For full functionality of this site it is necessary to enable JavaScript. Here are the instructions how to enable JavaScript in your web browser.
SIMS21, Poland 2017 - Danica Heller abstract

Danica Heller oral presentation (OA3-Tue3-3-2)

Multivariate Data Analysis to Simplify ToF-SIMS Data Analysis in an Industrial Context

Danica Heller1, Reinhard Kersting2, Michael Fartmann2, Rik ter Veen2, Carsten Engelhard3, Birgit Hagenhoff2

1 Tascon GmbH - University of Siegen, Mendelstraße 17, 48149 Münster, Germany
2 Tascon GmbH, Mendelstraße 17, 48149 Münster, Germany
3 Department Chemistry & Biology, University of Siegen, Adolf-Reichwein-Str. 2, 57076 Siegen, Germany


Time-of-Flight Secondary Ion Mass Spectrometry (ToF-SIMS) has developed into a popular tool in industry for failure analysis. Often the analyses are performed by a comparison of one 'good' and one 'bad' sample. Usually, not much is known about the chemical composition before the analysis. Therefore, a comprehensive technique, such as ToF-SIMS, is needed enabling a reliable elucidation of the chemical surface composition and thus contributing to the root cause analysis of e.g. the defect formation. However, the high amount of information included in each ToF-SIMS spectrum results in complex and time-consuming data analysis. Over the last decade, multivariate data analysis (MVA) has increasingly been used to simplify the data interpretation. However, in root cause analysis, in particular in an industrial context, this is rarely done as for an efficient MVA approach some challenges have to be overcome.

The most used methods in MVA are classification or calibration techniques. In our context the classification of 'good' and 'bad' samples was already observed by their physical behavior, which led to the initial failure. Calibration methods could be used to relate the external variables such as 'good' and 'bad' samples to the chemical surface composition. However, often they cannot be applied due to a very limited amount of samples and a limited amount of measurements per sample (due to e.g. sample size, sample topography, financial and time constraints.).

In addition, an appropriate data preprocessing is crucial for a successfully MVA. However, no general rules exist and the preprocessing always depends on the data set. In addition, the signals in the ToF-SIMS data vary over several orders of magnitude. The significant signals often lie in different intensity ranges. Therefore, an appropriate scaling is extremely difficult.

In this study we show an MVA approach, which takes these challenges into account and has provided solutions for many of our customers' problems. Although the classification is already known, we use classification techniques such as Principal Component Analysis (PCA), but focus on the loadings to investigate the signals (root) inducing this classification. We have developed a procedure for the efficient extraction of significant signals - in different intensity ranges - from these loadings. This strategy will be discussed along several examples.