Compounding Meta-Atoms into Metamolecules with Hybrid Artificial Intelligence Techniques

Area of Science:

• Optics and Photonics
• Materials Science
• Artificial Intelligence

Background:

• Metasurfaces offer advanced control over light properties (amplitude, phase, frequency) through spatially variant metamolecules.
• Designing complex metamolecules with emergent properties is challenging due to intricate mechanisms and numerous degrees of freedom.

Purpose of the Study:

• To propose a hybrid artificial intelligence (AI) framework for the inverse design of metamolecules in metasurfaces.
• To overcome limitations in designing multielement systems by decomposing the design process.

Main Methods:

• A hybrid AI framework combining compositional pattern-producing networks and cooperative coevolution was developed.
• The framework decomposes metamolecule design into independent meta-atom design tasks.
• Deep learning and evolutionary algorithms are used to solve individual meta-atom design problems.

Main Results:

• The framework successfully designed metallic metamolecules for arbitrary manipulation of light polarization and wavefront.
• Experimental validations confirmed the efficacy and reliability of the proposed design strategy.
• The approach demonstrated a labor-saving and systematic method for designing large-scale metasurfaces.

Conclusions:

• The hybrid AI framework offers a promising approach to expedite the inverse design of metamolecules for metasurfaces.
• This method simplifies the design of complex optical devices by addressing smaller, independent design tasks.
• The study highlights the potential of AI in advancing metasurface technology and optical manipulation.