Deep learning based atomic defect detection framework for two-dimensional materials.

Area of Science:

• Materials Science
• Nanotechnology
• Condensed Matter Physics

Background:

• Atomic defects in 2D transition metal dichalcogenides (TMDs) significantly impair field-effect transistor (FET) performance.
• Scanning tunneling microscopy (STM) is crucial for atomic defect identification but suffers from long analysis times.
• Existing automated defect detection methods face challenges due to low signal-noise ratios and limited data.

Purpose of the Study:

• To develop an efficient deep learning-based framework (DL-ADD) for automated atomic defect detection in TMDs.
• To improve the accuracy and generalizability of defect detection across different TMD materials.
• To address limitations of current STM analysis and automated detection systems.

Main Methods:

• Implemented a deep learning-based atomic defect detection framework (DL-ADD).
• Utilized data augmentation, color preprocessing, and noise filtering techniques.
• Developed a detection model trained on MoS2 and tested for generalizability on WS2.

Main Results:

• DL-ADD achieved precise atomic defect detection in MoS2 with an average F2-score of 0.86.
• The framework demonstrated good generalizability for defect detection in WS2, achieving an average F2-score of 0.89.
• The proposed methods effectively improved detection quality despite challenges like low signal-noise ratio.

Conclusions:

• The DL-ADD framework offers an efficient and accurate solution for atomic defect detection in 2D TMDs.
• The model's generalizability suggests its potential for widespread application across various TMD materials.
• This work advances automated analysis of defects in 2D materials, crucial for electronic device development.