Dissertations, Theses, and Capstone Projects
Date of Degree
2-2026
Document Type
Doctoral Dissertation
Degree Name
Doctor of Philosophy
Program
Physics
Advisor
Andrea Alù
Committee Members
Mohammad-Ali Miri
Matthew Sfeir
Karl G. Sandeman
Vinod Menon
Subject Categories
Optics | Physics
Keywords
Metamaterial, Multifunctional Metasurface, Photonic Inverse Design, Photonic Computing, Photonic Ising Machine, Analog Image Processing
Abstract
Rapid advancements in nanotechnology, coupled with the increasing demand for unconventional computing systems that are faster, more compact, and more energy-efficient than traditional classical computers, have sparked a surge of research interest in developing innovative light-based computing nanotechnologies. These technologies aim to leverage light’s inherent parallelism, analog mode of operation, and ultra-fast computing speeds enabled by light-speed propagation. In my dissertation, I present some of our research findings focused on light-based computing paradigms utilizing artificial photonic nanostructures known as metamaterials. These can be directly integrated into ultra-compact imaging systems for full-analog wave-based computations or into hybrid computing schemes that incorporate classical computers as optical computing hardware accelerators. This integration aims to enhance state-of-the-art digital processing for large-scale, real-world problems by executing components of computations as light-based operations, such as matrix multiplication, Fourier transforms, differentiation, and integration, which can be made passive, parallel, and instantaneous by using light.
Metamaterials—artificially engineered materials at sub-wavelength scales—exhibit, for our purposes, unconventional electromagnetic wave properties that enable efficient manipulation of light within highly miniaturized volumes. By manipulating the incident light waves, photonic metamaterials can produce an output electromagnetic wave profile proportional to a mathematical operation performed on the incident light wave, thereby functioning as analog optical computers. Nonlocal metasurfaces, which are two-dimensional metamaterials with angle-dependent electromagnetic responses, have recently garnered significant attention for their ability to implement arbitrary linear mathematical operations directly on incident electromagnetic wavefronts, eliminating the need for Fourier-transforming optical elements. This is achieved by engineering the angular-dependent transfer function of nonlocal metasurfaces to align with the desired operation kernel in Fourier space. Following this metamaterial design principle, operations such as differentiation, integration, and convolution have been optically demonstrated in ultra-compact, highly integrable, and easy-to-fabricate analog optical devices. While the designs proposed to date can implement a single specific mathematical operation, the potential to execute multiple operations with a single device has largely remained unexplored. As part of our research, I present a new multi-functional metasurface design strategy and demonstrate that optical analog computing for spatial differentiation and integration—useful for image processing operations like edge detection and blurring—can be achieved using a single multi-operation metasurface. This principle leverages the combination of nonlocal metasurface transmission responses to orthogonally polarized s-p plane waves, both of which arrive as components of the scattered electromagnetic fields that form the image. This results in the creation of two distinct effective isotropic Fourier filters for processing images based on the incident orthogonal x- or y-polarizations of light. The metasurface design is based on a silicon-on-glass photonic crystal slab featuring an amorphous silicon layer with etched holes placed on a glass substrate, while ensuring careful geometric x-y symmetry breaking of the metasurface unit cell. I present numerical results of multi-operation processing using the optimized metasurface designs on 1-D signals and a 2-D aperture image. Our results demonstrate, for the first time, that a single metasurface design can perform both high-pass and low-pass spatial Fourier filtering operations in both one and two dimensions with a large spatial bandwidth, thereby enabling the isotropic processing of two-dimensional images for high-quality edge detection and blurring based on this effect. Our research holds promise for profound applications in cutting-edge, highly miniaturized imaging systems such as LiDAR, automated medical microscopy, satellite imaging systems, computer vision, and other critical areas where high-throughput, rapid image processing is essential.
I also discuss the use of metamaterials as on-chip silicon photonics computing hardware accelerators, which can work synergistically with classical computers to realize photonic Ising machine platforms. Photonic Ising machines are unconventional computing mechanisms that encode Ising spin states and Ising Hamiltonians in various observable properties of light, such as phase, polarization, and intensity. Our Ising machine concept specifically belongs to a class of next-generation hybrid optical Ising machines that benefit from existing computing technology and advanced iterative algorithms, like Monte Carlo techniques, running on digital computers to reach exact Ising ground states or approach them for very large problems, and to solve several combinatorial optimization problems that are NP, NP-complete, or NP-hard and can be mapped as Ising problems. I describe our realization of a highly compact on-chip silicon photonics optical Ising machine platform using photonic inverse design to optimize the required metamaterial geometry and numerically demonstrate the operational principles for minimizing different spin graphs. Our Ising machine setup is based on a system of input and output optical waveguides connected to a metamaterial region with an engineered transmission matrix that encodes the Ising matrix of the spin graph to be emulated. A computer running a Monte Carlo algorithm, such as Metropolis or Simulated Annealing, iteratively updates the input binary phase combination (0 or π radians) of the light in each input waveguide - all of which are excited in the Ising machine platform simultaneously with equal amplitudes - such that the photodetectors in the output waveguides provide a net output intensity reading from all output waveguides, which the Ising machine uses to achieve an input phase combination that maximizes the net output intensity to a global maximum in the final step. This final input binary phase combination corresponds to the emulated spin ground state and represents the solution of the Ising machine.
Recommended Citation
Biswas, Aneek, "Metamaterials that Perform Computation Using Light" (2026). CUNY Academic Works.
https://academicworks.cuny.edu/gc_etds/6530
