Unpaired Learning-Enabled Nanotube Identification from AFM Images.

Area of Science:

• Materials Science
• Nanotechnology
• Data Science

Background:

• Single-walled carbon nanotubes (SWCNTs) offer exceptional properties but their network morphology is difficult to characterize.
• Accurate characterization is crucial for optimizing SWCNT-based devices and applications.

Purpose of the Study:

• To develop a robust deep learning approach for precise nanotube morphology extraction from AFM images.
• To overcome challenges posed by substrate roughness in nanotube characterization.

Main Methods:

• Utilized a cycleGAN-based image-to-image translation framework to generate pure substrate images from AFM data.
• Implemented a specialized loss function for accurate transformation of AFM images.
• Developed a subtraction method to isolate nanotube structures from the original images.

Main Results:

• Successfully extracted SWCNT morphologies, even on rough substrates exceeding nanotube diameter.
• Demonstrated superior sensitivity and accuracy compared to traditional and supervised learning methods.
• Validated the approach using simulations and real-world applications in flexible carbon nanotube transistors.

Conclusions:

• The deep learning method significantly improves nanotube network characterization.
• Provides valuable insights for optimizing fabrication processes in flexible electronics.
• The methodology is adaptable for other nanomaterial-based electronic devices.