Source-linked AI summary
Manifold Learning of Four-dimensional Scanning Transmission Electron Microscopy
Xin Li, Ondrej E. Dyck, Mark P. Oxley, Andrew R. Lupini, Leland McInnes, John Healy, Stephen Jesse, Sergei V. Kalinin
TL;DR
Interpreting large Ronchigram datasets remains challenging, and much of their information is not fully exploited. The paper applies data-driven manifold learning to 4D-STEM patterns, separating sublattice-dependent deflection patterns even in weak graphene signals.
Problem
Ronchigram datasets contain information that is not fully exploited, and how physical parameters affect electron distributions on pixelated detectors remains incompletely understood.
Source: Synthetic Dataset Analysis · p. 7Method
The authors use data-driven manifold learning to spatially map local inhomogeneities in Ronchigram datasets without a priori models or data constraints.
Source: Synthetic Dataset Analysis · p. 7Results
Manifold learning separates weak graphene Ronchigrams by local, sublattice-dependent symmetry, revealing real-space neighboring effects on deflection patterns.
Source: Synthetic Dataset Analysis · pp. 6–7Takeaways & Limitations
The approach could serve as a fast screening tool for experts analyzing large-scale Ronchigram patterns.
Source: Synthetic Dataset Analysis · p. 7
Abstract
from arXiv · showhide
Four-dimensional scanning transmission electron microscopy (4D-STEM) of local atomic diffraction patterns is emerging as a powerful technique for probing intricate details of atomic structure and atomic electric fields. However, efficient processing and interpretation of large volumes of data remain challenging, especially for two-dimensional or light materials because the diffraction signal recorded on the pixelated arrays is weak. Here we employ data-driven manifold leaning approaches for straightforward visualization and exploration analysis of the 4D-STEM datasets, distilling real-space neighboring effects on atomically resolved deflection patterns from single-layer graphene, with single dopant atoms, as recorded on a pixelated detector. These extracted patterns relate to both individual atom sites and sublattice structures, effectively discriminating single dopant anomalies via multi-mode views. We believe manifold learning analysis will accelerate physics discoveries coupled between data-rich imaging mechanisms and materials such as ferroelectric, topological spin and van der Waals heterostructures.
Synthetic Dataset Analysis
Manifold learning reveals sublattice-related neighboring effects in synthetic graphene Ronchigrams and distinguishes experimental sublattices and dopant anomalies despite weak signal.
Synthetic Dataset Analysis
- Synthetic Ronchigrams form clusters around graphene’s two sublattices, with opposite deflection patterns on mirrored atom sites and standard deviations (10-6-10-5) an order of magnitude below means (10-4).Cluster labels at Si dopants have irregular real-space positions.Source: Synthetic Dataset Analysis · p. 6
- Similarity loadings confirm opposite deflection patterns lie on mirrored sites across sublattices, not opposite sides of the same atom, and show distinct patterns around the central 4-fold Si dopant.The analysis separates deflection patterns into two spatially sublattice-related classes rather than radially symmetric patterns around each atom.Source: Synthetic Dataset Analysis · p. 6
Experimental Dataset Analysis
- Despite weak diffraction from monolayer graphene, manifold learning identifies sublattice structure in experimental Ronchigrams without signal preprocessing, and similarity loadings distinguish both sublattices and the 4-fold Si dopant.Clusters 0,1,5,6 are assigned to sublattice A and clusters 2,3,7,8 to sublattice B; clusters 4 and 9 are not indicative of either.Source: Synthetic Dataset Analysis · pp. 6–7
- Experimental cluster Ronchigrams, after subtracting the overall mean, show deflection patterns consistent with their spatial cluster-label positions relative to atom sites.The original cluster means and a compact illustration are provided in supplementary figures.Source: Synthetic Dataset Analysis · p. 7
Discussion
- At -10 nm defocus, the synthetic-data manifolds have only two groups and cluster-label positions no longer indicate graphene sublattices, underscoring the need to control and record experimental conditions.The discussion presents manifold learning as a way to spatially map Ronchigram inhomogeneities without a priori models or constraints.Source: Synthetic Dataset Analysis · p. 7
- Manifold learning efficiently extracts hidden information from weak 4D-STEM Ronchigrams, separating patterns by sublattice-dependent local symmetry rather than radial symmetry around each atom.The authors propose applications including denoising, interpolation, large-scale screening, and anomaly detection for real-time feedback and control.Source: Synthetic Dataset Analysis · pp. 7–8