When
Where
Title: The Simulation and Inverse Design of Metasurface-Based Angle Sensitive Pixels
Abstract:
Light field imaging, also known as plenoptic imaging, has emerged as a powerful and versatile technique, enabling the simultaneous capture of both the intensity and precise angle of incoming light. The fundamental building block of light field imaging is called the angle-sensitive pixel (ASP), which functions as a single unit on the image sensor. This remarkable capability presents light field imaging with a diverse array of post-imaging processing opportunities, such as on-demand refocusing, viewpoint adjustment, and reconstruction of depth maps. Nonetheless, the current mainstream light field imaging methods predominantly rely on microlens arrays, stacked transmission gratings, or holograms, which exhibit several inherent limitations. Such limitations encompass constraints in spatial and angular resolution, restricted angular range, and challenges in the fabrication process, thereby hindering their widespread applications.
Metasurfaces, characterized as emerging two-dimensional (2D) artificial photonic devices, exhibit an exceptional capacity to finely manipulate the phase, amplitude, and polarization of light at subwavelength scales, thereby presenting an opportunity to tackle prevailing challenges in current ASP design. Leveraging metasurfaces in ASP design allows for maintaining substantial angular detection range as well as high spatial and angular resolution within a compact footprint. The conventional metasurface design process confronts two significant challenges: the time-intensive nature of acquiring the electromagnetic response of metasurfaces through 3D full wave solvers and the computationally demanding task of optimizing metasurfaces characterized by high degree of freedom (DoF) design parameters.
In this dissertation, we introduce a pioneering methodology for rapid inverse design of angle-sensitive pixels based on the discrete dipole approximation (DDA) in conjunction with adjoint method. To this end, We have developed a frequency-domain solver for Maxwell's equations by integrating the discrete dipole approximation with analytical solutions for light scattered by a dipole near a flat surface. This solver provides accurate results at least six times faster than the finite difference time domain (FDTD) method. Moreover, the incorporation of the adjoint method into our DDA approach enables the analytical computation of gradients for the objective function using only one forward simulation and one backward simulation, regardless of the number of designing parameters. This capability facilitates rapid design of high DoFs or large-scale metasurfaces, significantly reducing the number of iterations and simulation time in photonic device optimization, which goes far beyond the scope of ASPs.
Moreover, we investigate the use of dipole approximations in light scattering calculations and optical forces acting on metallic nanoparticles. Through a comprehensive comparison of different dipole models to Mie scattering and FDTD simulations, we demonstrate that using the full volume of metallic nanoparticles, instead of an effective volume based on skin depth, yields more accurate results, considering that the complex relative permittivity of the metal already accounts for its skin depth.