Host: The Japan Society of Vacuum and Surface Science
RHEED (reflection high-energy electron diffraction) is a widely used method for in-situ surface structural analysis of thin films. Since it is difficult to interpret the entire patterns quantitatively, researchers often use limited information such as the intensity oscillation at a given diffraction spot in film thickness estimation. Here, we adopted machine learning techniques for feature extraction from the entire RHEED patterns. We performed peak fitting analysis of the luminance histogram obtained from the time-series image datasets of RHEED patterns of Si surface superstructures during Indium deposition using EM algorithm. One peak component corresponds to the background, and the other corresponds to the diffraction spots. By tracking the change in the dispersion value of the peak, the optimal time for preparing each surface superstructure could be estimated automatically. Our method is expected for the application in data-driven material synthesis.
