Multi-Dimensional Multiplexed Metasurface for Multifunctional Near-Field Modulation by Physics-Driven Intelligent...

Area of Science:

• Materials Science
• Computational Science
• Information Technology

Background:

• Metasurfaces offer tunable properties through engineered meta-atom arrangements.
• Advanced metasurfaces with multiple degrees of freedom (MDOF) present design challenges due to vast design spaces.
• Existing data-driven intelligent design methods struggle with data scarcity, interpretability, and generalization.

Purpose of the Study:

• To propose a physics-driven intelligent design (PDID) paradigm for creating MDOF multiplexed metasurfaces.
• To enhance the physical interpretability and reduce data dependency of metasurface design.
• To demonstrate a computationally efficient and versatile design tool.

Main Methods:

• Developed a PDID paradigm integrating physical prior knowledge into deep neural networks.
• Applied the PDID method to the design of MDOF multiplexed metasurfaces.
• Validated the designed metasurfaces through experimental testing.

Main Results:

• PDID reduced design time and database size by two orders of magnitude compared to traditional intelligent designs.
• Achieved enhanced physical interpretability and reduced reliance on extensive datasets.
• Experimental validation confirmed the versatility and computational efficiency of the PDID method.

Conclusions:

• The PDID paradigm offers a novel and effective tool for designing advanced metasurfaces.
• Integrating physical knowledge with machine learning addresses key challenges in computational material design.
• This interdisciplinary approach holds significant potential for future innovations in science and technology.