Advanced semiconductor production planning

This research investigates the issues of optimizing the production planning of the semiconductor wafer fabrication process in the face of arbitrary initial WIP distributions and complex process flows. Two different WIP scheduling models are developed and incorporated into the dynamic rate-based production planning model originated by Leachman. Both models schedule both initial WIP-out rates and fab-in rates so as to feasibly utilize resource capacities. Multi-objective optimization is used to achieve optimal on-time delivery performance as well as maximum discounted cash flow. These two WIP scheduling models fully cope with characteristics of advanced semiconductor process flows, e.g., re-entrant flows with non-integer cycle times, static or dynamic machine assignment constraints. The models differ in the flexibility allowed for the subsequent movement of WIP once it is scheduled to start moving.; A generalized mapping technique is introduced to calculate formulation coefficients accurately reflecting time-varying data (cycle time, yield) presented in varying formats. This technique generalizes and extends previous work by Hung and Leachman. Using detailed data on the initial WIP distribution, i.e., initial WIP levels at major and minor steps, step cycle times and factory working calendar, the technique provides an accurate assessment of workloads at bottleneck steps and accurate WIP-out schedules. Using the generalized mapping technique, discrete-time inventory balance equations are formulated that ensure exact continuous time inventory balance. A practical smaller set of discrete time inventory balance equations is generated to approximate the continuous time inventory balance. The dynamics of factory output and inventory levels of finished products are more accurately captured than in traditional formulations.; The parameters of the objective function are critically examined and a basic principle is introduced to set values of these parameters so as to avoid undesired results and ensure the results are consistent with the production planning goals, i.e., on time delivery and maximal resource utilization.; Software generating the linear programming formulations for the models described above has been implemented. This software takes as inputs data files describing the fab process technology and fab status in formats familiar to practitioners. Encouraging computational results from application of the models to actual industrial data are reported. The linear programs are solved in practical amounts of time using commercial LP software operating on a UNIX work station.