Abstract
The adoption of grazing-incidence wide-angle X-ray scattering (GIWAXS) has expanded rapidly, as it has proven invaluable for extracting a wealth of structural information from thin films and surfaces on the atomic and molecular scale. The amount of scattering data collected in such experiments creates new challenges associated with relatively slow conventional data analysis due to the need for human input. Existing Machine-Learning-based Bragg-peak detection approaches have demonstrated the potential of automation, but significant challenges such as weak peaks, peaks on top of Debye-Scherrer rings, and difficulties in differentiating closely spaced peaks remain. We introduce a powerful model for GIWAXS peak detection based on a visual transformer that achieves a performance strongly surpassing existing models. We evaluate our new approach using a diverse set of labelled, experimental GIWAXS patterns from perovskite and organic thin films. Notably, we demonstrate for the first time the reliable detection of Bragg peaks superimposed on other peaks and Debye-Scherrer rings, a significant improvement in detecting low-intensity peaks and a substantially stronger overall performance. In addition, we evaluate both pre-and postprocessing latency, confirming that the overall analysis speed is sufficient for real-time applications. With its ability to deliver a fast and high-performance analysis, our approach enables new types of GIWAXS measurements, including closed-loop experiments, that were previously limited by analysis bottlenecks. The studied model could potentially also be applied to other datasets with a large dynamic range and weak signal-to-noise ratio.
Funding Information
  • Bundesministerium für Forschung und Technologie (05D23VT1)
  • European Commission (101129751)
  • Deutsche Forschungsgemeinschaft (390727645)
  • Deutsche Forschungsgemeinschaft (460248799)
  • Deutsche Forschungsgemeinschaft (EXC 2064/1)

This publication has 40 references indexed in Scilit: