CAMTEC Guest talk

TITLE: Computational microscopy for extracting nanoscale 3D structural insights from 2D TEM micrographs

SPEAKERS: Dr. Deepan Balakrishnan (Senior Postdoctoral Fellow, Center for Bioimaging Science, National University of Singapore)

DATE: Thursday, July 11, 2024

TIME: 2:30 PM – 3:15 PM

LOCATION: ENGINEERING/COMPUTER SCIE 104

SEMINAR ABSTRACT:

Like how we extended our sensing beyond the visible spectrum with infrared imaging, X-ray, and electron microscopy, extending the inference with a quantitative brain can stretch the boundaries of understanding structure-function relationships, which is crucial for nanofabrication and materials research. Even though nanotechnology moves towards fabricating more complex 3D nanostructures such as metasurfaces or solid-state batteries, due to the lack of rapid, easily accessible nanoscale 3D imaging tools, most structural insights are currently limited to 2D. Our recent work, pop-out 3D metrology, is a computational optics technique that exploits the physics of electron-matter interaction and known material priors to determine the depth and thickness of the specimen’s local region to create a 3D reconstruction. Unlike tomography, pop-out metrology does not require a tilt series and can provide a 3D image of amorphous specimens from a single 2D specimen. The underlying principle shows we can extend the pop-out 3D metrology to imaging more complex multi-layered specimens. We demonstrated that with suitable automation, the pop-out principle can readily be applied to measure nanoscale 3D structural dynamics from a large field of view. For complex multi-layered heterogeneous materials, deriving 3D insights from a 2D projection is ill-posed. However, the pop-out metrology provides the critical spatial coarse-gaining that acts as a regularizer to render the ill-posed 2D to 3D reconstruction a well-posed problem. Now, we are exploring the possibilities of employing machine learning models to learn electron optics, electron-matter interaction, and material priors for inferring 3D insights directly from 2D images.