Related Experiment Video
Updated: Aug 4, 2025
11:05
Knowledge Based Cloud FE Simulation of Sheet Metal Forming Processes
Published on: December 13, 2016
12.2K
A knowledge-inherited learning for intelligent metasurface design and assembly
Yuetian Jia1,2,3, Chao Qian4,5,6, Zhixiang Fan1,2,3
1ZJU-UIUC Institute, Interdisciplinary Center for Quantum Information, State Key Laboratory of Extreme Photonics and Instrumentation, Zhejiang University, Hangzhou, 310027, China.
Light, Science & Applications
|March 30, 2023
Summary
A new knowledge-inherited deep learning method enables efficient, multi-object metasurface design. This approach overcomes limitations of traditional methods, allowing for adaptable and complex metadevices like intelligent origami for satellite communications.
Area of Science:
- Optics and Photonics
- Deep Learning Applications
- Metamaterial Design
Background:
- Deep learning is a powerful tool in optics and photonics for material design, system optimization, and automation.
- Current deep learning methods for metasurface design are limited by sample collection and training for individual metamaterials, failing for large problem sizes.
- Conventional numerical simulations and physics-based methods for metasurface design are time-consuming, inefficient, and experience-dependent.
Purpose of the Study:
- To propose a novel, knowledge-inherited paradigm for multi-object and shape-unbound metasurface inverse design.
- To overcome the limitations of current deep learning approaches in handling large-scale and complex metasurface designs.
- To enable the creation of adaptable and versatile intelligent metadevices.
Main Methods:
- A knowledge-inherited paradigm inspired by object-oriented programming is introduced.
- Neural networks inherit knowledge from parent metasurfaces and are assembled to create offspring metasurfaces.
- The paradigm is benchmarked through the free design of aperiodic and periodic metasurfaces.
Main Results:
- The proposed paradigm achieves high accuracy (86.7%) in the free design of metasurfaces.
- Demonstration of an intelligent origami metasurface for compatible and lightweight satellite communication facilities.
- The method effectively handles multi-object and shape-unbound metasurface inverse design.
Conclusions:
- The knowledge-inherited paradigm offers a new avenue for automatic metasurface design.
- The assemblability of the approach broadens the adaptability of intelligent metadevices.
- This work significantly advances the efficiency and scope of deep learning in metasurface engineering.
