Metasurface for "vision" improvement

Abstract

Metasurface, a miniaturized device with a meticulously designed permittivity distribution which accommodates nano-fabrication technique, has many novel optical functionalities. This thesis mainly presents the design and application of metasurface devices to improve "vision" in three aspects, namely, signal "vision", holographic "vision", and biomedical "vision". The design techniques spread from forward calculation, neural network, to adjoint optimization. These methods are dedicated to finalize the exact metasurface patterning for a desired far-field radiation, which, if properly defined, can be beam profiles in signal "vision", 2D intensity distribution in holographic "vision", or depth of focus distribution in biomedical "vision".

For signal "vision", an enhanced signal transmission is achieved through a metasurface to convert a plane wave into multiple Laguerre-Gaussian (LG) modes, thus expanding the communication channels by one more dimension. A 2D array of Au nano-antenna was fabricated on a silica chip. When the chip was illuminated by a plane wave, multiple orthogonal LG modes will be generated and demultiplexed into different angles. Each nano-antenna has a width of 200 nm and a length of 200-400 nm, which can achieve a complex modulation with 0-1 amplitude range and 0-2$\pi$ phase range. Error analysis suggests that this metasurface is robust for LG mode transmission over a 400-nm broadband range.

For holographic "vision", the arbitrary holographic display is accomplished through a metasurface predicted by a pre-trained neural network in a lensless projection modality. The design process through neural network is about 400 times faster than theoretical computation. The metasurface in question is capable of complex modulation. Here I developed a convolutional neural network with dual-output, featuring the complex pattern in terms of the amplitude and phase branch. Then the complex pattern, termed as computer generated hologram (CGH) was fabricated into a metasurface and positioned in an experimental setup to project a holographic image. The averaged training loss was 4.43% while the testing loss was 4.57%, suggesting a good generalization.

For biomedical "vision", an augmented biomedical imaging can be realized by a tailored fiber facet, which is inversely designed through adjoint optimization to deliver an extended depth of focus for blood vessel imaging in optical coherence tomography. Two modalities has been proposed. One is chip-modality: a phase mask has been optimized to deliver an extended focus with a center intensity 40% higher than its Bessel beam counterparts and a lowered side lobe. The other is fiber -modality: a grating structure has been calculated on top of a fiber facet to an extended focus in the forward direction. In the future, I will calculate the grating pattern for a fiber with side view and integrate it to an imaging modality.

All in all, metasurface design techniques have been demonstrated to improve "vision" in signal, holographic, and biomedical aspects. The design techniques spread from forward calculation for a complex map of signal demultiplexing, neural neural network prediction for an accelerated holographic display, and all the way to the inverse design achieved by adjoint optimization to finalize the design in one go to meet the desired requirement.