E-Thesis 301 views 86 downloads
Molecular dynamics simulations of nanoclusters in neuromorphic systems / WENKAI WU
Swansea University Author: WENKAI WU
DOI (Published version): 10.23889/SUthesis.63613
Abstract
Neuromorphic computing is a new computing paradigm that deals with computing tasks using inter-connected artificial neurons inspired by the natural neurons in the human brain. This computational architecture is more efficient in performing many complex tasks such a pattern recognition and has promis...
| Published: |
Swansea, Wales, UK
2023
|
|---|---|
| Institution: | Swansea University |
| Degree level: | Doctoral |
| Degree name: | Ph.D |
| Supervisor: | Palmer, Richard E. and Pavloudis, Theodoros. |
| URI: | https://cronfa.swan.ac.uk/Record/cronfa63613 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Abstract: |
Neuromorphic computing is a new computing paradigm that deals with computing tasks using inter-connected artificial neurons inspired by the natural neurons in the human brain. This computational architecture is more efficient in performing many complex tasks such a pattern recognition and has promise at overcoming some of the limitations of conventional computers. Among the emerging types of artificial neurons, a cluster-based neuromorphic device is a promising system with good costefficiency because of a simple fabrication process. This device functions using the formation and breakage of the connections (“synapses”) between clusters, driven by the bias voltage applied to the clusters. The mechanisms of the formation and breakage of these connections are thus of the utmost interest. In this thesis, the molecular dynamics simulation method is used to explore the mechanisms of the formation and breakage of the connections (“filaments”) between the clusters in a model of neuromorphic device. First, the Joule heating mechanism of filament breakage is explored using a model consisting of Au nanowire that connects two Au1415 clusters. Upon heating, the atoms of the nanofilament gradually aggregate towards the clusters, causing the middle of the wire to graduallythin and then suddenly break. Most of the system remains crystalline during this process, but the centre becomes molten. The terminal clusters increase the melting point of the nanowires by fixing them and act as recrystallisation regions. A strong dependence of the breaking temperature is found not only on the width of the nanowires but also their length and atomic structure. Secondly, the bridge formation and thermal breaking processes between Au1415 clusters on a graphite substrate are also simulated. The bridging process , which can heal a broken filament, is driven by diffusion of gold along the graphite substrate. The characteristic times of bridge formation are explored at elevated simulation temperatures to estimate the longer characteristic times of formation at room-temperature conditions. The width of the bridge formed has a power-law dependence on the simulation time, and the mechanism is a combination of diffusion and viscous flow. Simulations of bridgebreaking are also conducted and reveal the existence of a voltage threshold that must be reached to activate the breakage. The role of the substrate in the bridge formation and breakage processes is revealed as a medium of diffusion. Lastly, to explore future potential cluster materials, the thermal behaviour of Pb-Al core-shell clusters is studied. The core and shell are found to melt separately. In fact, the core atoms of nanoclusters tend to escape their shells and partially cover them, leading to a preference for a segregated state. The melting point of the core can either be depressed or elevated, depending on the thickness of the shell due to different mechanisms. |
|---|---|
| Keywords: |
Neuromorphic Computing, Cluster-Based Neuromorphic Devices, Molecular Dynamics Simulation, Filament Formation and Breakage, Joule Heating Mechanism, Nanocluster, Nanowire, Diffusion, Melting |
| College: |
Faculty of Science and Engineering |