Combining Deep Learning and Compressed Sensing Methods for the 3D Characterization of Ultra‐Thin Epitaxial Layers Grown on Controlled‐Shape Nano‐Oxides

Abstract

Using a nanostructured platform (a controlled‐shape nano‐oxide) and conventional wet impregnation techniques, powder‐type materials have been prepared in which atomically thin surface layers are deposited under very mild conditions. More importantly, an advanced methodology, combining energy dispersive X‐ray spectroscopy‐scanning transmission electron tomography (STEM‐EDX ET) and deep learning denoising techniques, has been developed for the 3D compositional characterization of these unique nanosystems. The complex case of LaOx‐coated CeO2 nanocubes is illustrated. For these, aberration corrected 2D STEM‐EDX evidence that ceria nanocubes become covered with a 2–4 atom‐thick layer of a La, Ce‐mixed oxide with spatially varying composition. However, STEM‐EDX ET reveals that this layer distributes unevenly, patching most of the available nanocube surface. The large flexibility and spread availability of the involved synthetic techniques enables, using the tools here developed, a wide exploration of the wealth of questions and applications of these intriguing, atomically thin, surface oxide phases.

Quantitative 3D Characterization of Functionally Relevant Parameters in Heavy-Oxide-Supported 4d Metal Nanocatalysts

“…For example, DL techniques have been successfully employed to reconstruct 3D structures of systems like ZnSe@ZnS quantum dots involving components with certain compositional proximity . This technique has also been utilized to dump the effects of the missing wedge and enhance resolution in ET, as demonstrated by Wang et al Furthermore, these methods have also been applied to denoise energy-dispersive X-ray (EDX) elemental maps and improve analytical electron tomography studies. , Recently, DL algorithms have also been successfully used for automatic segmentation in a large data set of ET experiments involving Pt nanoparticles supported on alumina …”

“…23 This technique has also been utilized to dump the effects of the missing wedge and enhance resolution in ET, as demonstrated by Wang et al 24 Furthermore, these methods have also been applied to denoise energy-dispersive X-ray (EDX) elemental maps and improve analytical electron tomography studies. 25,26 Recently, DL algorithms have also been successfully used for automatic segmentation in a large data set of ET experiments involving Pt nanoparticles supported on alumina. 27 Based on these recent findings, we hypothesize that DL methods can improve the analysis of the limited contrast differences observed in HAADF-STEM images of nanocatalysts with 3d-4d metals supported on ceria oxide particles.…”