AVS 50th International Symposium
    QSA-10 Topical Conference Monday Sessions
       Session QS-MoA

Paper QS-MoA10
Multivariate Statistical Analysis of Spatially Compressed Time-of-Flight Secondary Ion Mass Spectrometry Images@footnote 1@

Monday, November 3, 2003, 5:00 pm, Room 320

Session: Thin-Film Metrology
Presenter: J.A. Ohlhausen, Sandia National Laboratories
Authors: J.A. Ohlhausen, Sandia National Laboratories
M.R. Keenan, Sandia National Laboratories
P.G. Kotula, Sandia National Laboratories
D.E. Peebles, Sandia National Laboratories
Correspondent: Click to Email
Owing to the parallel nature of Time-of-Flight Secondary Ion Mass Spectrometry (TOF-SIMS), complex and very large datasets can easily be acquired. An example of such a large dataset is a spectral image where a complete spectrum is collected for each pixel. Ideally, the complete spectral image would be used to provide a comprehensive materials characterization. This is difficult to accomplish with traditional techniques. Previously, we have demonstrated the application of multivariate spectrum imaging techniques to TOF-SIMS. This technique, called Automated eXpert Spectral Image (and series) Analysis -- AXSIA, is based on the separation of a complex and very large spectral image dataset into physically realizable and intuitive chemical components, including both spectra and concentrations. The full analysis is performed without outside estimates of spectral shapes, concentrations or the number of components present. In TOF-SIMS, we have shown that spectral series in the form of depth profiles (1D), images (2D), and imaged depth profiles (3D) can be analyzed using AXSIA. Since datasets can be large (5MB-1GB), data compression must be performed in order to process the data on laboratory computers. While providing signal-to-noise and memory storage improvements, data compression can hide or dilute important and small features. In this talk, I will present some statistical advantages of using multivariate techniques directly to spatially compressed data while maintaining full image resolution. In addition, I will explore the trade-offs between spatial and spectral compression and small feature recognition. @FootnoteText@ @footnote 1@This work was completed at Sandia National Laboratories, a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin Company, for the United States Department of Energy's National Nuclear Security Administration under contract DE-AC04-94AL85000.