Item

Abstract

A ubiquitous goal in plasma-enhanced chemical vapour deposition (PECVD) is to describe the correlation between film properties and categorical and quantitative input variables. The correlations within the high-dimensional parameter space are described using a multivariate model. Bayesian group analysis is employed to assess the grouping structures of the set of data vectors. This allows to identify sub-groups or meta-groups of predefined groups of data sets, e.g. with respect to source gases. Outliers can be identified by the necessity to form a separate group. The Bayesian approach consistently allows the handling of missing data. The grouping probabilities were compared with classical approaches such as likelihood ratio tests, the Akaike information criterion and a Bayesian variant called Bayesian information criterion. The method was applied to PECVD data of rare-earth oxide film deposition and hydrocarbon film deposition to study the evidence of grouping structures attributed to categorical quantities such as rare-earth components or source gases and quantitative variates such as bias voltage.