Deep Neural Network Inverse Design of Integrated Photonic Power Splitters.

Area of Science:

• Nanophotonics and Metamaterials
• Computational Physics and Engineering

Background:

• Predicting the physical response of engineered materials is crucial for scientific and engineering advancements.
• Nanophotonic devices offer unique optical properties but their design can be complex.

Purpose of the Study:

• To employ deep learning for predicting and designing nanophotonic devices.
• To enable both forward prediction and inverse design of optical responses for engineered nanostructures.

Main Methods:

• Utilized a Deep Neural Network (DNN) for optical response prediction.
• Developed an approach for inverse design to achieve targeted optical functionalities.
• Trained the DNN to optimize parameters for power splitters, minimizing reflection and maximizing transmission.

Main Results:

• Successfully predicted the optical response of nanophotonic devices.
• Enabled inverse design of compact silicon-on-insulator (SOI)-based 1x2 power splitters (2.6 × 2.6 μm²).
• Achieved high transmission efficiency (>90%) and low reflection (< -20 dB) while meeting splitting ratio specifications.

Conclusions:

• Deep learning accelerates the design of complex nanophotonic devices.
• This AI-driven methodology facilitates rapid development of integrated photonic components.
• The approach shows potential for designing various nanophotonic devices with tailored optical properties.