Metalens Array for Optical Computation

ABSTRACT
Optical metasurfaces are two-dimensional materials that are artificially engineered to modulate light by sub-wavelength nanostructures. Their ability to precisely modulate wavefronts enables diverse applications in various photonic systems. Specifically, metasurfaces have shown to be effective as a platform for optical computing with successful applications such as solving differential equations or building neural networks. This thesis presents the design and optimization of a metasurface-based lens array for optical computing, specifically targeting complex matrix-vector multiplication. In chapter 2, we lay out the fundamentals, including the theory for geometric phase metasurfaces and a few methods for numerical simulations and focal pattern approximation. In chapter 3, we present a geometric phase metasurfaces device that implements an 8× 8 complex-to-complex DFT. We fabricate and experimentally benchmark the device accuracy, achieving a root-mean-square error (RMSE) of 4.73% with error correction enabled. Key innovations include an interferometric field reconstruction scheme for phase retrieval and an input control mechanism with a single spatial light modulator. In chapter 4, we discuss a few optimization strategies, such as truncated singular value decomposition and DFT basis masks. This further reduces the RMSE to 3.67%,enhancing the noise robustness. Additionally, input and output digitalization techniques are explored to mitigate noise. These advancements demonstrate the potential of metasurface lens arrays as compact, energy-efficient platforms for optical computing