Elucidating Metal and Oxide Nanoparticle Surface Dynamics through In Situ TEM and Artificial Intelligence

Nanoparticle systems often show high degrees of instability which are strongly influenced by size and ambient environment. These dynamic structural changes modify material properties such as reactivity, phase changes and catalysis. Atomic-level surface dynamics may play a significant role in defining particle structures and functionalities but characterizing nanoparticle surface structures with high spatial and temporal resolution simultaneously has proven challenging. Transmission electron microscopy (TEM) is a key tool to visualize local atomic structure of nanoparticles. Ultrafast TEM can now achieve picosecond temporal resolution but is limited to a spatial resolution of about 2 nm. Fortunately, modern detectors now provide readout rates in excess of 1000 frames per second, offering the potential to investigate atomic-level structural evolutions with time resolutions down to a millisecond. However, in most situations, the need to limit electron dose rates results in high temporal resolution movies that are dominated by shot noise, which often obscures surface structural dynamics. In order to address this challenge, we propose a fully-unsupervised AI denoising framework, which enables recovery of atomic-resolution information from such data. The unsupervised deep video denoising framework [1] improves the signal-to-noise ratio (SNR) by a factor of 30 at a spatial resolution of 1 Å and time resolution of 10 ms. For this investigation, we explore structural dynamics of metal and oxide nanoparticle particles under combinations of CO and O₂ at temperatures up to 300^oC. The enhanced time resolution reveals that supposedly stable, low-energy surfaces can display highly active dynamics, triggering nanoparticle instabilities resulting in rapid structural fluctuations. The new spatiotemporal capability enabled by the proposed framework dramatically enhances our ability to explore surface dynamics and the evolution of metastable states in nanoparticles at the atomic level, offering new insights into their functionality.<br/> <br/>References<br/>1. Sheth, D.Y., et al. <i>Unsupervised deep video denoising</i>. in <i>Proceedings of the IEEE/CVF International Conference on Computer Vision</i>. 2021.<br/>2. We gratefully acknowledge the support of the following NSF grants to ASU (OAC 1940263, 2104105, CBET 1604971, and DMR 184084 and 1920335) and NYU (HDR-1940097 and OAC-2103936). The authors acknowledge HPC resources available through ASU, and NYU as well as the John M. Cowley Center for High Resolution Electron Microscopy at Arizona State University.