Scaling In Lithium Niobite: Synaptic Devices for Neuromorphic Computing
Author(s)
McCrone, Timothy Michael
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Due to the end of Dennard scaling, problems with the Von Neumann bottleneck, and the unique challenges posed by the volume of data generated on the internet neuromorphic computing has emerged an energy efficient computing method. While standard neuromorphic solutions use matrix math to churn through tables of numbers, in the last several years dedicated hardware has emerged that has optimized the overhead these computations incur in digital systems. Research into memristors has emerged as an alternative method to create neuromorphic systems capable of implementations of synaptic inference more closely matching the computations made in the brain than faux digital or analog systems. Instead of a series of amplifiers, transistors, and resistors necessary for digital and analog systems memristive systems can implement a biomimetic computational paradigm within several two terminal devices. In the variety of memristive systems LiNbO2 stands out due to its already proven large change in resistivity, low power programmability, dynamic and static control of resistivity ranges and temporal response, ability to create volatile, non-volatile and even mixed volatility responses and the ability to implement inductive analogues via ionic momentum. In hopes of unlocking the lowest power neuromorphic implementation to date, LiNbO2 has been investigated as a promising class of memristor devices. Prior investigations were limited to optical lithography scale devices yet compared very well to nanoscale devices from other technologies. This thesis explored methods to a) produce suitable LiNbO2 films allowing scaling to nanometer length scales and b) LiNbO2 devices were scaled past optical lithography lengths resulting in state-of-the-art breakthroughs in power efficiency.
Sponsor
Date
2024-11-20
Extent
Resource Type
Text
Resource Subtype
Dissertation