Atom-by-atom analysis of sintering dynamics and stability of Pt nanoparticle catalysts in chemical reactions

Supported Pt nanoparticles are used extensively in chemical processes, including for fuel cells, fuels, pollution control and hydrogenation reactions. Atomic-level deactivation mechanisms play a critical role in the loss of performance. In this original research paper, we introduce real-time in-situ visualization and quantitative analysis of dynamic atom-by-atom sintering and stability of model Pt nanoparticles on a carbon support, under controlled chemical reaction conditions of temperature and continuously flowing gas. We use a novel environmental scanning transmission electron microscope with single-atom resolution, to understand the mechanisms. Our results track the areal density of dynamic single atoms on the support between nanoparticles and attached to them; both as migrating species in performance degradation and as potential new independent active species. We demonstrate that the decay of smaller nanoparticles is initiated by a local lack of single atoms; while a post decay increase in single-atom density suggests anchoring sites on the substrate before aggregation to larger particles. The analyses reveal a relationship between the density and mobility of single atoms, particle sizes and their nature in the immediate neighbourhood. The results are combined with practical catalysts important in technological processes. The findings illustrate the complex nature of sintering and deactivation. They are used to generate new fundamental insights into nanoparticle sintering dynamics at the single-atom level, important in the development of efficient supported nanoparticle systems for improved chemical processes and novel single-atom catalysis. This article is part of a discussion meeting issue ‘Dynamic in situ microscopy relating structure and function’.


Single Atom Detectability
In order to perform a statistical study on single atoms, the constraints determining their detectability in ESTEM must be mapped out. To begin with, the reaction conditions of temperature and gas environments will be expected to alter the number and detectability of single atoms.
Temperature is held constant at 20˚C, whilst single atom resolution in H2 has been previously shown to be at the sub-Angstrom level.
The next constraint on single atom detectability, as suggested in 1 , is the dependence on the migration speed of the atom versus the scan rate of the electron beam. For a scan speed much slower than the migration speed, the atom will only be present under the electron beam for a fraction of the dwell time. This scenario leads to the atom not being imaged at all, imaged with a very low SNR (signal to noise ratio) and/or being imaged multiple times. If the scan speed and migration speed are approximately equal, then a smearing of the atom intensity across several pixels may occur. There are two frequencies at which this can transpire, one for the horizontal scan speed and another for the vertical scan speed. The detectability of the atom then also becomes dependent on its direction, as well as its speed.
The scan rate must therefore be much faster than the single atom migration speed for a representative population of single atoms to be detected. As the migration speed may be affected by temperature and the support, the appropriate sampling rate must be determined on a case by case basis.
In addition to these detectability constraints, there is also the question of obtaining a statistically significant sample. For relating particles to single atoms, the requirement is that many single atoms 3 are detected per particle. The number of particles sampled must also be as large as possible. The challenge is therefore to use a low magnification, whilst maintaining single atom resolution by altering the pixel size and dwell time. The limiting factor in the case of Pt/C is the variation in the support height perpendicular to the electron beam. For magnifications <4MX, the electronic focus must be adjusted during image acquisition to account for height variations. By using a 4MX magnification, this issue is avoided and a sufficient sample size (approximately 200 single atoms and 100 particles) is obtained with this particle size distribution.
The next constraint is on sample drift. Generally, if the acquisition time is much larger than 2 minutes (clearly this value will fluctuate with microscope conditions), the position of features are perturbed relative to each other. The maximum acquisition time used is therefore 81.8s, but the proportion of this devoted to either number of pixels or dwell time can be chosen. To quantify this, pixel size and dwell time are independently varied at 4MX magnification. The signal from 30 single atoms is sampled along with 30 regions of the same size containing only the C-film. The region sampled is kept at 0.3x0.3nm (although the number of pixels varies) and is therefore comparable in diameter to a Pt atom 2 . To enable contrast measurements, the mean grey value of the sample region is calculated.
Here the contrast is defined as the signal difference between a single atom (SA) and the C-film (SB). The contrast to noise ratio can then be defined as: In an image a single atom occupies n pixels, each with a signal of SA with independent additive variation between the individual measurements of standard deviation σ0. The effective local SNR will be improved by the number of pixels making up a single atom. The CNR is modified by √n, thus defining the visibility vAB of an object. Rose 3 found that objects can be reliably distinguished when vAB is greater than 4 or 5.  The results of 30 repeats are then used to determine SA, SB and σ0.
Blurring of single atoms is not observed in the range of scan speeds used, as the intensity profile typically measured is Gaussian. Thus the scan rate is not the same as the migration speed and at least a fraction of the single atoms present are being detected with a scan rate much greater than the migration speed. Figure S1 shows the expected improvement in vAB as the acquisition time is increased from 5.1s to 81.8s. Note that there is no associated decrease in single atom number for dwell time or pixel size variations. This implies that the scan speeds used in the 5.1-81.8s range are not lower than the typical migration speed.
The maximum dwell time used is 312μs, corresponding to 512x512 pixels, with a pixel size of 0.07nm. If the pixel size exceeds the probe size (>0.1nm), then there will be gaps in the raster scan, potentially leading to undetected single atoms. As previously discussed, the acquisition time is limited to 81.8μs to reduce thermal drift. Maximising pixel count (2048x2048, 19.5μs) provides a vAB of 12, versus 10.4 when dwell time is increased to 312μs, 512x512 pixels. The difference between the two is detectable, but likely not significantly different when the temperature or support is altered. However, maximising pixel count does reduce the electron beam dose rate to the sample.
Most studies suggest that the significant factor in reducing beam effects is a dependency on dose rate, rather than total dose 4-7 . The following work therefore uses 2048x2048 pixels, at a dwell time 6 of 19.5μs. Note that rastering of the ebeam pauses at the beginning of each line, thus the acquisition times are slightly longer than those stated for comparison purposes above.
The final parameters of interest are the brightness and contrast levels used to interpret the detector signal. These are kept at constant values where contrast is maximised (for optimised differentiation between Pt and C) and brightness allows for visibility of internal particle structure, single atom resolution and some substrate visibility (brightness was approximately 1/3 of the maximum value).
Optimised conditions for single atom detectability at 20˚C have now been developed and an understanding of in situ OR at the atomic scale can be undertaken.

Electron Beam Effects
Careful calibration experiments [8][9][10][11] were employed in the experiments. Minimally invasive low dose electron beam techniques were used throughout. In situ ESTEM data under low dose imaging conditions were compared to regions unexposed to the electron beam by beam blanking 11 . The aim was to ensure non-invasive characterisation. From Fig. S2. there is no change in either the PSD shape or the mean particle size of the areas exposed or unexposed to the electron beam. This indicates that electron beam influence has been minimised by using a sampling rate of 1 acquisition per 5 minutes. Over the course of the 75 minute experiment, 16 high magnification (4MX) images were taken.
The positions of the single atoms and particles in each frame were calculated relative to 4 large 8 particles in the corners of each image. This minimises the influence of particle migration, although none was detected above the noise due to particle shape changes. The error in single atom position is thus the difference between alignment of the four particles in each image.

Voronoi Averaging Effects
The Voronoi analysis assumes that the local correlation model (using a Voronoi averaging technique) is applicable to this situation. To ensure that the single atom density changes are not an averaging effect, the observation area was changed to the individual Voronoi cells surrounding particles 17 and 18, Fig. S3. This shows a similar decrease in single atom density at t=25 mins, and increase at t=55 mins, showing that the trend is not an artefact introduced by surrounding particles and their nearest neighbours. The single atom density can also obviously be affected by focussing errors or other misalignments when acquiring images. If this was influencing the 'bathtub' trend from Fig. S4., then the single atom count across the image would follow the same trend. From Fig. S4., it can be seen that this is clearly not the case.

Acknowledgments
The authors acknowledge support from the UK Engineering and Physical Science Research