Automated Structure Discovery for Scanning Tunneling Microscopy.

Area of Science:

• Surface Science
• Materials Science
• Computational Chemistry

Background:

• Scanning Tunneling Microscopy (STM) provides simultaneous geometric and electronic structure information.
• Interpreting STM signals is challenging, limiting studies to simple, planar samples.
• Existing methods for structure discovery are primarily developed for noncontact atomic force microscopy (nc-AFM).

Purpose of the Study:

• To introduce a machine learning tool for automated atomic structure prediction directly from STM images.
• To adapt and apply successful nc-AFM structure discovery methods to STM data.
• To enable direct atomic structure determination for a wider range of samples using STM.

Main Methods:

• Development of automated structure discovery for STM (ASD-STM) using machine learning.
• Leveraging established structure discovery techniques from nc-AFM research.
• Application and validation of the ASD-STM tool on experimental STM images of organic molecules.

Main Results:

• Achieved good accuracy in predicting atomic structures from STM images.
• Demonstrated qualitative success in chemical identification of organic molecules.
• Highlighted areas for future development to enhance ASD-STM capabilities.

Conclusions:

• ASD-STM facilitates direct atomic structure discovery from STM images, broadening its applicability.
• The method makes advanced structure analysis accessible to a wider scanning probe microscopy community.
• This work paves the way for developing more sophisticated machine learning approaches for STM data analysis.