A Web-Based Facility for Evaluating Peak-Fitting Methods in X-ray Photoelectron Spectroscopy

X-ray photoelectron spectroscopy (XPS) is the most commonly used technique for determining the chemical composition of surfaces. It is customary to fit analytical functions to overlapping peaks in a measured spectrum to determine peak positions, which identify the chemical state, and peak intensities, which quantify the surface composition. Thus, the reliability of various peak-fitting methods for XPS quantification, such as fitting different non-linear functions, is of interest to the XPS community. In addition, analysts should know how reliably they are employing a particular peak analysis method from the uncertainties associated with operator choices such as the type of background function used in their fits.

Researchers in the Surface and Microanalysis Science Division recently developed a web-based facility to help analysts determine the veracity of peak-fitting approaches in XPS. The NIST researchers had earlier prepared simulated spectra, based on a factorial design that consisted mainly of overlapping doublets for carbon 1s photoelectrons from polymers. An international group of 20 surface analysts fitted variations of Gaussian or Gaussian-Lorentzian functions to the doublets. Their results were analyzed at NIST to yield measures of bias and random error in the peak positions and intensities for each doublet spectrum in the factorial design. These error measures were further analyzed with respect to seven peak-fitting methods used by the analysts. A statistical analysis of the errors showed that some peak-fitting methods are more accurate and precise than others depending on the extent of overlap between peaks in the spectrum and the relative heights of the peaks.�

Users of the XPS-STD Website can analyze the simulated spectra and then assess their results for their chosen peak-fitting approach as it matches one of the seven peak-fitting methods used previously. Analysts enter the peak parameters from their own fits at the web site, and then receive an on-line analysis of their results in terms of bias and random error. They can then compare their errors with statistics for the subgroup of the 20 analysts who employed the same peak-fitting method.