Virtual Metrology using Transfer Learning with Domain Knowledge

Virtual Metrology (VM) is essential in semiconductor manufacturing for predicting wafer properties using sensor data. However, building accurate VM is challenging due to issues like high dimensionality and nonstationarity. Traditional methods such as Lasso perform feature selection and parameter estimation simultaneously; however they may conflict with domain knowledge. We introduce Weighted Transfer Lasso, a novel method that incorporates domain knowledge into the regression model. Experiments using reactive ion etching data demonstrate improved accuracy and consistency compared to traditional methods, highlighting its potential for reliable VM updates.