), which will affect the WIP bubble and cycle time as well as the capacity utilization and profitability. This research aims to develop a manufacturing intelligence methodology to predict customer demands based on their behaviors and historical data to extract useful information to support the decision maker for manufacturing strategy and production plan in light of demand uncertainty and market fluctuation. In particular, we proposed a manufacturing intelligence framework in which distributed lags structure and neural tree were employed to analyze the relationships among customer finished goods, order, and other decision factors. An empirical study was conducted for validation. The derived empirical rules can effectively help the decision maker to make timely production decisions given different order situations while maintaining fab utilization and cycle time well. © Springer-Verlag Berlin Heidelberg 2012.