In many industries, various critical decisions are based on demand forecasts. However, these forecasts are subject to random error. Motivated by the semiconductor industry, we lay out a scheme estimating the variance and correlation of forecast errors. We model the evolution of forecasts and the resolution of demand uncertainty over time. We do not alter given forecasts. Our scheme (SeDFAM) allows correlations across time, products and technologies. It also addresses the case of nonstationary errors due to ramps (technology migrations). It can be used to simulate future demands for production planning/capacity expansion studies. We run various experiments statistically validating SeDFAM assumptions with real life forecast data and studying its robustness against forecasting parameters.; Next we present a novel approach to compute optimal machine, shop floor and shell space expansion times under uncertain demand. Our approach considers multiple machine types and allows for positive lead times for each type. Demand is assumed to be nondecreasing in a "weak" sense. A polynomial time algorithm (FIFEX) is developed and is illustrated with demands from two different semiconductor companies and real life semiconductor fabrication plant data. Using these data, we compare FIFEX with capacity planning heuristics used in practice. We also study the effects of forecast update frequency on costs with a rolling horizon experiment.
展开▼