Automated generation of analytical process time models for cluster tools in semiconductor manufacturing

In this paper, we present an approach to automatically create an analytical process time model for cluster tools using real-world data. The proposed model combines advantages of simple throughput models and discrete event simulation models. We consider the effect of small lot size and the slow down effect occurring when simultaneously processed lots interfere with each other. Especially the use of Slow Down Factors depending on a certain recipe combination and start delay adequately mirrors sequential and parallel processing mode. We also describe a modeling method that automatically leads to parameterized models with high accuracy. This study presents evaluation results gained from models, which we create from and test against real-world data gathered from past equipment events. We discuss exemplary processing behaviors by means of three examples. We conclude that the proposed analytical cluster tool model is suitable to predict process times with respect to accuracy and prediction coverage.