Review in situ transmission electron microscope with machine learning

Advanced electronic materials are the fundamental building blocks of integrated circuits (ICs). The microscale properties of electronic materials (e.g., crystal structures, defects, and chemical properties) can have a considerable impact on the performance of ICs. Comprehensive characterization and analysis of the material in real time with high-spatial resolution are indispensable. In situ transmission electron microscope (TEM) with atomic resolution and external field can be applied as a physical simulation platform to study the evolution of electronic material in working conditions. The high-speed camera of the in situ TEM generates a high frame rate video, resulting in a large dataset that is beyond the data processing ability of researchers using the traditional method. To overcome this challenge, many works on automated TEM analysis by using machine-learning algorithm have been proposed. In this review, we introduce the technical evolution of TEM data acquisition, including analysis, and we summarize the application of machine learning to TEM data analysis in the aspects of morphology, defect, structure, and spectra. Some of the challenges of automated TEM analysis are given in the conclusion.