Presentation Information
[20p-A311-5]Development of Classification Model of RHEED Patterns Using Transfer Learning
〇JINKWAN KWOEN1, MASAHIRO KAKUDA1,2, YASUHIKO ARAKAWA1 (1.NanoQuine, Univ. Tokyo, 2.AIO Core)
Keywords:
RHEED,MBE
The Reflection High Energy Electron Diffraction (RHEED) technique is used for in-situ observation of the surface state of materials during crystal growth by Molecular Beam Epitaxy (MBE). We have proposed a model that classifies such RHEED patterns. However, the model was only applicable to RHEED images from a single MBE device, making generalized model construction challenging. Additionally, the generalization process is not straightforward due to characteristics of each MBE device (changes in the angle and position due to substrate mounting, variation in substrate heater current, electron beam changes due to charge-ups, etc.) and differences between devices (electron gun, substrate, screen position, internal component arrangement, electron gun performance). Therefore, creating a generalized classification model for RHEED patterns is not easy. In this study, we report on the development and performance validation of a machine learning model trained on RHEED images obtained from three different MBE devices.