- 46 views
- 457 downloads
Phase Change Metamaterials for Fiber and Waveguide Integrated Compact Modulators with Built in Memory Functionality
-
- Author / Creator
- Cui, Yihao
-
Machine learning focuses on the development of algorithms and models that enable
a dynamic approach to making predictions or decisions on large data-sets which
cannot be readily described by analytical models. Artificial neural networks (ANN)
have become the main approach to providing advance functionalities such as pattern
recognition, and inference calculations with improved performance through experience and iterative training cycles. Inspired by the structural organization of a biological brain, ANNs consist of interconnected layers of artificial neurons, or nodes,
where input signals are weighted, pooled, and passed on to the next layer for analysis. Leveraging the computational hardware performance gains in application specific
integrated circuits (ASICs) such as graphics processing units (GPUs), ANNs have
proliferated in applications ranging from image recognition, natural language processing, autonomous vehicles, and medical diagnosis. However, the computational
requirements for floating-point operations of increasingly complex ANN models have
far exceeded performance trends predicted by Moore’s Law. Additionally, data movement between arithmetic and memory modules on metallic interconnects in ANN
hardware accelerators does not scale in terms of energy consumption, latency, and
bandwidth. The incorporation of chalcogenide phase-change metamaterials on photonic platforms such as optical fibers and photonic integrated circuits (PICs) offers a
compact, reversible, and non-volatile device platform to realize a largely distributed
and parallel optical computing architecture to accelerate training and inference calculations of ANNs. We demonstrate a cuboid-based germanium antimony telluride
(GST) metamaterial integrated on the tip of an optical fiber with a high intensity
contrast and switchable group delay dispersion between amorphous and crystalline
phases for long haul telecommunication signal transmission across network nodes.
Similarly, by eliciting an all-dielectric metamaterial resonance in a subwavelength
structured GST grating, an optically or electrically addressable synaptic weight was
realized on photonic silicon nitride waveguides. The resonator’s wavelength tunability enables symmetric 30% transmission modulation for both positive and negative
weighting values across amorphous and crystalline phases. Lastly, benefiting from
the high infrared transparency and the large refractive index contrast between the
two phases (∆n > 2.0 for certain alloy compositions), the thin film inclusion of phasechange materials on photonic waveguide circuits enables efficient and ultra-compact
phase-shifter elements with memory functionalities. Furthermore, effective medium
metamaterial design concepts allow for specified dispersion engineering required for
low insertion loss (< 0.3dB) and compact footprint (Lπ = 5 − 20µm) designs. Such
non-volatile phase-shifters can be embedded within a cascaded MZI mesh to implement a programmable zero static energy consumption matrix vector multiplication
layer for an interconnected neural network. -
- Subjects / Keywords
-
- Graduation date
- Fall 2023
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.