Change search
ReferencesLink to record
Permanent link

Direct link
Adaptive Computing based on FPGA Run-time Reconfigurability
KTH, School of Information and Communication Technology (ICT), Electronic Systems.
2011 (English)Doctoral thesis, monograph (Other academic)
Abstract [en]

In the past two decades, FPGA has been witnessed from its restricted use as glue logic towards real System-on-Chip (SoC) platforms. Profiting from the great development on semiconductor and IC technologies, the programmability of FPGAs enables themselves wide adoption in all kinds of aspects of embedded designs. Modern FPGAs provide the additional capability of being dynamically and partially reconfigured during the system run-time. The run-time reconfigurability enhances FPGA designs from the sole spatial to both spatial and temporal parallelism, providing more design flexibility for advanced system features.

Adaptive computing delegates an advanced computing paradigm in which computation tasks and resources are intelligently managed in correspondence with conditional requirements. In this thesis, we investigate adaptive designs on FPGA platforms: We present a comprehensive and practical design framework for adaptive computing based on the FPGA run-time reconfigurability. It concerns several design key issues in different hardware/software layers, specifically hardware architecture, run-time reconfiguration technical support, OS and device drivers, hardware process scheduler, context switching as well as Inter-Process Communications (IPC). Targeting a special application of data acquisition (DAQ) and trigger systems in nuclear and particle physics experiments, we set up the data streaming model and conduct theoretical analysis on the adaptive system. Three application studies are employed to verify the proposed adaptive design framework: The first application demonstrates a peripheral controller adaptable system aiming at general embedded designs. Through dynamically loading/unloading a NOR flash memory controller and an SRAM controller, both flash memory and SRAM accesses may be accomplished with less resource consumption than in traditional static designs. In the second case, two real algorithm processing engines are adaptively time-multiplexed in the same reconfigurable slot for particle recognition computation. Experimental results reveal the reduced on-chip resource requirements, as well as an approximate processing capability of the peer static design. Taking advantage of the FPGA dynamic reconfigurability, we present in the third application a novel on-FPGA interconnection microarchitecture named RouterLess NoC (RL-NoC). RL-NoC employs the novel design concept of Move Logic Not Data (MLND), and significantly distinguishes itself from the existing interconnection architectures such as buses, crossbars or NoCs. It does not rely on routers to deliver packets hop by hop as canonical NoCs do, but buffers data packets in virtual channels and brings various nodes using run-time reconfiguration to produce or consume them. In comparison with canonical packet-switching NoCs, the routerless architecture features lower design complexity, less resource consumption, higher work frequency, more efficient power dissipation as well as comparable or even higher packet delivery efficiency. It is regarded as a promising interconnection approach in some design scenarios on FPGAs, especially for light-weight applications.

Place, publisher, year, edition, pages
Stockholm: KTH Royal Institute of Technology , 2011. , xx, 165 p.
Series
Trita-ICT-ECS AVH, ISSN 1653-6363 ; 11:05
Keyword [en]
adaptive computing, FPGA partial reconfiguration
National Category
Computer and Information Science
Identifiers
URN: urn:nbn:se:kth:diva-33866ISBN: 978-91-7415-985-1OAI: oai:DiVA.org:kth-33866DiVA: diva2:418205
Public defence
2011-06-14, Sal D, Forum, Isafjordsgatan 39, Kista, 11:38 (English)
Opponent
Supervisors
Note
QC 20110531Available from: 2011-05-31 Created: 2011-05-20 Last updated: 2011-08-10Bibliographically approved

Open Access in DiVA

Ph.D thesis of Ming Liu(6299 kB)2645 downloads
File information
File name FULLTEXT01.pdfFile size 6299 kBChecksum SHA-512
0053f157d5572c4b19c3d42ab7c03d2378820ac7e82c42b6e7d956a34b3f8b312fee3df126fe6c08768d5deaa620d464c45c2cb09609f2273d51dbe478d12bb4
Type fulltextMimetype application/pdf

Search in DiVA

By author/editor
Liu, Ming
By organisation
Electronic Systems
Computer and Information Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 2645 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 299 hits
ReferencesLink to record
Permanent link

Direct link