2:30 PM - 2:45 PM
▲ [17p-A408-7] Machine learning analysis of RHEED images for structural phase mapping
Keywords:RHEED, phase mapping, machine learning
Optimizing thin film growth conditions (temperature, pressure, composition, growth rate, etc.) is a time consuming part of thin film materials design because a large number of growth experiments is required and structural analysis by x-ray diffraction is slow. We show that real-time structural phase mapping as a function of film growth conditions is possible based on Reflection High-Energy Electron Diffraction (RHEED) image analysis and for the chosen model system of FeOx near the Fe2O3/Fe3O4 phase boundary, the RHEED phase mapping produces an equivalent result with the much slower x-ray diffraction analysis. Since quantitative analysis of a large number of RHEED images is not feasible by hand, we have developed a machine learning workflow for automatically generating a phase composition map based on the RHEED. Deep learning (U-Net convolutional neural network) was used to find the diffraction features in the images, followed by feature location fitting and periodicity analysis. Assuming that each distinct period corresponds to a different crystalline phase, intensity analysis and clustering was used for automated phase map construction.