| Previous research on semiconductor manufacturing has concluded that input control, the decision of when to release a new lot into the system, is significantly more important than scheduling, the decision of which lot to process next at a particular work station. Despite the conclusions drawn by researchers, many factories (known as fabs ) invest a great deal of effort in scheduling decisions. This study addresses two questions related to this phenomenon. First, why do some fabs emphasize scheduling over input control? That is, what are the characteristics and performance objectives of fabs that emphasize scheduling? Second, what is the optimal scheduling policy for a system with a given set of characteristics and performance objectives?; The first question is addressed by analyzing data from U.C. Berkeley's Competitive Semiconductor Manufacturing survey. The results show that demand variability, process technology, and the number of product types manufactured are all associated with high emphasis on shop-floor scheduling. This clearly indicates a need for scheduling policies that account for multiple products. Previous research has focused almost exclusively on single-product, high-volume fabs with an emphasis on cycle time performance.; The second part of the dissertation addresses the need for scheduling policies for multiple-product systems with an emphasis on yield-related performance measures. Results indicate that under reasonable assumptions about the expected profits and the state transition probabilities, an intuitive, monotone policy will be optimal. Specifically, the optimal policy calls for producing progressively less-sensitive products as the machine deteriorates, and there is a threshold at which it is optimal to stop production and repair the machine. Different solution methods, each using different information about the machine state, are compared. In the cases studied, combining the production and maintenance scheduling decisions increased expected profit by an average of more than 20 percent.; The final part of the dissertation extends the results of the single-machine model to a multiple-machine, multiple-stage system. Results show that substantial benefits can be attained by incorporating information about the equipment condition in production scheduling decisions. The bigger the differences between products, the bigger the opportunity for improvement. In the cases studied, improvements ranged from 3 percent to more than 600 percent.; This research has significant implications for managers as well as scholars. For managers, the policies developed provide significant improvement over heuristic scheduling rules currently employed in many fabs and shed light on important relationships between production, maintenance, and engineering. For scholars, the dissertation extends the boundaries of semiconductor manufacturing research by addressing issues not previously considered. More importantly, it establishes a link between two important areas of operations management research: production scheduling and equipment maintenance. (Abstract shortened by UMI.)... |
