Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.
Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.
Author ORCID Identifier
https://orcid.org/0000-0002-4043-6701
AccessType
Campus-Only Access for Five (5) Years
Document Type
dissertation
Degree Name
Doctor of Philosophy (PhD)
Degree Program
Electrical and Computer Engineering
Year Degree Awarded
2021
Month Degree Awarded
May
First Advisor
Jianhua Joshua Yang
Subject Categories
Electronic Devices and Semiconductor Manufacturing
Abstract
Neuromorphic computing describes the use of electrical circuits to mimic biological architecture presented in nervous systems. Crossbar-structured artificial neural network is a promising emulator of biological neural network due to its high density, massive connectivity and efficient implementation of weighted sums.
Depending on the types of signals the artificial neural network is processing, it can be classified into non-spiking neural network and spiking neural network. In a typical working scheme of a crossbar non-spiking neural network, the electrical conductance of the cross-point device is used to represent the weight so that the total output current at the end of each column is a weighted sum of the input voltages to the rows. As a result, programming the conductance (weight) is critical for neural network training. Memristor is a two-terminal passive circuit element that exhibits a tunable conductance driven by input voltages. Due to its great performance in multilevel, speed, endurance, retention, stackability and scalability, the memristor has become one of the most promising candidates for future artificial synapse.
By analyzing and understanding the memristive behavior under electrical stimulations, we proposed an algorithm that can tune the device conductance precisely and enhance the stability of resistance levels by reducing the random telegraph noise. With the help of the algorithm, we have successfully demonstrated 2048 discrete conductance levels in memristors, which is the highest among all types of non-volatile resistive switching devices. We have also shown that the device can maintain its conductance within a very narrow fluctuation range for a long time after stabilization.
We also invented a capacitive switching device that processes spiking signals naturally with the leaky integrate-and-fire function similar to a biological spiking neuron. We further demonstrated spatial summation and signal propagation capability by integrating such neuron with a transistor to form a neuro-transistor. Finally, we built the first capacitive spiking neural networks, with which we have successfully demonstrated pattern classification with capacitive version of dot-product.
DOI
https://doi.org/10.7275/22100134.0
Recommended Citation
Rao, Mingyi, "Memristive Devices and Integrations for Neuromorphic Computing" (2021). Doctoral Dissertations. 2214.
https://doi.org/10.7275/22100134.0
https://scholarworks.umass.edu/dissertations_2/2214
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.