Endpoint Detection in Plasma Etching Using Principal Component Analysis and Expanded Hidden Markov Model

Abstract

In current semiconductor manufacturing, plasma processes such as etch and CVD take the portion at least 40% throughout of integration processes. As the feature size of integrated circuit (IC) devices continuously shrinks, detecting endpoint in low open area plasma etch process becomes more difficult. To solve this problem, a combination of Principal Component Analysis (PCA) and Expanded Hidden Markov model (eHMM) technique is applied to optical emission spectroscopy (OES) signals. Selected patterns are used in PCA, which reduces dimension of the raw data and increases gap between classes. The eHMM is employed to detect endpoint using output of PCA. The eHMM combines the semi-Markov model to enable an arbitrary distribution on the location of the change-point and the segmental HMM to model the configuration in each segment. After modeling using eHMM, real-time OES data were fed to this model to detect endpoint.