Gradient-based optimization of core-shell particles with discrete materials for directional scattering.

Area of Science:

• Nanophotonics
• Computational Physics
• Machine Learning

Background:

• Designing nanophotonic structures with discrete parameters (e.g., real materials) is computationally expensive, often requiring global optimization.
• Existing methods struggle with the discrete nature of material selection and geometric parameters.

Purpose of the Study:

• To develop a computationally efficient framework for designing nanophotonic structures with discrete parameters.
• To enable direct gradient-based optimization for inverse problems involving discrete variables.

Main Methods:

• Leveraging generative deep learning to map discrete parameter sets into a continuous latent space.
• Employing a neural network as a differentiable surrogate model for non-differentiable physics evaluation.
• Optimizing directional scattering properties of core-shell nanoparticles using realistic materials.

Main Results:

• Achieved direct gradient-based optimization by mapping discrete parameters to a continuous latent space.
• Successfully optimized core-shell nanoparticle geometries for enhanced forward scattering and reduced backscattering.
• Demonstrated significant improvements in computational efficiency and performance compared to traditional global optimization techniques.

Conclusions:

• The proposed deep learning framework offers a more efficient and effective approach to designing nanophotonic structures.
• This methodology is broadly applicable to various inverse problems constrained by discrete variables.
• The findings pave the way for accelerated discovery and design in materials science and photonics.