Back to search

National Technical University of Athens (NTUA)

Ηλεκτρονικές διατάξεις νανοσωματιδίων

Abstract

dc:description

In recent years the need to develop new computing architectures beyond the established von Neumann architecture has become clear mainly due to the need for the large volume of new data to be analyzed and processed in a short period of time. In this regard, the implementation of neuromorphic computing systems seems to offer increased efficiency and accuracy in calculations. Such implementations that use resistive switching memories (ReRAM) as the link to the biologically equivalent elements (synapses/neurons) have been the subject of extensive research in recent years. These devices seem to have, in addition to their trouble-free compatibility with CMOS techniques, the ability to function simply as non-volatile memories and store multiple bits of information, as well as to comprise the means of implementing machine learning techniques into neuromorphic devices.This memory technology, which was the item of research in this dissertation, can and does mimic, based on the process of development and rupture of a conductive filament, the way in which the changes of synaptic weight in biological synapses are made. Having this as a main concern, then, a number of issues are brought up, such as the stochasticity of the switching phenomenon and the statistical dispersion of the results, the low power consumption and the maximization of the memory window. One understands that in a high-density neuromorphic implementation such issues, especially of power, are vital. In this direction and by investigating both subcategories of ReRAM, namely valence change memory (VCM) and conductive bridge (CBRAM), an attempt was made to optimize the operation of memory nanodevices so that they can be used seamlessly in dense crossbar structures.A variety of materials were used to investigate the effect of complex deposition parameters on the electrical performance of memory nanodevices, focusing mainly on titanium oxide (TiOx), hafnium oxide (HfOx), tantalum oxide (TaOx) and silicon dioxide (SiO2). The use of multiple layers of active material within a VCM device was investigated as a way to achieve not only low statistical dispersion and reduction of operating currents, but also to enhance synaptic properties. This is achieved by depositing low, but also high, in oxygen content dielectric layers, with the former acting as series resistors and reducing power consumption. Consequently, the resistive switching phenomenon is limited to a smaller volume of switching oxide, i.e. the layer with the highest oxygen concentration, thus improving the statistical variability. In addition, the incorporation of metal nanocrystals (NCs) inside the dielectric matrix makes the localization of the electric field possible, thus increasing the effective diameter of a conductive filament (CF). This in turn can lead to larger memory windows, even in low bias conditions. The enhancement of the electric field is also a key factor in limiting the possible percolation paths, where the resistive switching phenomenon takes place.The use of SiO2 as a switching layer in CBRAM devices demonstrated a variety of properties both in terms of memory function and of synaptic plasticity. Very low sweep bias amplitudes (< 500 mV) are required for the switching effect to transpire with large enough memory windows and no need to electroform the device. In the comparison between the two electrochemically active Ag and Cu top electrodes, it was shown that the difference in the melting point of the formed metal ion nanoclusters that constitute the CF is the origin of the threshold switching effect. The nature of the diffusion of the moving ions and the dynamics of the formation/rupture of the CF demonstrated the ability to modulate the conductivity of the device under the application of pulses with the same polarity albeit of different amplitude. The embedding of Ag/Cu nanocrystals underneath the top Ag/Cu electrode showed that it somewhat restricts the movement of metal ions in the intermediate space between NCs resulting in the need for higher switching and sweeping voltages. In addition, being the same electrochemically active material as the top electrode, it is reasonable to observe some variation in the results after some sweep cycles. Moreover, the introduction of Pt NCs further improved statistical variability and spatially confined the random evolution of CFs while at the same time making it possible to regulate the CF diameter. Through this research, it was possible to achieve low-power multibit properties, as well as steeper transitions due to the spatial confinement of the CF’s evolution, highlighting the different evolution dynamics of silver nanowires in the presence of metallic NCs and its beneficial effect in memristive device operation.The thorough investigation of synaptic performance in all cases has shown that the utilization of these memory devices in high density neuromorphic implementations has great potential. The proposed optimizations of this dissertation can lead to high-performance neural networks while utilizing the multitude of synaptic properties exhibited by these devices, as well as the extremely low power consumption per synaptic event observed.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2021

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Sakellaropoulos, Dionysios
  • Σακελλαρόπουλος, Διονύσιος

Subjects

dc:subject × 28

Rights

dc:rights
Statement dc:rights
  • BY_NC_4_0
Language dc:language
gre

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/49396

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Sakellaropoulos, Dionysios; Σακελλαρόπουλος, Διονύσιος. Ηλεκτρονικές διατάξεις νανοσωματιδίων. National Technical University of Athens (NTUA), 2021. http://hdl.handle.net/10442/hedi/49396