Session Overview |
Wednesday, May 29 |
14:00 |
Lithography-Free Integrated Photonics for Reconfigurable Information Processing
Tianwei Wu, University of Pennsylvania * Liang Feng, University of Pennsylvania, United States of America We report an unprecedented lithography-free paradigm for an integrated photonic processor, which establishes remarkable field programmability and functionality from a global perspective, enabling accurate multiply-accumulate operations and in-situ training of optical neural networks. |
14:25 |
Inverse design and forward modelling in nanophotonics using deep-learning
* Junsuk Rho, Pohang University of Science and Technology (POSTECH), South Korea Recent introduction of deep learning into nanophotonics has enabled efficient inverse design process. Once the deep learning network is trained, it allows fast inverse design for multiple design tasks. In this talk, we show several inverse designing nanophotonic structures using deep learning. We firstly discuss inverse design methods that increase the degree of freedom of design possibilities. These attempts include designing arbitrary shapes of nanophotonic structures, that are not limited to pre-defined structures, and designing both types of materials and structural parameters simultaneously. In order to design arbitrary shapes of structures, cross-sectional design images are designed by generative model. Also, for simultaneous design of materials and structural parameters, we developed a novel objective function that combines regression and classification problems. After then, we also discuss optimizing nanophotonic structures using deep learning. We use reinforcement learning to optimize structure parameters. Using reinforcement learning, an agent learns parameter space of an environment through the exploration and exploitation of the reward. After learning, the agent can provide the optimized design parameters from its own experience. Several meta-devices including dielectric color filter, high efficiency hologram, perfect absorber, plasmonic structures, dielectric gratings and microwave antenna are designed using this method. We also intorduce recent work on real-time index sensing in microfluidic system by processing the optical information retrieved from metasurfaces using machine learning. |
14:50 |
Efficient inverse design techniques for topological photonic crystal waveguides
Eric Nussbaum, Queen's University, Canada Antonio Neill, Queen's University, Canada Nir Rotenberg, Queen's University, Canada * Stephen Hughes, Queen's University, Canada We discuss the key optical metrics of topological photonic crystal waveguides, and present fast inverse design techniques to significantly improve the various device figures-of-merit, including operation bandwidth, chiral coupling efficiently to quantum dot emitters, and reduced backscatter losses. |
15:15 |
Accelerated Optimization of Robust Nanophotonic Devices via Deep Learning
* Sawyer D. Campbell, The Pennsylvania State University, United States of America Ronald P. Jenkins, The Pennsylvania State University Pingjuan L. Werner, The Pennsylvania State University Douglas H. Werner, The Pennsylvania State University The potential for nanophotonic devices to disrupt existing and create new commercial applications has led to a surge in design and manufacturing research in recent years. Yet, new technologies must not only demonstrate performance advantages over legacy solutions, but also improved fabrication cost and reliability. Therefore, inverse-design techniques which optimize guaranteed performance in the presence of fabrication uncertainties are needed to maximize the yield achievable within a given process window. However, simulating a nominal structure and all potential perturbations caused by a variety of highly complex and coupled fabrication uncertainties results in a combinatorial explosion of solutions that ultimately makes direct optimization intractable. To overcome this, we exploit deep learning and demonstrate a neural network that accurately predicts the performance of a representative metasurface supercell in the presence of fabrication uncertainties. The trained neural network is subsequently paired with a modified multi-objective optimization procedure which enables one to study the tradeoffs between nominal performance and guaranteed performance. |