Classification of Reflection High-Energy Electron Diffraction Pattern Using Machine Learning
Jinkwan Kwoen, Yasuhiko Arakawa
Abstract
Reflection high-energy electron diffraction (RHEED) has wide application because it allows in situ observation of the sample surface behavior during molecular beam epitaxy growth. In particular, the RHEED pattern has been used as a milestone for growth condition calibration because it dynamically changes depending on the sample temperature, material supply rate, and supply ratio. However, RHEED pattern analysis depends on the accumulated know-how of the operator and has a time limitation; thus, its application to real-time feedback control is difficult. Moreover, with the conventional computerization method, it is difficult to correctly reflect and recognize the changes in RHEED due to changes in the observation conditions. On the other hand, the machine learning method using the convolutional neural network (CNN) recognizes feature points in the input database and is suitable for the classification of images with variability. In this study, we propose a measurement method for identifying the RHEED pattern of GaAs substrates during continuous rotation and build a data set of the growth conditions. A classification model is established by training the deep learning model using CNN, and is found to be more than 99% accurate. This is expected to be useful in the field of highquality III–V growth on GaAs.