Optimal STEM Convergence Angle Selection Using a Convolutional Neural Network and the Strehl Ratio.

Area of Science:

• Materials Science
• Physics
• Microscopy

Background:

• High-resolution imaging in scanning transmission electron microscopy (STEM) depends on precise convergence angle selection.
• Current methods often use heuristics like Rayleigh's quarter-phase rule, leading to suboptimal probe sizes and reduced resolution due to measurement uncertainties.

Purpose of the Study:

• To develop an accurate and efficient method for evaluating probe size in STEM.
• To improve the selection of convergence angles for optimal STEM imaging.
• To enable rapid, automated alignment of aberration-corrected electron microscopes.

Main Methods:

• Utilized the Strehl ratio as a metric for evaluating probe quality.
• Developed and trained a convolutional neural network (CNN) on simulated electron Ronchigram datasets.
• Trained the CNN to select optimal convergence angles based on the Strehl ratio.

Main Results:

• The CNN, trained on the Strehl ratio, outperformed experienced microscopists in selecting convergence angles from simulated Ronchigrams.
• The AI-selected convergence angles resulted in probe sizes 85% closer to optimal on average.
• The CNN achieved this assessment at millisecond speeds, significantly faster than human evaluation.

Conclusions:

• The Strehl ratio, combined with machine learning, provides an accurate and efficient criterion for evaluating STEM probe size.
• AI-driven assessment of Ronchigrams offers a viable path toward automated alignment of electron microscopes.
• While trends are well-modeled, high accuracy on experimental data requires extensive training datasets.