Phase Partition Based Virtual Metrology for Material Removal Rate Prediction in Chemical Mechanical Planarization Process

Abstract

In semiconductor manufacturing, the average Material Removal Rate (MRR) in Chemical Mechanical Planarization (CMP) process is important but difficult to measure. A useful method to predict MRR is to build a data-driven model for virtual metrology and a common step in virtual metrology is feature generation. However, few researches notice that the CMP process is a multi-phase batch process. Variable correlation varies with the phase variation. Thus, feature generation for the CMP process should consider the phase change to make the most of the process data. Inspired by this view, a phase partition-based virtual metrology method is proposed to predict MRR. In the proposed methodology, a novel phase partition and phase match method is first proposed to identify the stable phases in the batch process. And then statistical features are extracted from the time series of process variables and the filter-based feature selection method is designed to remove the irrelevant and redundant features. Based on the selected feature subset, a regression model is trained to predict the MRR. The effectiveness of the proposed method is evaluated on a challenge dataset.