Dynamic feature analysis of an industrial PECVD tool with connection to operation-dependent degradation modeling

An analysis that is based on the monitoring of dynamic features from in-situ sensors of an industrial PECVD tool is presented. Linear Discriminant Analysis is used to determine which features are the most sensitive to various changes in the tool condition. The concept of Confidence Values (CVs) is used to quantify statistical changes of these dynamic features as the condition of the tool changed. Two data sets were collected from a PECVD tool in the facilities of a well-known equipment supplier. Dynamic features coming from the RF plasma power and matching capacitors’ sensors are shown to be sensitive to various changes in the cleaning cycles for Si-N, Si-O₂, and TEOS depositions. Quantifying the statistical distributions of the sensitive sensor features during tool condition changes is important for determining which sensor features are necessary to monitor in order to predict the tool chamber health. Results show that these RF plasma sensors could be used to track changes inside the tool chamber.