Inverse design of nanophotonic devices based on numerical optimization methods

Abstract

Modern nanophotonics has introduced innovative technologies and components for controlling light. Notably, plasmonics, which utilizes metals to confine light into spaces much smaller than its wavelength, and metasurfaces, possessing optical properties not found in nature, are representative examples. In this way, the field of nanophotonics has advanced by increasing functional complexity and miniaturizing component sizes. However, designing and realizing increasingly complex and sophisticated nanophotonic components requires more than simple optical structure design based on intuition, as has been done for decades. Recently, numerical calculation methods such as the adjoint method and deep learning for device modeling have emerged. These approaches aim to inverse design free-form devices that satisfy desired performance metrics, going beyond the limitations of intuitive design based on decades of experience. In this thesis, state-of-the-art technologies for inverse design of free-form nanophotonic components are reviewed. The latest papers are examined in the order of classical optimization methods, adjoint methods, and deep learning methods to understand how these design methods are utilized. Additionally, the thesis explores what additional design methods are needed to ensure that the designed devices meet the minimum feature size limits of actual microfabrication processes. Subsequently, utilizing these advanced optimization techniques, the thesis demonstrates the inverse design of unprecedented high-performance nanophotonic components in three main topics: light trapping in a plasmonic waveguide, complete $2 \pi$ phase modulation, and single-gate electrically tunable beam switching. Firstly, the performance of light trapping in a linear plasmonic waveguide is enhanced to a level comparable to the latest resonator-based light trapping performance by designing free-form waveguide structures using classical optimization methods. The general relationship between light trapping performance and waveguide length and metal loss is derived. Secondly, an active metasurface based on silicon-graphene is designed to exhibit the avoided crossing phenomenon between qBIC (quasi-bound states in the continuum) mode and graphene plasmon mode using classical optimization methods. The designed metasurface showed high reflectance and phase modulation performance up to $3 \pi$, depending on the variable Fermi level. Lastly, using the adjoint method, a silicon-graphene-based active metasurface is optimized to refract light in a specific direction depending on the variable Fermi level. The designed metasurface exhibited near 100% directivity and high diffraction efficiency in 2- and 3-level beam switching. Inspired by the modulation limitation discovered in the design process of the 4-level switching metasurface, the necessary conditions for designing multi-level switching devices are derived.