Accurate Inverse Design of Broadband Solar Metamaterial Absorbers via Joint Forward-Inverse Deep Learning.

Area of Science:

• Materials Science
• Optics
• Artificial Intelligence

Background:

• Designing broadband, high-efficiency solar absorbers is complex due to inverse design challenges.
• Traditional methods struggle with the ill-posed mapping from optical properties to physical structures.

Purpose of the Study:

• To propose a joint forward-inverse deep learning framework for rapid and accurate optimization of multilayer metamaterial absorbers.
• To address the one-to-many ambiguity and improve prediction accuracy in inverse design.

Main Methods:

• Integrated an inverse network (Modified Swin Transformer) with a forward proxy (Multilayer Perceptron).
• Employed end-to-end training in a consistency-driven cycle for joint optimization.
• Optimized W/SiO2 hyperbolic metamaterial stacks with TiO2/SiO2 anti-reflection coatings.

Main Results:

• Achieved 97.4% average absorptivity over the 400-1750 nm solar spectrum.
• Demonstrated polarization insensitivity and wide-angle performance up to 60° incidence.
• Fabricated absorber reached 86.3 °C under natural sunlight.

Conclusions:

• The joint forward-inverse deep learning framework accelerates the design of high-performance solar thermal metamaterials.
• This approach offers a powerful and scalable paradigm for intelligent material design.
• The optimized absorber shows significant potential for solar energy applications.