Deep neural network-based automatic metasurface design with a wide frequency range.

Area of Science:

• Metasurface design
• Applied electromagnetics
• Machine learning in engineering

Background:

• Conventional metasurface design is computationally intensive and time-consuming.
• Machine learning offers a promising alternative for efficient metasurface inverse design.
• Deep Neural Networks (DNNs) can potentially streamline the design process.

Purpose of the Study:

• To present a DNN-based inverse design procedure for metasurfaces.
• To achieve an ultra-wide working frequency band for metasurface applications.
• To reduce computational resources and design time compared to traditional methods.

Main Methods:

• Utilized Deep Neural Networks (DNNs) for inverse design.
• Employed 8 ring-shaped patterns to generate resonant notches across 4-45 GHz.
• Proposed and evaluated two distinct DNN network architectures.
• Restricted DNN output in one architecture for faster computation.

Main Results:

• Achieved an average accuracy of over 90% in generating metasurface structures.
• Demonstrated a network accuracy above 91% with restricted output.
• Successfully designed metasurfaces for an ultra-wide working frequency band.
• Significantly reduced computational time for metasurface structure determination.

Conclusions:

• The DNN-based approach provides a direct method for metasurface structure generation.
• The developed model enables ultra-wide working frequencies with high accuracy.
• This platform facilitates engineering projects by simplifying complex electromagnetic theory requirements.