This award provides funding to investigate how the random elements of semiconductor manufacturing impact overall factory productivity. Algorithms will be created and tested for simulating dependent, transient random processes such as product demand, tool reliability, operator availability, and step yield. Statistical procedures will be developed for determining the types and amount of data that are needed to accurately represent these processes in factory queueing models and simulations. A central part of the research program will be designing efficient experimental procedures for optimizing the performance and validity of factory models. Experimental and analytical techniques for using these models to support the optimal timing of resource investments such as tool purchases and personnel cross-training will also be developed. If successful, this research will provide engineers in the semiconductor industry with fundamental knowledge necessary for the creation of valid models and efficient analytical tools useful in designing semiconductor fabrication facilities. A key to improving the cycle times and throughput of these multi-billion dollar factories is the identification and control of the random factors of production. Conventional semiconductor factory models assume that independent and identically distributed random input processes produce steady-state output processes with known performance objectives, economic parameters, and constraints. The realities of semiconductor fabrication are quite different. In the semiconductor industry, capacity requirements, product demands, and competitive cycle times are always changing, as are the very products, processes, and personnel skills involved. Currently, factory-level models are overly optimistic because they ignore many of the transient dependencies in the random phenomena found in real semiconductor manufacturing. This consistent optimism has had serious economic consequences. Valid factory models based on this research will be critical to the continued economic health of this vital industry. Furthermore, the costs of data collection are enormous. This research will provide guidelines for estimating how much data is necessary and how it should be used in engineering design. The modeling methodologies developed in this research will also help engineers to determine optimal settings of critical factory operating parameters that can make the difference between financial success and disaster.
