Unidirectional imaging using deep learning-designed materials

Area of Science:

• Optics and Photonics
• Metamaterials
• Deep Learning Applications

Background:

• Traditional imaging systems lack directional control, allowing light propagation in both forward and backward directions.
• Unidirectional imagers are crucial for applications requiring controlled light flow, such as optical isolators and one-way communication.

Purpose of the Study:

• To demonstrate the first polarization-insensitive and broadband unidirectional imager.
• To explore the use of deep learning-trained diffractive layers for achieving unidirectional imaging.
• To investigate wavelength-selective unidirectional imaging capabilities.

Main Methods:

• Designing and fabricating successive linear and isotropic diffractive layers.
• Employing deep learning algorithms for training the diffractive layers.
• Experimental validation using terahertz (THz) radiation under broadband illumination.
• Developing a wavelength-selective imager by multiplexing operations with different wavelengths.

Main Results:

• Successful demonstration of a polarization-insensitive, broadband unidirectional imager.
• The fabricated imager maintained functionality across a wide spectral band, despite monochromatic training.
• Experimental results with THz radiation closely matched numerical simulations.
• A wavelength-selective unidirectional imager was successfully created.

Conclusions:

• Diffractive unidirectional imaging offers a novel approach to controlling light propagation.
• The developed technology has significant potential for applications in security, defense, telecommunications, and privacy protection.
• Structured materials and deep learning provide a powerful platform for advanced optical device design.