ATOMIC is an innovative AI-powered optical microscope platform developed at Duke University that integrates foundation models like OpenAI's ChatGPT and Meta's Segment Anything Model to enable autonomous and highly accurate analysis of two-dimensional materials.

This system achieves up to 99.4% accuracy in identifying material features such as defects and layer overlaps, even under poor imaging conditions, surpassing many traditional methods.

ATOMIC manages tasks like sample focusing, defect detection, and layer identification through a combination of AI models, significantly reducing analysis time while maintaining high precision.

Designed specifically for analyzing ultra-thin 2D materials critical for next-generation electronics, sensors, and quantum devices, ATOMIC simplifies a process that traditionally requires extensive expert knowledge.

The adoption of AI in microscopy represents a transformative shift towards autonomous scientific research, accelerating discoveries across fields like materials science, chemistry, and biology.

Wang emphasizes that AI is intended to complement and amplify human expertise, not replace it, allowing scientists to focus on complex problem-solving and innovative research.

The development of ATOMIC highlights the importance of combining AI capabilities with human judgment to reduce analysis time and foster new scientific insights.

The platform employs a zero-shot learning approach, leveraging pre-trained foundation models, which allows it to adapt without needing thousands of labeled images, making it highly efficient and flexible.

ATOMIC automates workflows by handling sample movement, focusing, and lighting adjustments, and uses a specialized topological correction algorithm to recognize overlapping layers in 2D materials.

The system was developed by Haozhe 'Harry' Wang's lab at Duke University, showcasing its ability to analyze 2D materials with accuracy comparable to human experts in a fraction of the time.

While validated across various samples and conditions, the system's robustness still requires human oversight to interpret AI findings and manage potential unpredictable outcomes.

ATOMIC was specifically customized to address challenges like overlapping layers in microscopic images, incorporating a topological correction algorithm to accurately isolate single-layer regions.