Understanding lattice deformations is crucial for determining the properties of nanomaterials. This paper demonstrates a rapid and semi-automated unsupervised machine learning approach using divisive hierarchical clustering to uncover multi-scale deformations in materials from four-dimensional scanning transmission electron microscopy (4D-STEM) diffraction data. The method overcomes limitations of large 4D data analysis without prior sample knowledge, revealing various deformations (strain, lattice distortion, bending contour) impacting nanomaterial properties and device performance. This data-driven procedure offers insights into materials' intrinsic structures and accelerates materials discovery.

Chuqiao Shi, Michael C. Cao, Sarah M. Rehn, Sang-Hoon Bae, Jeehwan Kim, Matthew R. Jones, David A. Muller, Yimo Han