Design of optical meta-structures with applications to beam engineering using deep learning.

Area of Science:

• Nanophotonics and optical engineering
• Computational electromagnetics
• Machine learning applications in photonics

Background:

• Designing on-chip nanophotonic components for light manipulation involves high-dimensional design spaces.
• Conventional optimization methods struggle to find global optima for complex meta-optical structures.
• On-chip photonic beam engineering is crucial for advanced spectroscopic analysis.

Purpose of the Study:

• To explore a Machine Learning (ML)-based method for the inverse design of meta-optical structures.
• To develop a data-driven approach for modeling grating meta-structures for photonic beam engineering.
• To enable rapid estimation of meta-surface design parameters for desired electromagnetic field outcomes.

Main Methods:

• Utilized a feedforward deep neural network (DNN) and convolutional neural network (CNN) architecture for inverse modeling.
• Trained neural networks on a dataset relating meta-surface parameters to electromagnetic field outcomes.
• Employed a forward model in conjunction with the trained inverse model for validation.

Main Results:

• Achieved a high correlation coefficient (up to 0.996) in predicting diffraction profiles.
• Successfully predicted optimal design parameters (period, height, scatterer size) for meta-surfaces.
• Demonstrated the capability of the ML model to rapidly estimate design parameters.

Conclusions:

• The developed ML-based inverse design method offers a significant speed advantage over conventional optimization techniques.
• This approach facilitates efficient photonic beam engineering for lab-on-chip applications.
• The data-driven model provides a powerful tool for accelerating the design cycle of nanophotonic devices.