분회초청강연


     나노매뉴팩쳐링




  최시영 교수 포항공과대학교


     ■ Education


2000.9 - 2004.4
Ph. D.
Department of Materials Science and Engineering Korea Advanced Institute of Science and Technology (KAIST)
1999.3 - 2000.8
M.S.
Department of Materials Science and Engineering KAIST
1993.3 - 1999.2
B.S.
Department of Ceramic Engineering Pusan National University, Busan, Korea




■ Professional Career


2017.7 - Present
Associate Professor
Department of Materials Science & Engineering, POSTECH
2016.1 - 2017.6
Head of Department
Department of Materials Modeling & Characterization Korea Institute of Materials Science (KIMS)
2013.1 - 2015.1
Principal Researcher, KIMS
2007.12 - 2012.12
Senior Researcher, KIMS
2006 - 2007.12
JSPS Fellow Researcher
Department of Materials Science & Engineering The University of Tokyo




Atomistic analysis via deep machine learning



The use of Scanning Transmission Electron Microscopy (STEM) has led to a much deeper understanding of the materials via the ability to obtain the infinitesimal atomic structure. Of great importance is determining the structure-property relationships, i.e. how the atomic structure gives rise to the measured properties of the material and how structural changes affect these properties. For the last decades, the STEM technique has been drastically improved along with the higher-order aberration corrector and the state-of-art detectors such as pixelate CCD detector, the segmented detector, and so on. In addition, in order to determine more precisely the structure-property relationships, it is necessary to extract the atomic coordinate information and the slight difference in the atomic structure by using deep machine learning. To this end, we introduce the new methods to extract atomistic information at length scales even lower than the current resolution limits of from the instrument itself, which is currently 50-100pm; furthermore, we also propose the analytical method based on deep machine learning to recognize the subtle distinction in the atomic structures and thus to classify the type of octahedral tilt or the point defect. Moreover, the whole procedure is highly automated leading to improved efficiency in addition to a reliable analysis.