Data enhanced iterative few-sample learning algorithm-based inverse design of 2D programmable chiral metamaterials.

Area of Science:

• Metamaterials Science
• Computational Photonics
• Nanophotonics

Background:

• Metamaterials offer unique optical properties through engineered structures.
• Inverse design of chiral metamaterials is crucial for advanced optical devices but computationally intensive.
• Existing methods often require large datasets and significant computational resources.

Purpose of the Study:

• To develop an accurate and efficient algorithm for the inverse design of multi-shaped 2D chiral metamaterials.
• To reduce the computational cost and data dependency in metamaterial inverse design.
• To enhance the interpretability and generalization ability of inverse design algorithms.

Main Methods:

• A data enhanced iterative few-sample (DEIFS) algorithm combining data enhancement and iterative prediction.
• Utilizing rigorous coupled wave analysis (RCWA) for training data generation.
• Employing a forward prediction network to generate pseudo-data and reinforce datasets.
• Iterative refinement of geometric parameters based on circular dichroism (CD) spectra.

Main Results:

• The DEIFS algorithm achieves accurate inverse design with significantly reduced dataset sizes (1-2 orders of magnitude less than previous deep learning methods).
• Demonstrated fast inverse design for multiple 2D chiral metamaterial shapes, enabling diverse light manipulation.
• Improved interpretability of the inverse design process by analyzing the impact of geometric parameters on chiroptical response.

Conclusions:

• DEIFS offers a faster, simpler, and more data-efficient alternative to traditional iterative optimization for metamaterial inverse design.
• The algorithm shows great potential for applications in optical coding and information processing.
• This study thoroughly characterizes the flexibility, interpretability, and generalization of DEIFS for chiroptical effects in metamaterials.