Continuous three-dimensional imaging of nanoscale dynamics by in situ electron tomography

TL;DR

This paper presents a novel framework for continuous 3D electron tomography that enables real-time nanoscale imaging of dynamic transformations with reduced electron dose, combining continuous tilting and deep learning reconstruction.

Contribution

It introduces a self-supervised deep learning method that incorporates temporal information, allowing 3D imaging of evolving structures from a single tilt series.

Findings

Successfully validated with simulations and experiments.

Captured morphological evolution of Au nanostars and alloying in nanocubes.

Achieved dose-efficient, in situ 3D imaging of nanomaterials.

Abstract

Direct observation of nanoscale transformations in three dimensions (3D) is essential for understanding materials evolution under operating conditions, yet dynamic electron tomography remains limited by slow tilt series acquisition and by reconstruction methods that assume static specimens. These constraints prevent continuous 3D imaging of evolving structures and require electron doses that can alter the specimens and their dynamics. Here, we introduce a framework for dynamic electron tomography that combines continuous tilting with a self-supervised deep-learning reconstruction strategy. Our approach incorporates the temporal aspect into the electron tomography reconstruction process to recover 3D volumes at arbitrary time points from a single tilt series. We validate the method using simulations and demonstrate its merit in experimental studies of heat-induced transformations, including morphological evolution of Au nanostars and alloying in Au@Ag nanocubes. Our results establish a practical framework for dynamic, dose-efficient electron tomography, enabling in situ 3D investigation of nanomaterial transformations as well as the characterization of beam-sensitive structures.