1T-TaS2 for Nonlinear Applications in Optical Neural Networks

Date
2020-04-17
Journal Title
Journal ISSN
Volume Title
Publisher
Description
Abstract

Artificial neural networks (ANNs) are a set of algorithms which can realize many different functions, such as patterns recognition, classification and prediction. ANNs has become a well-worthy field of study in multiple industries over the past years. Much effort has been contributed to this area and it turns back with significant values in terms of commerce as well as research. However, nowadays most of the ANNs are completed based on electronic hardware, which have some natural limitations such as computation speed and energy cost.

Optical neural networks (ONNs) use photons, instead of electrons to transport and compute information. ONNs can easily realize linear computation just by the interference effect of lights and directly use optical nonlinear behavior to realize the nonlinear computation. Besides, ONNs own faster speed and lower energy cost compared with electronic ANNs. However, most of the ONNs are still in the developing stage and the nonlinear activation function has not been applied to real word application. Although there exist some optical nonlinear phenomena, currently the nonliearities are either too small or too slow. Besides, it is very hard to change the optical properties at room temperature with small stimulus, which limits the application of nonlinear applications. Since most of the existing devices requests strict restrictions, researchers keep exploring new materials for better efficiency and lower cost.

Here, we propose 1T-TaS2, a strongly correlated material for nonlinear behavior. The optical properties of this material are very sensitive to the external stimulus, and the response speed of this material is very fast, reaching to nano seconds, which make these materials promising in nonlinear applications in ONNs. 1T-TaS2 is one kind of the Transition Metal Dichalcogenides (TMDs). TMDs are one of the strongly correlated materials involving abundant phases, such as distorted phase, metallic phase, insulating phase, charge density wave (CDW) phase, superconducting phase and topological phase. Due to these different phases, TMDs exhibit possibilities for optical tunabilities. More attractively, 1T-TaS2 shows nearly commensurate charge density wave (NCCDWs) at room temperature, which means that its nonequilibrium states can be tuned under stimulus in a relatively easy way.

In this thesis, we show that the refractive index of 1T-TaS2 can be tuned under white light excitation with intensity from 2.5 mW/cm2 to 250 mW/cm2. We also find that this tunability comes from the out-of-plane, not the in-plane, stacking difference of the material. By implementing this tunability, we design three nanophotonic nonlinear devices: reflection-tunable, transmission-tunable and angle-tunable nonlinear device. By changing the parameters of the device structures dynamically, we collect, analyze and summarize the simulation results to finalize the implementation of the most efficient devices that are meeting the expectations. And we find that by giving a specific structure, the optical limiter can offer more than 10% reflection or transmission difference under different incident light intensity. The grating device could have a significant difference on the diffraction angles with a value of 6 degree under different incident light intensities.

Description
Degree
Master of Science
Type
Thesis
Keywords
Optical neural networks, Nonlinear activation function, 1T-TaS2, Nanophotonic device
Citation

Wang, Yuning. "1T-TaS2 for Nonlinear Applications in Optical Neural Networks." (2020) Master’s Thesis, Rice University. https://hdl.handle.net/1911/108301.

Has part(s)
Forms part of
Published Version
Rights
Copyright is held by the author, unless otherwise indicated. Permission to reuse, publish, or reproduce the work beyond the bounds of fair use or other exemptions to copyright law must be obtained from the copyright holder.
Link to license
Citable link to this page