Photonics and AI - 2Virtual room: Photons Canada - 4
|Wednesday, May 27|
AI-2-27-1 / Photonic inverse design for robust next generation components
* James Pond, Lumerical Inc., Canada
Jens Niegemann, Lumerical Inc.
Milad Mahpeykar, Lumerical Inc.
Taylor Robertson, Lumerical Inc.
Adam Reid, Lumerical Inc.
Dylan McGuire, Lumerical Inc.
Lukas Chrostowski, University of British Columbia
Mustafa Hammood, University of British Columbia
We show how photonic inverse design can be used to create high performance yet robust and manufacturable designs, including novel components with footprints that are orders of magnitude smaller than what is currently available.
AI-2-27-2 / Machine learning pattern recognition in integrated silicon photonics design
* Dan-Xia Xu, National Research Council Canada, Canada
The optimization of complex high-dimensional photonic structures is often limited by computational resources. Current techniques based on global optimization algorithms or shape/topology inverse design treat design variables as entirely independent. However, there is often correlation between the input variables and patterns in the design outcomes. We review our strategy of using machine learning pattern recognition for building the performance map of a high-dimensional design space, thereby quickly guiding the search to a small region of interest and significantly improving the computational efficiency. This strategy is also beneficial in overcoming the limitations of local optima in all gradient-decent based optimization methods including adjoint inverse design.
AI-2-27-3 / Transport vs. Deep Neural Networks in OAM Underwater Communications
* Patrick Neary, Utah State University, United States
Nicholas Flann, Utah State University
Abbie Watnik, Naval Research Laboratory
Peter Judd, Naval Research Laboratory
Ryan Lindle, Naval Research Laboratory
Signal attenuation, resulting in low signal to noise ratio (SNR), in underwater optical communications (UWOC) is a problem that degrades classification performance. We develop and contrast several novel ways to create machine learning (ML) and optimal transport-based attenuation models and insert these models in a convolutional neural network (CNN) classification training pipeline. We show that including these ML-based attenuation models in the CNN classifier training, significantly improves classification performance.
AI-2-27-4 / Approaching the fundamental limits of photonic control via inverse methods
Alejandro Rodriguez, Princeton University, United States
* Sean Molesky, Princeton University, United States
We present a framework based on Lagrange duality and scattering constraints to set physical bounds on any single material electromagnetic design problem that can be framed as a net power emission, scattering or absorption process. We demonstrate that the aforementioned developments can be used to derive limits on light extraction and light–matter interactions (e.g., achievable scattering cross sections, Purcell enhancement, and optical forces) in nanostructured media. The presented relevant scattering constraints and optimization techniques have immediate applications for guiding, understanding, and improving inverse methods.