Machine Learning-Based Reward-Driven Tuning of Scanning Probe Microscopy: Toward Fully Automated Microscopy

Area of Science:

• Materials Science
• Nanotechnology
• Surface Science

Background:

• Tapping mode (intermittent contact mode) is a widely used imaging technique in scanning probe microscopy (SPM).
• Manual optimization of tapping mode is time-consuming, operator-dependent, and can lead to sample/probe damage and poor reproducibility.
• Existing control and machine learning methods struggle with the complex optimization of tapping mode imaging.

Purpose of the Study:

• To develop a novel reward-driven workflow for automating the optimization of SPM tapping mode.
• To create a sample-agnostic measure of image quality that mimics human operator decision-making.
• To enhance the efficiency, consistency, and reliability of tapping mode SPM operation.

Main Methods:

• A reward-driven workflow was designed to automate SPM tapping mode parameter optimization.
• A reward function was developed, incorporating physical and empirical knowledge of scan quality.
• The workflow was tested across diverse probes and sample types.

Main Results:

• The automated workflow successfully determined optimal scanning parameters for tapping mode.
• The system produced consistent, high-quality images across various probes and samples.
• The reward function provided a sample-agnostic measure of image quality.

Conclusions:

• The developed reward-driven workflow offers an efficient and reliable method for optimizing SPM tapping mode.
• This automation reduces the need for manual adjustments, saving time and minimizing damage.
• The approach demonstrates potential for improving SPM usability for new users and complex samples.