SKM 2021 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 3: AKPIK Postersession
AKPIK 3.3: Poster
Donnerstag, 30. September 2021, 13:30–15:30, P
Convolutional Neural Network Framework for the Analysis of X-ray photoelectron spectra — •Lukas Pielsticker1, Rachel L. Nicholls1, Gudrun Klihm1, Robert Schlögl1,2, and Mark Greiner1 — 1Department Heterogeneous Reactions, Max Planck Institute for Chemical Energy Conversion, Mülheim an der Ruhr — 2Department Inorganic Chemistry, Fritz Haber Institute of the Max Planck Society, Berlin
X-ray photoelectron spectroscopy (XPS) enables studying the electronic structure and chemical state of solid materials and their surfaces. Quantitative analysis of the phases present in XP spectra is typically performed by manual peak fitting. However, such elemental quantification often suffers from, among other things, superposition of core-levels of different elements, incorrect instrument calibration, poor choice of backgrounds and lineshapes, as well as from noise in the data. Moreover, as XPS instruments are becoming increasingly automated and capable of producing large amounts of data, an equally automated approach to elemental quantification is desirable.
Here, a scalable automation framework for XPS analysis using Convolutional Neural Networks (CNNs) is presented. For model training, synthetic mixed metal-oxide spectra were generated based on known reference spectra. CNNs are shown to be capable of quantitatively determining the presence of metallic and oxide phases, as well as identifying morphological features such as over- and sublayers, exhibiting more reliable performance than standard XPS users. The use of Bayesian CNNs for the determination of quantification uncertainty is illustrated.