Resolving overlapping peaks in ARXPS data: The effect of noise and fitting method

Abstract

Peak-fitting of X-ray photoelectron spectroscopy (XPS) data can be very sensitive to noise when the difference on the binding energy among the peaks is smaller than the width of the peaks. This sensitivity depends on the fitting algorithm. Angle-resolved XPS (ARXPS) analysis offers the opportunity of employing the combined information contained in the data at the various angles to reduce the sensitivity to noise. The assumption of shared peak parameters (center and width) among the spectra for the different angles, and how it is introduced into the analysis, plays a basic role. Sequential fitting is the usual practice in ARXPS data peak-fitting. It consist on first estimating the center and width of the peaks from the data acquired at one of the angles, and then using those parameters as a starting approximation for fitting the data for each of the rest of the angles. An improvement of this method consists of averaging the centers and widths of the peaks obtained at the different angles, and then employing these values to assess the areas of the peaks for each angle. Another strategy for using the combined information is by assessing the peak parameters from the sum of the experimental data. The complete use of the combined information contained in the data-set is optimized by the simultaneous fitting method. It consists of the assessment of the center and width of the peaks by fitting the data at all the angles simultaneously. Computer-generated data was employed to compare the sensitivity with respect to noise between the sequential, averaged-sequential, sum, and simultaneous fitting methods. It is shown that the latter is significantly more robust and could provide reliable results even for noisy data and small peak separation. The fundamentals for the robustness of the simultaneous method are discussed, as well as the possibility of fitting many parameters at the same time. As an added feature of ARXPS, it was found that the estimation of the error intervals on the peak parameters is done remarkably more precisely by employing XPS data at various angles.