Imaging 3D polarization dynamics via deep learning 4D-STEM

TL;DR

This paper introduces a deep learning-based 4D-STEM method to reconstruct 3D polarization maps in ferroelectric thin films, enabling detailed nanoscale analysis of polarization dynamics and topological features.

Contribution

The study presents a novel deep learning framework combined with 4D-STEM for high-precision 3D polarization mapping in ferroelectric materials, overcoming previous depth resolution limitations.

Findings

Reconstructed 3D polarization maps with picometer accuracy.

Observed polarization switching via vector rotation toward <111> minima.

Identified topological protection effects in polarization variation.

Abstract

Recent advances in ferroelectrics highlight the role of three-dimensional (3D) polar entities in forming topological polar textures and generating giant electromechanical responses, during polarization rotation. However, current electron microscopy methods lack the depth resolution to resolve the polarization component along the electron beam direction, which restricts full characterization. Here, we present a deep learning framework combined with four-dimensional scanning transmission electron microscopy to reconstruct 3D polarization maps in Ba0.5Sr0.5TiO3 thin-film capacitors with picometer-level accuracy under applied electric fields. Our approach enables observation of polar nanodomains consistent with the polar slush model and shows that switching occurs through coordinated vector rotation toward <111> energy minima, rather than magnitude changes. Furthermore, regions with higher topological density exhibit smaller polarization variation when the electric field changes, indicating topological protection. Our work reveals the value of 3D polarization mapping in elucidating complex nanoscale polar phenomena, with broad implications for emergent ferroelectrics.