Chirality Detection in Scanning Tunneling Microscopy Data Using Artificial Intelligence.

Area of Science:

• Surface chemistry
• Supramolecular chemistry
• Materials science

Background:

• Chiral molecular networks self-assemble at surfaces, crucial for advanced applications.
• Scanning probe microscopy (SPM) images these networks but suffers from low contrast and high noise.
• Traditional image analysis is hindered by long acquisition times and manual labor, leading to errors.

Purpose of the Study:

• To develop an AI-driven method for accurate analysis of chiral molecular networks imaged by SPM.
• To overcome limitations of low contrast, high noise, and manual analysis in SPM data.
• To reduce reliance on extensive real-world datasets for AI model training.

Main Methods:

• Generation of realistic synthetic SPM images of chiral molecular networks.
• Training state-of-the-art object detection architectures (e.g., Faster R-CNN) on synthetic data.
• Evaluating model performance on real SPM data, assessing robustness to noise and zoom variations.

Main Results:

• A Faster R-CNN model trained solely on synthetic data achieved 99% mean average precision on real data.
• Synthetic data training outperformed augmented datasets for chiral unit-cell detection.
• The AI approach demonstrated high robustness against experimental noise and varying zoom levels.

Conclusions:

• AI models trained on synthetic SPM images provide a robust and accurate method for analyzing chiral molecular networks.
• This approach significantly reduces the need for manual analysis and large real datasets.
• The method shows generalizability to different chiral network structures.