Alvaro J. Lizarbe¹, Kristopher Wright¹, Garrett Lewis¹, Aaron J. Tippitts¹, Stanislav Prusa², Jeff Terry³, David Aspnes⁴, Matthew R. Linford^1
1Brigham Young University, Provo, UT
²Brno University of Technology, Brno, Czech Republic
³Illinois Institute of Technology, Chicago, IL
⁴North Carolina State University, Raleigh, NC
XPS is the most widely used method for chemically characterizing surfaces. Because current advances in XPS are leading to the collection of larger data sets, mathematical and algorithmic methods for analyzing XPS data are increasingly important. There will not be time to analyze every bit of every spectrum in the large data sets being collected, and, like many areas in science, there are ‘reproducibility’ concerns about the quality of much of the data analysis being performed. In this talk, I will discuss various aspects of collecting and analyzing XPS data. These include:

1. Peak fitting. In general, the peak fitting of an XPS spectrum is the modeling of the chemical content of a surface. The choice of background and peak shapes is important here. Common choices for backgrounds include the Shirley, Tougaard, and linear backgrounds. Their use depends on the spectrum and material analyzed. Most synthetic peaks used in peak fitting are mixtures of Gaussians (G) and Lorentzians (L), i.e., GL sum, GL product, and GL convolution (Voigt) functions. Asymmetry is necessary in some synthetic fit components, especially from metallic materials that have high densities of states at their Fermi energies.
2. Data science tools. The comparison of large numbers of spectra, e.g., from images, depth profiles, operando studies, and damage studies is facilitated by the use of data science tools that include principal component analysis (PCA), multivariate curve resolution (MCR), and cluster analysis (CA). These tools provide graphical results that reveal the chemical similarities and differences between spectra and especially between groups (clusters) of related spectra.
3. Fourier denoising. There are times when it is not possible or not practical to collect high signal-to-noise data. Under these conditions, it may be advantageous to Fourier denoise XPS data. This is best done in reciprocal space with a Guass-Hermite filter function. This approach has now been applied to a variety of different peaks.
4. Unique depiction of XPS data. Time permitting, I will discuss a new method of depicting XPS data.