Cluster based analytical method for the lot delivery forecast in semiconductor fab with wide product range

The usual forecast method in semiconductor industry is simulation. Due to the manufacturing environment, the number of processes and the multitude of disturbing factors the development of high-fidelity simulation model is time-consuming and requires a huge amount of high quality basic data. The simulation facilitates a detailed prediction possible, but in many cases this level of detail of the forecast information is not required. In this paper, we present an alternative forecast method. It is considerably faster and the results for a subset of parameters are comparable to simulation. The solution does not need a complete fab model but a limited mathematical system and some fast algorithms which make the forecast of important parameters or characteristics possible. The prediction is based completely on statistics extracted from historical lot data traces. It is already implemented and tested in a real semiconductor fab environment and we also present some validation results.