Back to search

Aristotle University Of Thessaloniki (AUTH)

Design methodologies for sustainable hardware acceleration at the edge

Abstract

dc:description

The rapid rise of Internet of Things (IoT) devices, along with the proliferation of applications and services requiring low latency and enhanced privacy, has led to increased demand for computational power closer to where data originates, at the edge of the network. This trend has given rise to the edge computing paradigm, where local computational and storage capabilities on devices are utilized for inference of tasks that are unsuitable for execution at centralized data centers. However, the heterogeneity of computing devices and their limited computational capabilities pose significant challenges in meeting the performance, power, and privacy requirements of diverse tasks executed within this ecosystem. Specialized hardware devices, such as ASICs, GPUs, and FPGAs, are ideal candidates for accelerating time-critical applications due to their custom architectures (ASICs, FPGAs), high parallelism (GPUs), and rapid development cycles (GPUs, FPGAs). Furthermore, the increasing availability of System-on-Chip (SoC) devices facilitates the deployment of such specialized accelerators in the resource- and power-constrained environments typically found at the network edge, while providing flexible design and deployment solutions. However, the growing demand for "green" sustainable computing often conflicts with the increasing computational and storage demands characteristic of edge computing. Sustainable computing requires minimal resource usage, reduced power budgets, and a low carbon footprint, while edge computing often demands high computational intensity. Exploiting the capabilities of specialized hardware devices and proposing effective design and deployment methodologies to satisfy both the strict constraints of edge execution and sustainability goals is the central objective of this dissertation. The proposed methodologies aim to cover the entire spectrum of the edge computing ecosystem—from extremely resource-constrained environments up to Mobile Edge Computing (MEC) servers deployed close to Radio Access Network (RAN) equipment, providing cloud-like capabilities to mobile users requiring time-sensitive processing. The dissertation follows an approach that begins from extremely constrained environments and progresses toward edge servers—often referred to as fog computing—proposing design methodologies across the entire stack. Specifically, design methodologies are introduced for ASICs fabricated using printed electronics, leveraging the unique features of these technologies to automate the creation of area- and power-efficient Multi-Layer Perceptron (MLP) classifiers. Additionally, methodologies for designing FPGA accelerators are proposed for both the TinyML domain and more resource-abundant environments. In the case of TinyML, a resource-utilization estimation methodology is proposed to enable the rapid and efficient design-space exploration of FPGA-based MLP accelerators designed through widely used acceleration workflows. For environments supporting more extensive resources, the dissertation presents a methodology to consolidate multiple FPGA accelerators onto a single platform, ensuring reliable execution and efficient sharing of FPGA on-chip resources. Furthermore, a Mobile Edge Computing acceleration architecture combined with an Optical Circuit Switch (OCS) is presented, focusing on the deployment of multiple accelerated applications onto MEC heterogeneous resources. Emphasis is given to performance, energy efficiency, isolated execution, and the associated control schemes developed for managing the underlying MEC infrastructure.

Degree

thesis:*
Grantor dc:publisher
Aristotle University Of Thessaloniki (AUTH)
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kokkinis, Argyrios
  • Κοκκίνης, Αργύριος

Subjects

dc:subject × 16

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/60031

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

Kokkinis, Argyrios; Κοκκίνης, Αργύριος. Design methodologies for sustainable hardware acceleration at the edge. Aristotle University Of Thessaloniki (AUTH), 2025. http://hdl.handle.net/10442/hedi/60031