A4 Vertaisarvioitu artikkeli konferenssijulkaisussa
Design of Fault - Tolerant and Reliable Networks-on-Chip
Tekijät: Wang JS, Ebrahimi M, Huang L, Jantsch A, Li GJ
Toimittaja: Aida Todri-Sanial, Patrick Girard, Giorgio Di Natale
Konferenssin vakiintunut nimi: IEEE Computer Society Annual Symposium on VLSI
Julkaisuvuosi: 2015
Kokoomateoksen nimi: 2015 IEEE COMPUTER SOCIETY ANNUAL SYMPOSIUM ON VLSI
Tietokannassa oleva lehden nimi: 2015 IEEE COMPUTER SOCIETY ANNUAL SYMPOSIUM ON VLSI
Lehden akronyymi: IEEE COMPUT SOC
Sarjan nimi: IEEE Computer Society Annual Symposium on VLSI
Aloitussivu: 545
Lopetussivu: 550
Sivujen määrä: 6
ISBN: 978-1-4799-8718-4
ISSN: 2159-3469
DOI: https://doi.org/10.1109/ISVLSI.2015.33
Tiivistelmä
Networks-on-Chips (NoCs) are at the core of high performance multi-processor systems-on-chips. As the number of cores and sub-systems on chip grows, the size and complexity of NoCs increase as well. Due to the process variation, aging effects and soft-errors in current and expected future process generations, the probability of failure in the NoCs rises and has to be fought at all levels: circuit, architecture, and communication protocols.This paper discusses appropriate fault models for NoCs and their effects on the architecture and network levels. A method to design fault-tolerant NoCs comprising of techniques at the link level, the routing level, and the end-to-end level of the communication is presented. In addition, the proposed method offers an isolation technique where the computing cores are decoupled from the faults in the network. This technique avoids or at least attenuates the severe impacts of faults on the network performance and functionality. These point techniques are combined together to design fault-tolerant and reliable NoCs.
Networks-on-Chips (NoCs) are at the core of high performance multi-processor systems-on-chips. As the number of cores and sub-systems on chip grows, the size and complexity of NoCs increase as well. Due to the process variation, aging effects and soft-errors in current and expected future process generations, the probability of failure in the NoCs rises and has to be fought at all levels: circuit, architecture, and communication protocols.This paper discusses appropriate fault models for NoCs and their effects on the architecture and network levels. A method to design fault-tolerant NoCs comprising of techniques at the link level, the routing level, and the end-to-end level of the communication is presented. In addition, the proposed method offers an isolation technique where the computing cores are decoupled from the faults in the network. This technique avoids or at least attenuates the severe impacts of faults on the network performance and functionality. These point techniques are combined together to design fault-tolerant and reliable NoCs.