A. I Kirkland1,2,3
1 Department of Materials, Parks Road, Oxford, UK.
2 Electron Physical Sciences Imaging Centre (ePSIC), Diamond Light Source, Oxford, UK.
3 The Rosalind Franklin Institute, Harwell Campus, Didcot, OX11 0FA, UK.
This lecture will describe firstly recent developments in the use of Scanning Transmission and Transmission Electron Microscopy applied quantitative structural studies of materials where data is acquired under low dose conditions and / or at high frame rates to study dynamics
I will describe recent work using high speed direct electron detectors and artificial intelligence / machine learning to automatically map defect and adatom migrations in graphene from large data sets. I will then show how this approach can be extended to probe the local kinetics of defect transitions.
In the first part of this lecture, I will describe recent work using high speed direct electron detectors and artificial intelligence / machine learning to automatically map defect and adatom migrations in graphene from large data sets. I will then show how this approach can be extended to probe the local kinetics of defect transitions. I will also describe the use of a related approach to understanding industrial catalysts
For both applications the extremely large datasets (typically 106 images or greater) that can be routinely acquired makes conventional manual image processing, with human intervention intractable. Theses also put a hard restriction on the scalability of data processing in some important areas, where the collection of large image series is routine. I will describe how this can be overcome using a deep learning neural network which performs atomic model abstraction from low dose high framerate experimental graphene images Although the training of such neural networks requires significantly more effort than classical image processing, this method is more general in that it can autonomously process large datasets and can readily be extended to studies of other two-dimensional materials. Using this approach, it is possible to identify many instances of specific defect transitions and to map the lifetimes of defect states as a probe of local kinetics. In turn these can be used as input to density functional theory to model the potential energy landscape for the transitions. In studies of metallic nanoparticles catalysts there is an urgent need to “handshake” EM data with bulk structural probes using realistic heterogeneous samples.
For many materials, including biological molecules and larger structures the development of dose efficient strategies is required. I will highlight the use of fast detectors for optimised phase retrieval. Important the challenges in applying electron Ptychography under low dose conditions which requires the collection of a 4D dataset of far field diffraction patterns as a function of probe position at the specimen plane to recover the complex specimen wavefunction will be discussed.
Back to Computational Microscopy Tutorials