Inverse Design of Unitary Transmission Matrices in Silicon Photonic Coupled Waveguide Arrays Using a Neural Adjoint...

Area of Science:

• Integrated photonics
• Quantum computing
• Optical neural networks

Background:

• Low-loss reconfigurable integrated optical devices are crucial for advancing photonic signal processing, analogue quantum computing, and optical neural networks.
• Designing these devices requires computationally intensive inverse design protocols to determine optimal geometries for specific optical outputs.

Purpose of the Study:

• To introduce a novel platform for implementing unitary matrix operations using digital patterning of coupled waveguide arrays.
• To present a robust and versatile inverse design protocol for predicting device geometries.

Main Methods:

• Utilizing high-speed neural network surrogate-based gradient optimization.
• Predicting refractive index perturbations based on switching the chalcogenide phase change material, antimony triselinide (Sb2Se3).
• Applying data set augmentation and random noise supplementation for neural network optimization.

Main Results:

• Demonstrated control of both amplitude and phase for each transmission matrix element in a 3x3 silicon waveguide array.
• Achieved an average fidelity of 0.94 for unitary matrix targets.
• Showcased programmable unitary operators with a reduced footprint compared to conventional technologies.

Conclusions:

• Coupled waveguide arrays with digital perturbation patterns offer a new pathway for creating programmable unitary operators.
• This approach is suitable for developing compact Hamiltonians for quantum simulators.
• The developed inverse design protocol enhances the design process for complex integrated optical devices.