Broadband Solar Metamaterial Absorbers Empowered by Transformer-Based Deep Learning

Area of Science:

• Metamaterials
• Solar Energy Harvesting
• Nanotechnology

Background:

• Solar metamaterial absorbers (SMAs) are crucial for efficient solar energy harvesting.
• Traditional SMA design involves complex, time-consuming optimization processes.
• Deep learning (DL) applications in broadband SMA design are limited due to spectral complexity.

Purpose of the Study:

• To develop a deep learning (DL) model for high-performance solar metamaterial absorber (SMA) design.
• To overcome the limitations of traditional DL methods in handling broadband spectral features.
• To create a flexible tool for on-demand metamaterial design with diverse optical functions.

Main Methods:

• Developed a metamaterial spectrum transformer (MST) DL model.
• MST divides optical spectra into patches to enhance learning capability and prevent overfitting.
• Applied the model to design and fabricate graded-refractive-index nanostructure SMAs.

Main Results:

• Achieved high average absorptance of 94% across a broad solar spectrum.
• Demonstrated superior performance compared to existing state-of-the-art metamaterial absorbers.
• Outdoor testing indicated high annual solar energy collection efficiency (1061 kW h m⁻²).

Conclusions:

• The developed DL model enables rapid and intelligent design of SMAs.
• The metamaterial spectrum transformer (MST) approach significantly boosts DL performance for broadband applications.
• This work provides a versatile tool for designing various metamaterials and metadevices.