Automatic 3D Reconstruction by Deep Learning Neural Networks Using Images Acquired via 4D-STEM Stereo Imaging

The stereo-pair approach for revealing the 3D distribution of dislocation structures from the transmission electron micrographs proved to be an effective alternative approach compared to laborious electron tomography Further improvements occurred in the field of stereo 3D reconstruction with the development of a tilt-less imaging technique where a stereo-pair of images can be acquired in a single sample tilt using a convergent probe in scanning TEM mode (STEM) This technique brings many advantages such as the elimination of sample movement between two images and opens the avenue to dynamic imaging However, existing algorithms for post-processing and following 3D reconstruction were still slow and required human intervention.