Reevaluating the conventional approach for analyzing spectroscopic ellipsometry psi/delta versus time data. Additional statistical rigor may often be appropriate

• Published in: Surface and interface analysis Vol. 48; no. 4; pp. 186 - 195
• Main Authors: Bagley, Jacob D., Dennis Tolley, H., Linford, Matthew R.
• Format: Journal Article
• Language: English
• Published: Bognor Regis Blackwell Publishing Ltd 01.04.2016; Wiley Subscription Services, Inc
• Subjects: chemometrics; cluster analysis; Constants; Deltas; distance analysis; Ellipsometry; Materials selection; Principal components analysis; spectroscopic ellipsometry; Spectroscopy; Spectrum analysis; stability; Statistical analysis; Thin films
• ISSN: 0142-2421; 1096-9918
• DOI: 10.1002/sia.5938

A standard approach for confirming the stability of thin films consists of probing them over an extended period of time by spectroscopic ellipsometry (SE) and plotting the resulting psi (Ψ) and delta (Δ) values versus time at a few selected wavelengths across the spectral range. In general, if Ψ and Δ remain constant with time, or essentially constant, the material is deemed to be stable. Here, we suggest that in addition to this ‘eyeball’ approach, a statistical analysis of the data may often be appropriate. In particular, we reevaluate a set of Ψ/Δ versus time data from a sputtered bismuth–tellurium–selenium film that appear to remain essentially constant by the traditional approach, subjecting it to a distance analysis, a principal components analysis, and a cluster analysis. The application of these statistical tools, especially to range‐selected data (300–989 nm), reveals previously unobserved changes, suggesting that either the film in question or the analytical approach itself had changed during the course of the measurements. Weighted distance formulas for Ψ and Δ were also applied so that the data did not need to be range selected for the key effects to be observed. The distance approach was similarly applied (i) to a set of SE data from a thin carbon film, which revealed consistent changes in the material that were not apparent in the standard approach, and (ii) to samples that were deliberately contaminated or improperly measured. Thus, we recommend a more rigorous, statistical approach to the analysis of SE Ψ/Δ versus time data. Copyright © 2016 John Wiley & Sons, Ltd.