Nowadays, the development of algorithms focuses on performance-efficient and energy-efficient computations. Technologies such as field programmable gate array (FPGA) and system on chip (SoC) based on FPGA (FPGA/SoC) have shown their ability to accelerate intensive computing applications while saving power consumption, owing to their capability of high parallelism and reconfiguration of the architecture. Currently, the existing design cycles for FPGA/SoC are time-consuming, owing to the complexity of the architecture. Therefore, to address the gap between applications and FPGA/SoC architectures and to obtain an efficient hardware design for image analysis and highly demanding applications using the high-level synthesis tool, two complementary strategies are considered: ad-hoc techniques and performance estimator. Regarding ad-hoc techniques, three highly demanding applications were accelerated through HLS tools: pulse shape discriminator for cosmic rays, automatic pest classification, and re-ranking for information retrieval, emphasizing the benefits when this type of applications are traversed by compression techniques when targeting FPGA/SoC devices. Furthermore, a comprehensive performance estimator for hardware acceleration is proposed in this thesis to effectively predict the resource utilization and latency for FPGA/SoC, building a bridge between the application and architectural domains. The tool integrates analytical models for performance prediction, and a design space explorer (DSE) engine for providing high-level insights to hardware developers, composed of two independent sub-engines: DSE based on single-objective optimization and DSE based on evolutionary multi-objective optimization.

Al giorno d'oggi, lo sviluppo di algoritmi si concentra su calcoli efficienti in termini di prestazioni ed efficienza energetica. Tecnologie come il field programmable gate array (FPGA) e il system on chip (SoC) basato su FPGA (FPGA/SoC) hanno dimostrato la loro capacità di accelerare applicazioni di calcolo intensive risparmiando al contempo il consumo energetico, grazie alla loro capacità di elevato parallelismo e riconfigurazione dell'architettura. Attualmente, i cicli di progettazione esistenti per FPGA/SoC sono lunghi, a causa della complessità dell'architettura. Pertanto, per colmare il divario tra le applicazioni e le architetture FPGA/SoC e ottenere un design hardware efficiente per l'analisi delle immagini e altri applicazioni altamente demandanti utilizzando lo strumento di sintesi di alto livello, vengono prese in considerazione due strategie complementari: tecniche ad hoc e stima delle prestazioni. Per quanto riguarda le tecniche ad-hoc, tre applicazioni molto impegnative sono state accelerate attraverso gli strumenti HLS: discriminatore di forme di impulso per i raggi cosmici, classificazione automatica degli insetti e re-ranking per il recupero delle informazioni, sottolineando i vantaggi quando questo tipo di applicazioni viene attraversato da tecniche di compressione durante il targeting dispositivi FPGA/SoC. Inoltre, in questa tesi viene proposto uno stimatore delle prestazioni per l'accelerazione hardware per prevedere efficacemente l'utilizzo delle risorse e la latenza per FPGA/SoC, costruendo un ponte tra l'applicazione e i domini architetturali. Lo strumento integra modelli analitici per la previsione delle prestazioni e un motore design space explorer (DSE) per fornire approfondimenti di alto livello agli sviluppatori di hardware, composto da due motori indipendenti: DSE basato sull'ottimizzazione a singolo obiettivo e DSE basato sull'ottimizzazione evolutiva multiobiettivo.

SoC-based FPGA architecture for image analysis and other highly demanding applications / Molina, ROMINA SOLEDAD. - (2023 Mar 01).

SoC-based FPGA architecture for image analysis and other highly demanding applications

MOLINA, ROMINA SOLEDAD
2023-03-01

Abstract

Nowadays, the development of algorithms focuses on performance-efficient and energy-efficient computations. Technologies such as field programmable gate array (FPGA) and system on chip (SoC) based on FPGA (FPGA/SoC) have shown their ability to accelerate intensive computing applications while saving power consumption, owing to their capability of high parallelism and reconfiguration of the architecture. Currently, the existing design cycles for FPGA/SoC are time-consuming, owing to the complexity of the architecture. Therefore, to address the gap between applications and FPGA/SoC architectures and to obtain an efficient hardware design for image analysis and highly demanding applications using the high-level synthesis tool, two complementary strategies are considered: ad-hoc techniques and performance estimator. Regarding ad-hoc techniques, three highly demanding applications were accelerated through HLS tools: pulse shape discriminator for cosmic rays, automatic pest classification, and re-ranking for information retrieval, emphasizing the benefits when this type of applications are traversed by compression techniques when targeting FPGA/SoC devices. Furthermore, a comprehensive performance estimator for hardware acceleration is proposed in this thesis to effectively predict the resource utilization and latency for FPGA/SoC, building a bridge between the application and architectural domains. The tool integrates analytical models for performance prediction, and a design space explorer (DSE) engine for providing high-level insights to hardware developers, composed of two independent sub-engines: DSE based on single-objective optimization and DSE based on evolutionary multi-objective optimization.
1-mar-2023
RAMPONI, GIOVANNI
35
2021/2022
Settore ING-INF/01 - Elettronica
Università degli Studi di Trieste
File in questo prodotto:
File Dimensione Formato  
RM_thesis2022.pdf

accesso aperto

Descrizione: SoC-based FPGA architecture for image analysis and other highly demanding applications
Tipologia: Tesi di dottorato
Dimensione 6.85 MB
Formato Adobe PDF
6.85 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11368/3043618
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
social impact