Researchers from Kings College London, UK, have indicated that machine learning techniques can be used to enhance metasurfaces for nonlinear optics and optomechanics.
In his new research paper, head of Photonics & Nanotechnology Group at King’s College London, Anatoly Zayats suggests that metasurfaces can be enhanced using machine learning techniques. The end goal of this enhancement is to optimise metasurfaces for nonlinear optics and optomechanics.
What are metasurfaces?
Metasurfaces are platforms that are often used to manipulate the scattering, colour, phase, or intensity of light that can be used for light emission, detection, modulation, control and/or amplification at the nanoscale.
Recently, metasurfaces have been a subject of intense study as their optical properties can be adapted to a diverse set of applications, including superlenses, tuneable images, and holograms.
According to a paper published in Advanced Photonics: ‘Currently, one of the typical approaches for designing a metasurface is to optimise one or two variables among a vast number of fixed parameters, such as various materials’ properties and coupling effects, as well as the geometrical parameters.
‘Ideally, this would require multidimensional space optimisation through direct numerical simulations. Recently, an alternative, popular approach allows for reducing the computational cost significantly based on a deep-learning-assisted method.’
Optimising metasurfaces for optics
The discovery made by Zayats has promising possibilities for the development of a wide range of photonic devices and applications including those involved in optical sensing, optoacoustic vibrations, and narrowband filtering.
According to Zayats, this work marks an exciting advancement in nanophotonics. “Optimisation of metasurfaces and metamaterials for particular applications is an important and time-consuming problem.
“With traditional approaches, only few parameters can be optimised, so that the resulting performance is better than for some other designs but not necessarily the best. Using machine learning, one can search for the best design and cover the space of parameters not possible with traditional approaches.” Commented Zayats.
