Abstract: In this study, we explore the possibility to use deep learning for the reconstruction of phase images from 4D scanning transmission electron microscopy (4D-STEM) data. The process can be divided into two main steps. First, the complex electron wave function is recovered for a convergent beam electron diffraction pattern (CBED) using a convolutional neural network (CNN). Subsequently, a corresponding patch of the phase object is recovered using the phase object approximation. Repeating this for each scan position in a 4D-STEM dataset and combining the patches by complex summation yields the full-phase object. Each patch is recovered from a kernel of 3x3 adjacent CBEDs only, which eliminates common, large memory requirements and enables live processing during an experiment. The machine learning pipeline, data generation, and the reconstruction algorithm are presented. We demonstrate that the CNN can retrieve phase information beyond the aperture angle, enabling super-resolution imaging. The image contrast formation is evaluated showing a dependence on the thickness and atomic column type. Columns containing light and heavy elements can be imaged simultaneously and are distinguishable. The combination of super-resolution, good noise robustness, and intuitive image contrast characteristics makes the approach unique among live imaging methods in 4D-STEM.

Abstract: CsPbI3 is a promising perovskite material for photovoltaic applications in its photoactive perovskite or black phase. However, the material degrades to a photovoltaically inactive or yellow phase at room temperature. Various mitigation strategies are currently being developed to increase the lifetime of the black phase, many of which rely on inducing strains in the material that hinder the black-to-yellow phase transition. Physical insight into how these strategies exactly induce strain as well as knowledge of the spatial extent over which these strains impact the material is crucial to optimize these approaches but is still lacking. Herein, we combine machine learning potential-based molecular dynamics simulations with our in silico strain engineering approach to accurately quantify strained large-scale atomic structures on a nanosecond time scale. To this end, we first model the strain fields introduced by atomic substitutions as they form the most elementary strain sources. We demonstrate that the magnitude of the induced strain fields decays exponentially with the distance from the strain source, following a decay rate that is largely independent of the specific substitution. Second, we show that the total strain field induced by multiple strain sources can be predicted to an excellent approximation by summing the strain fields of each individual source. Finally, through a case study, we illustrate how this additive character allows us to explain how complex strain fields, induced by spatially extended strain sources, can be predicted by adequately combining the strain fields caused by local strain sources. Hence, the strain additivity proposed here can be adopted to further our insight into the complex strain behavior in perovskites and to design strain from the atomic level onward to enhance their sought-after phase stability.

Abstract: In this study, the multi-layer self-assembled molecular network formation of an alkylated tetrathiafulvalene compound is studied at the liquid-solid interface between 1-phenyloctane and graphite. A combined theoretical/experimental approach associating force-field and quantum-chemical calculations with scanning tunnelling microscopy is used to determine the two-dimensional self-assembly beyond the monolayer, but also to further the understanding of the molecular adsorption conformation and its impact on the molecular packing within the assemblies at the monolayer and bilayer level.

Abstract: In this paper, a methodology is presented to count the number of atoms in heterogeneous nanoparticles based on the combination of multiple annular dark field scanning transmission electron microscopy (ADF STEM) images. The different non-overlapping annular detector collection regions are selected based on the principles of optimal statistical experiment design for the atom-counting problem. To count the number of atoms, the total intensities of scattered electrons for each atomic column, the so-called scattering cross-sections, are simultaneously compared with simulated library values for the different detector regions by minimising the squared differences. The performance of the method is evaluated for simulated Ni@Pt and Au@Ag core-shell nanoparticles. Our approach turns out to be a dose efficient alternative for the investigation of beam-sensitive heterogeneous materials as compared to the combination of ADF STEM and energy dispersive X-ray spectroscopy.