STEM Image Analysis Based on Deep Learning: Identification of Vacancy Defects and Polymorphs of MoS2

Kihyun Lee; Jinsub Park; Soyeon Choi; Yangjin Lee; Sol Lee; Joowon Jung; Jong Young Lee; Farman Ullah; Zeeshan Tahir; Yong Soo Kim; Gwan Hyoung Lee; Kwanpyo Kim

doi:10.1021/acs.nanolett.2c00550

STEM Image Analysis Based on Deep Learning: Identification of Vacancy Defects and Polymorphs of MoS₂

Kihyun Lee, Jinsub Park, Soyeon Choi, Yangjin Lee, Sol Lee, Joowon Jung, Jong Young Lee, Farman Ullah, Zeeshan Tahir, Yong Soo Kim, Gwan Hyoung Lee, Kwanpyo Kim

Research output: Contribution to journal › Article › peer-review

3 Citations (Scopus)

Abstract

Scanning transmission electron microscopy (STEM) is an indispensable tool for atomic-resolution structural analysis for a wide range of materials. The conventional analysis of STEM images is an extensive hands-on process, which limits efficient handling of high-throughput data. Here, we apply a fully convolutional network (FCN) for identification of important structural features of two-dimensional crystals. ResUNet, a type of FCN, is utilized in identifying sulfur vacancies and polymorph types of MoS2 from atomic resolution STEM images. Efficient models are achieved based on training with simulated images in the presence of different levels of noise, aberrations, and carbon contamination. The accuracy of the FCN models toward extensive experimental STEM images is comparable to that of careful hands-on analysis. Our work provides a guideline on best practices to train a deep learning model for STEM image analysis and demonstrates FCN's application for efficient processing of a large volume of STEM data.

Original language	English
Pages (from-to)	4677-4685
Number of pages	9
Journal	Nano letters
Volume	22
Issue number	12
DOIs	https://doi.org/10.1021/acs.nanolett.2c00550
Publication status	Published - 2022 Jun 22

Bibliographical note

Funding Information:
We thank Prof. Hwidong Yoo for providing computing resources (YHEP server) and helpful comments and discussions. We also thank Prof. Sejung Yang for her insightful comments and advice on data preprocessing and machine learning. This work was mainly supported by the Basic Science Research Program of the National Research Foundation of Korea (NRF-2017R1A5A1014862 and 2022R1A2C4002559) and by the Institute for Basic Science (IBS-R026-D1). Y.L. received support from the Basic Science Research Program at the National Research Foundation of Korea which was funded by the Ministry of Science and ICT (NRF-2021R1C1C2006785). Y.S.K. acknowledges support from the Priority Research Centers Program (2019R1A6A1A11053838) and the Basic Science Research Programs (2021R1A2C1004209) through the National Research Foundation of Korea (NRF). G.H.L acknowledges the support from Creative-Pioneering Researchers Program through Seoul National University (SNU).

Publisher Copyright:
© 2022 The Authors. Published by American Chemical Society.

All Science Journal Classification (ASJC) codes

Bioengineering
Chemistry(all)
Materials Science(all)
Condensed Matter Physics
Mechanical Engineering

Access to Document

10.1021/acs.nanolett.2c00550

Cite this

@article{964c4775beef46fca95efc5d5b53d3fe,

title = "STEM Image Analysis Based on Deep Learning: Identification of Vacancy Defects and Polymorphs of MoS2",

abstract = "Scanning transmission electron microscopy (STEM) is an indispensable tool for atomic-resolution structural analysis for a wide range of materials. The conventional analysis of STEM images is an extensive hands-on process, which limits efficient handling of high-throughput data. Here, we apply a fully convolutional network (FCN) for identification of important structural features of two-dimensional crystals. ResUNet, a type of FCN, is utilized in identifying sulfur vacancies and polymorph types of MoS2 from atomic resolution STEM images. Efficient models are achieved based on training with simulated images in the presence of different levels of noise, aberrations, and carbon contamination. The accuracy of the FCN models toward extensive experimental STEM images is comparable to that of careful hands-on analysis. Our work provides a guideline on best practices to train a deep learning model for STEM image analysis and demonstrates FCN's application for efficient processing of a large volume of STEM data.",

author = "Kihyun Lee and Jinsub Park and Soyeon Choi and Yangjin Lee and Sol Lee and Joowon Jung and Lee, {Jong Young} and Farman Ullah and Zeeshan Tahir and Kim, {Yong Soo} and Lee, {Gwan Hyoung} and Kwanpyo Kim",

note = "Funding Information: We thank Prof. Hwidong Yoo for providing computing resources (YHEP server) and helpful comments and discussions. We also thank Prof. Sejung Yang for her insightful comments and advice on data preprocessing and machine learning. This work was mainly supported by the Basic Science Research Program of the National Research Foundation of Korea (NRF-2017R1A5A1014862 and 2022R1A2C4002559) and by the Institute for Basic Science (IBS-R026-D1). Y.L. received support from the Basic Science Research Program at the National Research Foundation of Korea which was funded by the Ministry of Science and ICT (NRF-2021R1C1C2006785). Y.S.K. acknowledges support from the Priority Research Centers Program (2019R1A6A1A11053838) and the Basic Science Research Programs (2021R1A2C1004209) through the National Research Foundation of Korea (NRF). G.H.L acknowledges the support from Creative-Pioneering Researchers Program through Seoul National University (SNU). Publisher Copyright: {\textcopyright} 2022 The Authors. Published by American Chemical Society.",

year = "2022",

month = jun,

day = "22",

doi = "10.1021/acs.nanolett.2c00550",

language = "English",

volume = "22",

pages = "4677--4685",

journal = "Nano Letters",

issn = "1530-6984",

publisher = "American Chemical Society",

number = "12",

}

TY - JOUR

T1 - STEM Image Analysis Based on Deep Learning

T2 - Identification of Vacancy Defects and Polymorphs of MoS2

AU - Lee, Kihyun

AU - Park, Jinsub

AU - Choi, Soyeon

AU - Lee, Yangjin

AU - Lee, Sol

AU - Jung, Joowon

AU - Lee, Jong Young

AU - Ullah, Farman

AU - Tahir, Zeeshan

AU - Kim, Yong Soo

AU - Lee, Gwan Hyoung

AU - Kim, Kwanpyo

N1 - Funding Information: We thank Prof. Hwidong Yoo for providing computing resources (YHEP server) and helpful comments and discussions. We also thank Prof. Sejung Yang for her insightful comments and advice on data preprocessing and machine learning. This work was mainly supported by the Basic Science Research Program of the National Research Foundation of Korea (NRF-2017R1A5A1014862 and 2022R1A2C4002559) and by the Institute for Basic Science (IBS-R026-D1). Y.L. received support from the Basic Science Research Program at the National Research Foundation of Korea which was funded by the Ministry of Science and ICT (NRF-2021R1C1C2006785). Y.S.K. acknowledges support from the Priority Research Centers Program (2019R1A6A1A11053838) and the Basic Science Research Programs (2021R1A2C1004209) through the National Research Foundation of Korea (NRF). G.H.L acknowledges the support from Creative-Pioneering Researchers Program through Seoul National University (SNU). Publisher Copyright: © 2022 The Authors. Published by American Chemical Society.

PY - 2022/6/22

Y1 - 2022/6/22

N2 - Scanning transmission electron microscopy (STEM) is an indispensable tool for atomic-resolution structural analysis for a wide range of materials. The conventional analysis of STEM images is an extensive hands-on process, which limits efficient handling of high-throughput data. Here, we apply a fully convolutional network (FCN) for identification of important structural features of two-dimensional crystals. ResUNet, a type of FCN, is utilized in identifying sulfur vacancies and polymorph types of MoS2 from atomic resolution STEM images. Efficient models are achieved based on training with simulated images in the presence of different levels of noise, aberrations, and carbon contamination. The accuracy of the FCN models toward extensive experimental STEM images is comparable to that of careful hands-on analysis. Our work provides a guideline on best practices to train a deep learning model for STEM image analysis and demonstrates FCN's application for efficient processing of a large volume of STEM data.

AB - Scanning transmission electron microscopy (STEM) is an indispensable tool for atomic-resolution structural analysis for a wide range of materials. The conventional analysis of STEM images is an extensive hands-on process, which limits efficient handling of high-throughput data. Here, we apply a fully convolutional network (FCN) for identification of important structural features of two-dimensional crystals. ResUNet, a type of FCN, is utilized in identifying sulfur vacancies and polymorph types of MoS2 from atomic resolution STEM images. Efficient models are achieved based on training with simulated images in the presence of different levels of noise, aberrations, and carbon contamination. The accuracy of the FCN models toward extensive experimental STEM images is comparable to that of careful hands-on analysis. Our work provides a guideline on best practices to train a deep learning model for STEM image analysis and demonstrates FCN's application for efficient processing of a large volume of STEM data.

UR - http://www.scopus.com/inward/record.url?scp=85132453474&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85132453474&partnerID=8YFLogxK

U2 - 10.1021/acs.nanolett.2c00550

DO - 10.1021/acs.nanolett.2c00550

M3 - Article

C2 - 35674452

AN - SCOPUS:85132453474

SN - 1530-6984

VL - 22

SP - 4677

EP - 4685

JO - Nano Letters

JF - Nano Letters

IS - 12

ER -