Knowledge Discovery in Nanophotonics Using Geometric Deep Learning

Abstract

Herein, a new approach for using the intelligence aspects of artificial intelligence for knowledge discovery rather than device optimization in electromagnetic (EM) nanostructures is presented. This approach uses training data obtained through full‐wave EM simulations of a series of nanostructures to train geometric deep learning algorithms to assess the range of feasible responses as well as the feasibility of a desired response from a class of EM nanostructures. To facilitate the knowledge discovery, this approach combines the dimensionality reduction technique with convex‐hull and one‐class support‐vector‐machine (SVM) algorithms to find the range of the feasible responses in the latent response space of the EM nanostructure. More importantly, the one‐class SVM algorithm can be trained to provide the degree of feasibility of a response from a given nanostructure. This important information can be used to modify the initial structure to an alternative one that can enable an initially unfeasible response. To show the applicability of this approach, it is applied to two important classes of binary metasurfaces (MSs), formed by an array of plasmonic nanostructures, and periodic MSs formed by an array of dielectric nanopillars. These theoretical and experimental results confirm the unique features of this approach for knowledge discovery in EM nanostructures.

The duality between particle methods and artificial neural networks

“…for dimensionality-reduction 20,21 . In the second group, the ML algorithm generates a data-driven model that is fed into the FP model 22,23 . These approaches have their benefits, but, because they link the FP model and the ML algorithm in series, they still require either a complete knowledge of the FPs or a large quantity of data.…”

Meta-optics for spatial optical analog computing

“…Recently, inverse design approaches based on local and global optimization techniques have attracted significant attention in enabling nontrivial high-performance meta-optic configurations targeted to a wide range of applications. Among several inverse design approaches, global step-by-step searching algorithms (such as genetic or particle swarm), adjoint-based topology optimization implementations, and neural network-assisted optimization approaches have proven to be compelling candidates to push forward highperformance nanophotonic devices [100][101][102][103][104][105][106][107][108][109][110][111][112][113][114][115][116]. Coupled to recent advances in nanofabrication technologies, physical modeling, and computational power, such single and multiobjective optimization approaches can benefit nextgeneration computational metaprocessors.…”

Tunable nanophotonics enabled by chalcogenide phase-change materials

“…Although, they have been widely used in nanophotonics design and optimization problems, they suffer from remarkable computation complexity and often result in local optimum (rather than global optimum) solutions; (ii) algorithms employing artificial neural networks (ANNs) to optimize topologies of nanophotonic structures. While being more reliable in providing the global optimum designs, such data-driven algorithms require a large amount of training instances to be practical for the real-world applications [249][250][251][252][253][254][255][256][257][258][259][260][261][262][263][264][265][266][267].…”