Determination of Surface Dynamics on CeCh Nanoparticles Using Time-Resolved High-Resolution TEM

Abstract

Characterizing the structural evolution of catalyst nanoparticle surfaces can provide insights into the active motifs responsible for catalysis. High-resolution transmission electron microscopy (HRTEM) has been used extensively to study surface dynamics on nanoparticles using in situ imaging techniques [1]. Recent advances in imaging detector technology, namely direct detection cameras with fast acquisition modes, have enabled many new materials phenomena to be investigated, such as the imaging of each sequential step of the synthesis of Ni silicide nanostructures within Si nanowires [2]. With the advent of fast image acquisition, large datasets, otherwise known as "big data", are becoming increasingly common and new developments in data storage, data mining, and processing methods are necessary [3]. Thus, batch processing techniques are essential for reducing processing time and extracting useful information from large, often noisy image datasets. By combining advanced acquisition and processing techniques, timeresolved HRTEM will enable new insights into the structural evolution of nanoparticle surfaces. In this work, we use time-resolved HRTEM to achieve high spatial and temporal resolution of surface atom migration on CeO2 nanoparticles, a material that is used extensively in catalysis applications due to its oxygen exchange properties [4].An aberration-corrected FEI Titan ETEM equipped with a Gatan K2 IS direct detection camera (with high detection quantum efficiency) was used to image CeO2 nanoparticles at 400 frames/second and 10 4 e -/(Å 2 s) in vacuum. A single frame of 1/400 second exposure is shown in Figure 1a), and due to the fast acquisition rate, the signal-to-noise ratio of the image was low. To reduce noise in each individual frame, a Kalman filter was applied to an image stack of ~9000 frames and Figure 1b) shows the resulting filtered image from Figure 1a). For each pixel in the image, the Kalman filter uses a prediction step to produce an estimate of the pixel intensity and its uncertainty. The prediction is then averaged with the measured value from the image through a weighted average, with more weight given to estimates with higher certainty, producing a "filtered" value. The "filtered" value is then used to update the prediction for the next image frame. Thus, the Kalman filter is a computationally light, recursive filtering technique which adapts at each time step to significantly reduce noise in each frame. MIPAR, a commercially available software with a recipe-based image processing method, was used to identify atomic column positions and quantify column intensities [5]. Using MIPAR's batch processing feature, each image from the ~9000 image stack was analyzed with the same recipe. Figure 2a) shows the resulting mask of atomic columns (shown in red) that was detected from a single frame during processing. The mask was then used to generate measurements, including the position, area, intensity mean, and integrated intensity of each atomic column, from the unfiltered image frames. The in...