Veranstaltungen

01. Juli 2025, 16:00 bis 17:00

Florian Simperl, TU Wien, IAP, FB Atom- und Plasmaphysik

Seminar

From Spectra to Structure (and back): A neural network for high-throughout materials characterization with X-ray photoelectron spectroscopy

X-ray photoelectron spectroscopy (XPS) is a characterisation technique that is sensitive to surface properties (down to 10 nm). It is used to investigate material properties, including chemical composition, chemical depth distribution or core-shell nanoparticles [1, 2]. In recent years, XPS has become a reliable and advanced experimental technique in various scientific and engineering disciplines. Traditionally, extracting accurate quantitative information from XPS data required trained experts to perform time consuming empirical peak fitting for each individual spectrum. In response to the growing need for reliable, instantaneous spectral analysis, we present an automated, quantitative XPS analysis pipeline that combines the Simulation of Electron Spectra for Surface Analysis (SESSA) software with a transformer-based neural network architecture. For this study, SESSA was used to generate approximately 8.1 million spectra of 7,587 inorganic and organic bulk compounds and single elements spanning the periodic table (Z = 1–92). These simulated spectra, together with their corresponding chemical abundance labels, were used to train a supervised transformer to classify the stoichiometry of non-crystalline bulk materials. The current model can already correctly predict 90% of all compounds in the test set with a mean absolute error of 0.2 in the chemical concentration for 80% of the predicted elements.

[1] D. Nanda Gopala Krishna, & John Philip (2022). Applied Surface Science Advances, 12, 100332.
[2] Greczynski, L. (2023). Nature Reviews Methods Primers, 3(1), 40.

Kalendereintrag

Veranstaltungsort

SEM.R. DB gelb 05 B
1040 Wien
Wiedner Hauptstraße 8-10/E134

 

Veranstalter

IAP
Manuela Marik
marik@iap.tuwien.ac.at

 

Öffentlich

Ja

 

Kostenpflichtig

Nein

 

Anmeldung erforderlich

Nein