Professor Yang Yong-soo of the Department of Physics at KAIST developed atomic resolution electron tomography technology based on scanning transmission electron microscope (STEM) using artificial neural networks. Futhermore through the adapting the development they precisely identified the platinum nano-particles surface and inside of the three-dimensional atomic structure by 15 pm (Picometer). 1 pm (Picometer) is a unit of 1 over one trillion of 1 meter accurate 15 pm is high level of one over 3 of the hydrogen atomic radius.

The electron tomography is a technique used to acquire 3D images from projected 2D images measured from various angles with electron microscopes. Recently by the technology development of the STEM and 3D reconstruction algorithm the resolution of the electron tomography has reached the level where it can be categorized to a single atom. This allowed us to understand the structure and physical properties of many nano-materials.

However in a general electron tomography experiment it is impossible to measure the image at a high angle (About 75 degrees or more) due to the experimental constraint that the holder or grid equipped with the design piece blinds the electron beam. Consequently resolution in the high angle direction is reduced and unwanted noise is generated in the reconstructed 3D image. This phenomenon is called the missing wedge problem which makes it difficult to measure the 3D atomic structure of the surface and interface by conventional electron tomography with high resolution.

Professor Yangs research team successfully solved the missing wedge problem by restoring high-angle data using artificial neural networks. This made it possible to determine a 3D surface and interface atomic structure with high resolution and fundamentally analyze the mechanism of physical properties appearing on the surface and interface of nano-materials at a single atomic level.

The research that Lee Joo-hyeok combined MS/PhD student of Department of Physics participated as the first author of the research was published in the international journal Nature Communications.

The research team generated atomic structure tomography 3D data through simulation based on atomicity that all materials were composed of atoms. In order to teach the correlation between incomplete atomic structure tomography 3D data with high-angle data loss and ideal atomic structure 3D data an artificial intelligence neural network (3d-unet based model) was coached. Artificial intelligence neural networks taught by the basis of atomicity solved the problem of resolution degradation due to damaged wedge problems by restoring lost high-angle data. This allows the identification of the 3D surface interface atomic structure with high accuracy.

The research team was able to determine the 3-dimensional surface and internal structure of platinum nano-particles at a single atomic level using the developed electron tomography technology based on the artificial neural network. The accuracy of the atomic structure has improved significantly from 26pm before the artificial neural network application to 15pm after the application.


Professor Yang Yong-soo who led the research said Electron tomography based on the artificial neural network is a general method that does not depend on the structure and form of the material and can be applied immediately to atomic structure volume data acquired by the electron tomograph. and mentioned the significant result by explaining Through this various materials 3D surface and interface atomic structures are identified accurately and with the fundamental understanding of properties occurring on the surface and interface and related mechanism it will be applied to development of high performance catalyst.

Meanwhile this research was carried out with the support of the National Research Foundation of Koreas individual basic research project and the KAIST Global Specificity Project (M3I3).
