Iterative Backward Elimination PLSR: A novel PLS-based modeling technique to eliminate noise components for VM solutions

Virtual Metrology (VM) solutions are becoming increasingly popular in the semiconductor manufacturing industry with the growing demand for cost reduction and cycle-time improvements at advanced technology nodes. Standard statistical procedures such as Principal Component Analysis (PCA), Partial Least Squares Regression (PLSR) are widely employed in building VM models and analysis of wafer fab data. These mathematical tools assist the fab engineers to identify critical process variables and provide robust equipment monitoring solutions. However, the prediction accuracies of standard PLS versions like the Non-Linear Iterative Partial Least Squares (NIPALS) is limited by their inability to suppress the noise components. We propose an Iterative Backward Elimination PLSR (IBE-PLSR), a novel modeling technique with noise reduction capability, as a real-time solution with superior performance for VM problems. A case study for the prediction of Chemical Vapor Deposition (CVD) film thickness using the IBE-PLSR is also presented.