A deep learning approach for determining the chiral indices of carbon nanotubes from high-resolution transmission electron microscopy images

Stimulated by the progress made in the fabrication of electronic devices based on carbon nanotubes, our objective is to facilitate the identification of the structure of nanotubes, characterized by the so-calledHamada indices (n, m), in experimental samples. High-resolution transmission electron microscopy (HRTEM) is a powerful tool to achieve this goal. However, image analysis is delicate and can be tedious for large samples. We have developed an automated image recognition system based on a convolutional neural network that allows an easy and fast identification of the helicity of nanotubes.
Neural networks have been proven in image classification in many fields. However, it is often difficult to obtain high quality training data in sufficient numbers. In this paper, we generate a large set of reference images using well-established numerical methods: molecular dynamics to generate realistic atomic structures and the multi-slice technique to simulate the corresponding HRTEM images. After training the deep learning system on this database, the application to experimental HRTEM images is reliable and fast. A demo of our software for the analysis of the structure of carbon nanotubes from high-resolution TEM images is available at http://hrtem-analysis.fr

Référence
Georg Daniel Förster, Alice Castan, Annick Loiseau, Jaysen Nelayah, Damien Alloyeau, Frédéric Fossard, Christophe Bichara, Hakim Amara
A deep learning approach for determining the chiral indices of carbon nanotubes from high-resolution transmission electron microscopy images. Carbon, 169, 465–474, 2020.
https://doi.org/10.1016/j.carbon.2020.06.086