Yield Prediction with Machine Learning and Parameter Limits in Semiconductor Production
Rebecca Busch, M. Wahl, Peter Czerner, Bhaskar Choubey
Abstract
Yield is an important cost factor in wafer production. Therefore, continuous data-driven yield monitoring and optimization provides opportunities to reduce production costs. Predicting yield during production would reveal its relationships with production parameters enabling dynamic optimization with a preventive and active increase in yield. In our investigations, we will first predict the yield based on one yield critical process step and later on with the data of four process steps. We will use different machine learning methods for this. Furthermore, we will look at whether the classification into good and bad yield values with these methods provides better results for the prediction. Another point of our investigations are the parameter limits of the individual methods. We show that these can be controlled by a simple method and optimised, if necessary.