Abstract
Brain inspired computing enabled by memristors have gained prominence over the years due to its nano-scale footprint, reduced complexity for implementing synapses and neurons. Several demonstrations show two-dimensional (2D) materials as a promising platform for realization of robust and energy-efficient memristive synapses. Ideally, a synapse should exhibit low cycle-to-cycle (C-C) and device-to-device (D-D) variability along with high maximum /minimum conductance (Gmax/Gmin) ratio, linearity and symmetry in weight update for obtaining high learning accuracy in neural networks (NNs). However, the demonstration of neuromorphic circuits using conventional materials systems has been limited by high C-C and D-D variability and non-linearity in the weight updates. In this study, we have realized robust memristive synapses using 2D molybdenum disulfide (MoS2) to address the concerns like high variability and non-linear weight update and asymmetry. We have utilized engineering techniques like electrode and stack engineering to realize ultra-low variability and linear weight update in MoS2 synapses. The ultra-low C-C and D-D variability in SET voltage, RESET power and weight update is demonstrated in Au/MoS2/Ti/Au synapses. Further, these synapses were integrated with MoS2 leaky-integrate and fire (LIF) neurons to realize AND, OR and NOT logic gates proving the viability of these synapses for in-memory computing. However, these MoS2 synapses suffer from low Gmax/Gmin ratio. We have employed stack engineering to increase Gmax/Gmin ratio while preserving low variability. In that regard, the active medium is modified to a heterogenous stack of MoS2/SiOx with Ti/Au bottom and top electrodes. We observe an increase in the Gmax/Gmin ratio from 2 to ~10. Further, electrode engineering is used to realize graphene/MoS2/SiOx/Ni to obtain linear weight update with identical pulses essential for online training of NNs. This work substantiates the necessity of engineering techniques to implement essential synaptic characteristics like ultra-low variability and linear and symmetric weight update.
Notes
If this is your thesis or dissertation, and want to learn how to access it or for more information about readership statistics, contact us at STARS@ucf.edu
Graduation Date
2022
Semester
Summer
Advisor
Roy, Tania
Degree
Doctor of Philosophy (Ph.D.)
College
College of Engineering and Computer Science
Department
Electrical and Computer Engineering
Degree Program
Electrical Engineering
Identifier
CFE0009210; DP0026814
URL
https://purls.library.ucf.edu/go/DP0026814
Language
English
Release Date
August 2022
Length of Campus-only Access
None
Access Status
Doctoral Dissertation (Open Access)
STARS Citation
Krishnaprasad Sharada, Adithi Pandrahal, "MoS2 based Memristive Synapses for Neuromorphic Computing" (2022). Electronic Theses and Dissertations, 2020-2023. 1239.
https://stars.library.ucf.edu/etd2020/1239