Multivariate Analysis for Spectromicroscopy : Getting More From XPS Imaging

The spherical mirror analyser (SMA) has been employed on a commercially available photoelectron spectrometer for ten years.  During this time numerous examples of both elemental and chemical state images have been published and x-ray photoelectron imaging has become a routine technique for the determination of lateral distribution of elements and chemical species at the surface. 

Here we review the properties of the SMA including fast parallel image acquisition, high spatial and energy resolution and provide examples of the capabilities of such an imaging analyser.  In the last three years the combination of the SMA with a two-dimensional, pulse counting electron detector has again increased the level of information available for surface characterisation.  The delay-line detector (DLD) represents the next generation of photoelectron detection for XPS imaging and has allowed the realisation of quantitative surface chemical state microscopy by x-ray photoelectron spectroscopy.  To generate such information requires the acquisition of multi-spectral datasets comprising a series of images incremented in energy so that each pixel contains photoelectron intensity as a function of energy.  The datasets generated by this method contain >65,500 spectra and are therefore ideally suited to multivariate analysis to analyse the information content of the dataset and as a tool for noise reduction in individual images or spectra.  Several examples including a plasma modified surface and corrosion of a medical grade steel are used to demonstrate parallel imaging using the SMA and the additional information that can be gained by using multivariate statistical analysis of multi-spectral datasets.

Principle Component Analysis
Principle component analysis (PCA) assumes that any dataset can be described by a linear combination of one or more pure components.  As described in the paper by Walton [1] multiplying the data matrix by its transpose, a covariance matrix is formed which can then be decomposed into an orthogonal dataset, using singular value decomposition (SVD) sort.  From this the maximum variation in the data is partitioned into abstract components with the largest eigenvalues.  The abstract factors without any obvious features can be attributed to noise.  If the original dataset is reconstructed from only those abstract factors containing significant information, the result is a new dataset where the influence of the noise is reduced in magnitude.  One limitation of the PCA approach to noise reduction is the significant computation time required although substantial decreases can be achieved by operating on a subset of each images at a time.  In the example shown to the right the dataset of 256 images is processed in groups of 16 images such that the SVD is applied to adjacent images and then stepped through the dataset instead of being applied to the entire dataset at once.  The result is to move the vector containing the most information to the top

Summary
The SMA allows high energy and spatial resolution images to be acquired very rapidly.  With the use of a pulse counting delay-line detector genuine quantitative images are acquired to characterise surface distribution of elemental or chemical state species.  Acquiring a set of images incremented in energy over photoemission peaks a large 3-dimensional dataset can be generated easily.  The use of multivariate statistical analysis to extract the information content of the multi-spectral dataset and as a tool for noise reduction in images or spectra has been demonstrated.

An A4 pdf version of the poster can be downloaded directly from the 'downloads>General Reference' section of the Members Area or by contacting the Applications Specialists at Kratos Analytical Ltd, Manchester.