Meta-lenses are 2D optical elements with nanostructures that control light waves, allowing for rapid phase and amplitude changes. They offer features like polarization control and higher spatial resolution compared to traditional lenses, making them valuable for applications such as biomedical imaging and virtual reality. However, meta-lenses have drawbacks including stray light issues, tradeoffs between numerical aperture and efficiency, and limitations with single-surface designs.
Freeform meta-surfaces, using more flexible pillar shapes, can address these issues and improve efficiency. The challenge in meta-lens design lies in simulating complex physics due to computation barriers and balancing optimization with manufacturability.
To overcome these challenges, artificial intelligence techniques such as artificial neural networks (ANNs) and Physics-Informed Neural Networks (PINNs) are being explored. ANNs can replace multi-step simulations with single-shot models, while PINNs leverage underlying physics for better generalization and extrapolation. PINNs are particularly promising for developing fast surrogate models and inverse models, potentially speeding up the design process significantly. Despite their current limitations in accuracy and network complexity, PINNs offer a promising approach for solving high-dimensional and nonlinear problems in meta-lens design.
Contact
 |
Esmaeil Nadimi Professor and Head of Unit SDU Applied AI and Data Science esmi@mmmi.sdu.dk |
 |
Vinay Chakravarthi Gogineni SDU Applied Ai and Data Science vigo@mmmi.sdu.dk |