Machine Learning-Enabled Fast-Scan STEM for the Precise Identification of Labile Single-Atom Sites.

Area of Science:

• Materials Science
• Catalysis
• Nanotechnology

Background:

• Single-atom catalysts (SACs) performance depends critically on their local atomic environment.
• Resolving transient atomic configurations is challenging due to thermal fluctuations and electron-beam effects.
• Conventional scanning transmission electron microscopy (STEM) suffers from noise and long dwell times, obscuring dynamic processes.

Purpose of the Study:

• To introduce a machine-learning-enhanced fast-scan STEM methodology for accurate single-atom localization.
• To investigate the dynamic structural disorder and local electric field heterogeneity of platinum atoms on MoS2.
• To establish a computational framework for atomic-scale structure resolution under noisy imaging conditions.

Main Methods:

• Development and application of a machine-learning-enhanced fast-scan STEM technique.
• Differential phase contrast (DPC) imaging for analyzing local electric fields.
• Computational framework for resolving atomic structures from noisy images.

Main Results:

• Accurate identification and localization of individual platinum atoms on MoS2 supports were achieved.
• Random atomic displacements of approximately 3.2% relative to ideal lattice sites were quantified.
• Persistent heterogeneity in the local electric field across platinum atomic columns was observed.

Conclusions:

• Single-atom sites exist in a highly dynamic, non-ideal equilibrium state.
• The developed methodology overcomes limitations of high-noise imaging in STEM.
• This study provides crucial insights into the dynamic nature of SACs at the atomic scale.