Electro-optical Spiking Neurons in Silicon Photonics
Loading...
Authors
Tamura, Marcus
Date
Type
thesis
Language
eng
Keyword
Spiking Neurons , Silicon Photonics , Graphene , Nonlinear Dynamics , Neuromorphics
Alternative Title
Abstract
Artificial intelligence (AI) has shown it is capable of completing a large variety of tasks, such as prediction, control, classification, simplification, memory, and generation. These tasks are diverse and have applications in self-driving cars, medical diagnosis, physical system simplification, language tasks, and art creation. The versatility of AI is one of its greatest strengths. Currently, most machine learning tasks are performed using digital computers with graphic processing units (GPUs). These devices can perform these tasks, but their energy usage is atrocious, since these devices were not originally designed to perform AI computations. A neuromorphic system is hardware designed specifically to run AI tasks. In particular, spiking neuromorphic systems have great promise for incredible energy efficiency. Artificial spiking neurons more closely emulate the neurons in the human brain and primarily consume power when they fire a spike and are in use. The construction of spiking neurons is non-trivial due to the inherently non-linear nature of their time dynamics. There are different types of spiking neurons and their dynamical behavior involve sudden shifts (known as bifurcations) from steady state to oscillatory behavior. Simulations of these devices can be difficult due to convergence issues of studying a system that intentionally switches from stable to unstable behavior. This thesis focuses on how to create and simulate spiking systems using physics present in silicon photonic platforms. Silicon photonics is known for its scalable, high bandwidth, low-loss interconnects. Since computations in neural networks are distributed over a large number of interconnected neurons in large networks, silicon photonics is well-suited for neuromorphic devices. This thesis discusses the various types of bifurcations required for different classes of spiking neurons and gives an in depth look at simulations of three different silicon photonic devices capable of spiking. The devices are electrooptical and exhibit nonlinear optical and electrical physics. In particular, there are plenty of applications at cryogenic temperatures for spiking neural networks, so the cryo-compatibility of these effects are discussed.
Description
Citation
Publisher
License
Queen's University's Thesis/Dissertation Non-Exclusive License for Deposit to QSpace and Library and Archives Canada
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Attribution 3.0 United States
ProQuest PhD and Master's Theses International Dissemination Agreement
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Attribution 3.0 United States