A new deep-learning approach based on dimensionality reduction techniques for the design and knowledge discovery in nanophotonic structures will be presented. It is shown that reducing the dimensionality of the response and design spaces in a class of nanophotonic structures can provide new insight into the physics of light-matter interaction in such nanostructures while facilitating their inverse design. These unique features are achieved while considerably reducing the computation complexity through dimensionality reduction. It is also shown that this approach can enable an evolutionary design method in which the initial design can be evolved intelligently into an alternative with favorable specification like less complexity, more robustness, less power consumption, etc. In addition to providing the details about the fundamental aspects of the latent learning approach, its application to design of reconfigurable metasurfaces will be demonstrated.
We present a new approach for design of novel loss functions and introduce an optimal similarity-metric design for machine-learning-based design and knowledge discovery in nanophotonics. Machine-learning algorithms estimate the input-output relation in a photonic nanostructure by minimizing a loss function. We show that careful selection (from the available loss functions) or design of novel loss functions that are optimized for specific tasks can considerably improve the performance of machine-learning algorithms for design and knowledge discovery in photonic nanostructures. We also discuss the limitations and inefficacies of conventional loss functions that are currently being used for machine learning algorithms.
We present a new approach based on manifold learning for breaking the geometrical complexity of the photonic nanostructures during solving the inverse design problem. By encoding the high-dimensional spectral responses of a class of nanostructures into the latent space, we provide intuitive information about the underlying physics of these structures. We discuss the relations between the non-Euclidean distances in the latent space and changes in the optical responses and relate the movements in the latent space to the modifications of the optical responses for a class of nanostructures. Finally, we provide a new approach to use the insight about the role of design parameters to design nanostructures with minimal design complexity for a given functionality.
This Conference Presentation, "Sample-efficient machine-learning method for designing photonic nanostructures," was recorded at Photonics West 2020 held in San Francisco, California, United States.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.