Automated structure discovery in atomic force microscopy.

Area of Science:

• Surface science
• Scanning probe microscopy
• Computational chemistry

Background:

• Atomic force microscopy (AFM) is crucial for imaging organic molecules on surfaces.
• Interpreting AFM images of nonplanar molecules is challenging due to image distortion.
• Current methods are limited to analyzing nearly planar molecules.

Purpose of the Study:

• To develop a deep learning approach for analyzing complex AFM images.
• To enable atomic and chemical structural resolution of nonplanar molecules on surfaces.
• To expand the application of high-resolution AFM to diverse molecular systems.

Main Methods:

• Developed a deep learning infrastructure to correlate AFM images with molecular configuration descriptors.
• Applied the methodology to low-temperature AFM measurements of 1S-camphor on Cu(111).
• Utilized machine learning to predict molecular structures directly from AFM data.

Main Results:

• Successfully resolved distinct adsorption configurations of 1S-camphor on a Cu(111) surface.
• Demonstrated the ability to predict molecular structure directly from AFM images.
• Overcame limitations in interpreting distorted AFM images of nonplanar molecules.

Conclusions:

• The developed deep learning method enhances AFM's capability for molecular structure determination.
• This approach allows for routine atomic and chemical structural resolution of individual nonplanar molecules.
• Opens new avenues for high-resolution AFM studies across various chemical systems.