Deep Learning Inverse Design of Phase-Change Reconfigurable Terahertz Metadevices for Multidimensional Secure...

Area of Science:

• Physics
• Electrical Engineering
• Computer Science

Background:

• Next-generation 6G networks face increasing data demands and cyber threats, necessitating enhanced physical-layer security.
• Terahertz (THz) waves offer unique properties like high bandwidth and directionality, ideal for secure, high-capacity communication.
• Current design methods for THz metadevices are often iterative and time-consuming.

Purpose of the Study:

• To introduce a deep-learning-enabled inverse-design framework for creating dynamically reconfigurable THz metadevices.
• To enable adaptive, multidimensional encryption at the physical layer for secure THz communication.
• To develop versatile meta-architectures for advanced wireless applications.

Main Methods:

• Utilized a residual neural network for direct mapping of electromagnetic responses to device geometries.
• Incorporated continuous material phase transitions in Ge2Sb2Te5 (GST) for dynamic reconfiguration.
• Implemented an inverse-design approach to bypass traditional iterative design bottlenecks.

Main Results:

• Generated versatile meta-architectures with high precision and speed.
• Achieved eight-channel encrypted holography with multiplexed control over polarization, depth, and phase transitions in GST.
• Demonstrated a reconfigurable diffractive THz neural metadevice performing universal logic operations under a dual-key security protocol.

Conclusions:

• The deep-learning framework enables rapid, high-precision design of adaptive THz metadevices for secure communication.
• The developed metadevices offer advanced physical-layer encryption capabilities, including encrypted holography and secure logic operations.
• This work establishes a new paradigm for secure, adaptive THz communication systems, crucial for 6G and beyond.