Metamaterial perfect absorber with morphology-engineered meta-atoms using deep learning.

Area of Science:

• Electromagnetics and Materials Science
• Metamaterials and Metasurfaces

Background:

• Conventional metamaterial perfect absorbers (MPAs) are limited by regularly-shaped unit structures.
• Existing analysis methods restrict the design flexibility and absorption properties of MPAs.

Purpose of the Study:

• To propose a novel MPA structure utilizing general polygon-shaped meta-atoms.
• To enable flexible electromagnetic properties, including dual-band absorption, through irregular unit structures.
• To develop a deep neural network for predicting MPA parameters and absorptivity.

Main Methods:

• Introduced a metamaterial perfect absorber (MPA) with a general polygon-shaped meta-atom.
• Leveraged the degrees-of-freedom offered by irregular unit structures for enhanced properties.
• Constructed a deep neural network for inverse and forward prediction of MPA parameters and absorptivity.

Main Results:

• The proposed MPA structure demonstrates flexible properties, including dual-band absorption.
• The deep neural network achieved a low mean-square error of 0.0017 on the validation set.
• Successfully predicted MPA parameters from absorptivity and vice versa.

Conclusions:

• Irregularly shaped meta-atoms in MPAs offer enhanced design flexibility and tunable properties.
• Deep neural networks are effective tools for designing complex artificial electromagnetic structures.
• This study lays the groundwork for advanced metamaterial and metasurface applications.