Closing the loop: autonomous experiments enabled by machine-learning-based online data analysis in synchrotron...

Area of Science:

• Materials Science
• Data Science
• Physics

Background:

• Increasing data volumes in X-ray scattering necessitate advanced analysis techniques.
• Machine learning (ML) offers potential for automated, real-time data interpretation.
• Closed-loop feedback systems can enhance experimental efficiency and control.

Purpose of the Study:

• To integrate ML-based online data analysis into a closed-loop workflow for X-ray reflectometry (XRR).
• To demonstrate the autonomous control of a vacuum deposition setup using ML analysis of XRR data.
• To extract physical thin-film parameters (thickness, density, roughness) using ML.

Main Methods:

• Development of a combined one-dimensional convolutional neural network (CNN) and multilayer perceptron model.
• Training the ML model to extract thin-film parameters from XRR data, incorporating prior knowledge.
• Integration of ML-based online analysis into a beamline for real-time feedback control of a vacuum deposition system.

Main Results:

• Accurate and robust analysis of X-ray reflectometry curves and Bragg reflections using ML methods.
• Successful demonstration of autonomous control over a vacuum deposition setup based on ML analysis.
• Validation of ML for real-time monitoring and decision-making in thin-film growth.

Conclusions:

• ML techniques, specifically CNNs and MLPs, are effective for automated XRR data analysis.
• Beamline integration of ML enables closed-loop feedback for real-time experimental control.
• This approach accelerates thin-film characterization and deposition processes.