Four-dimensional scanning transmission electron microscopy (4D-STEM) provides a unique opportunity to probe local crystal structure in diffraction space by scanning a converged probe across a designated field of view and recording a diffraction pattern at each location. This technique has the ability to provide a wealth of information on materials with complex arrangements. To interpret and analyze this data, it is necessary to develop methods to separate diffraction patterns into groups with physically meaningful similarities. New approaches currently focus on classifying patterns based on Bragg disks in the diffractograms, which has made headway in creating interpretable collections of patterns of bulk-structures with strong Bragg scattering. However, a robust and easily interpretable clustering method for 4D-STEM datasets containing both ordered and disordered regions has yet to be realized. We apply unsupervised learning on features extracted from the data to improve the ability to identify disordered structures that lack strong Bragg scattering. We present a detailed analysis of the features necessary to optimize classification of a wide variety of structures, showing a case study of the ultrathin (&lt; 2 nm) Au-based nanowires taking on complex double helical structures. This outcome reveals the presence of molecular icosahedral structures, which are systemically excluded in the Bragg disk classification approach.

Direct observation of icosahedral packing in double helical Au-based nanowires using unsupervised learning on four-dimensional scanning transmission electron microscopy (4D-STEM) datasets

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.