From dataflow models to energy efficient application specific processors
Thesis event information
Date and time of the thesis defence
Place of the thesis defence
L10, Linnanmaa, University of Oulu
Topic of the dissertation
From dataflow models to energy efficient application specific processors
Doctoral candidate
Master of Science (Tech) Ilkka Hautala
Faculty and unit
University of Oulu Graduate School, Faculty of Information Technology and Electrical Engineering, Center for Machine Vision and Signal Analysis
Subject of study
Computer science
Opponent
Professor Christoph Kessler, Linköping University
Custos
Associate Professor Jani Boutellier, University of Vaasa
Toolchain for implementing energy-efficient and programmable signal processing systems
The subject of the thesis is the development of high-level design methods for the implementation of energy-efficient and programmable signal processing systems. The development methods presented in the thesis have been adopted for traditional application areas of digital signal processing such as video coding, baseband processing, and machine vision.
One of the biggest challenges of implementing new generation signal processing applications is their energy requirements. There have been attempts to tackle energy efficiency problems by introducing parallel and application tailored hardware resources. The downside of those approaches is that they generally hamper the software and hardware development. It is because the traditional design flows tend to describe the software and hardware using a low level of abstraction. For this reason, exploring the design space of different hardware options for finding the optimal ones are economically unfeasible. By raising the level of abstraction of signal processing software and taking advantage of the automatized hardware description tools, it is possible to come up with multiple different solutions for the same application and finally choose a solution whose characteristic is the most suitable. The thesis proposes a toolchain, which can be used for designing signal processing systems through software models with a high level of abstraction together with energy-efficient application-specific processors.
The results of the thesis suggest that raising the abstraction level of the signal processing software is possible without a significant negative impact on overall system performance. Furthermore, the research presents the energy-efficient and programmable implementations for high-end video-coding algorithms. The results of this research help to find a solution that balances between highly efficient yet inflexible fixed hardware and highly flexible but inefficient general-purpose processors.
One of the biggest challenges of implementing new generation signal processing applications is their energy requirements. There have been attempts to tackle energy efficiency problems by introducing parallel and application tailored hardware resources. The downside of those approaches is that they generally hamper the software and hardware development. It is because the traditional design flows tend to describe the software and hardware using a low level of abstraction. For this reason, exploring the design space of different hardware options for finding the optimal ones are economically unfeasible. By raising the level of abstraction of signal processing software and taking advantage of the automatized hardware description tools, it is possible to come up with multiple different solutions for the same application and finally choose a solution whose characteristic is the most suitable. The thesis proposes a toolchain, which can be used for designing signal processing systems through software models with a high level of abstraction together with energy-efficient application-specific processors.
The results of the thesis suggest that raising the abstraction level of the signal processing software is possible without a significant negative impact on overall system performance. Furthermore, the research presents the energy-efficient and programmable implementations for high-end video-coding algorithms. The results of this research help to find a solution that balances between highly efficient yet inflexible fixed hardware and highly flexible but inefficient general-purpose processors.
Last updated: 1.3.2023