Identification and Structural Characterization of Twisted Atomically Thin Bilayer Materials by Deep Learning.

Area of Science:

• Materials Science
• Condensed Matter Physics
• Nanotechnology

Background:

• Two-dimensional materials like transition metal dichalcogenides are crucial for next-generation electronics and optoelectronics.
• Twisted bilayer structures exhibit unique physical properties, driving significant research interest.

Purpose of the Study:

• To develop an automated, rapid, and accurate method for characterizing chemical vapor deposition (CVD)-grown molybdenum disulfide (MoS2) flakes.
• To precisely determine the thickness and twist angles of MoS2 bilayers using artificial intelligence.

Main Methods:

• Utilized optical microscopy to capture the color space of CVD-grown MoS2.
• Applied a semantic segmentation convolutional neural network (CNN) for MoS2 thickness identification.
• Trained a second CNN model on over 10,000 synthetic images to predict twist angles in bilayer MoS2.
• Validated deep learning predictions using second harmonic generation and Raman spectroscopy.

Main Results:

• Demonstrated accurate and rapid identification of MoS2 flake thicknesses via CNN.
• Achieved precise predictions of twist angles in CVD-grown bilayer MoS2 flakes.
• Validated the deep learning models' performance through experimental spectroscopic techniques.

Conclusions:

• Introduced a scalable deep learning methodology for automated inspection of twisted atomically thin CVD-grown bilayers.
• The developed approach facilitates high-throughput characterization of 2D materials for electronic applications.