Cycle Time Prediction for Semiconductor Manufacturing via Simulation on Demand

Traditionally, competition between semiconductor manufacturers has primarily focused on product design and cost. Recently, speed of delivery has also become an important differentiator among these firms which has led to manufacturing cycle time becoming a critical performance measure. This paper presents a methodology that performs a limited set of simulation runs for a complex wafer fabrication system and then uses the results to develop metamodels that predict mean steady-state cycle time as a function of product mix and throughput. These predictions can be made on demand, i.e., without performing any additional simulation runs, for product mixes and throughput levels not previously simulated. The goal is to support medium and longrange planning by providing results with the fidelity of a detailed simulation model but with the speed of a queueing approximation or simple capacity model.