Personnel
Overall Objectives
Research Program
Application Domains
New Software and Platforms
New Results
Bilateral Contracts and Grants with Industry
Partnerships and Cooperations
Dissemination
Bibliography
XML PDF e-pub
PDF e-Pub


Section: New Results

Error-Rate Prediction and Radiation Experiments on a 28nm Many-Core Processor

Participants : Vanessa Vargas [TIMA Labs Grenoble & ESPE, Ecuador] , Pablo Ramos [TIMA Labs Grenoble & ESPE, Ecuador] , Vincent Ray [kalray, Montbonnot Saint-Martin] , Camille Jalier [kalray, Montbonnot Saint-Martin] , Renaud Stevens [kalray, Montbonnot Saint-Martin] , Benoît Dupont de Dinechin [kalray, Montbonnot Saint-Martin] , Maud Baylac [LSPC Labs, Université Grenoble Alpes, CNRS/IN2P3] , Francesca Villa [LSPC Labs, Université Grenoble Alpes, CNRS/IN2P3] , Solenne Rey [LSPC Labs, Université Grenoble Alpes, CNRS/IN2P3] , Nacer-Eddine Zergainoh [TIMA Labs, Université Grenoble Alpes & CNRS] , Jean-Francois Mehaut, Raoul Velazco [TIMA Labs, Université Grenoble Alpes & CNRS] .

This work analyses the 14 MeV neutron sensitivity of the MPPA-256 many-core processor. Analysis results suggest that ECC and interleaving implemented in the shared memories of clusters are very effective to mitigate SEUs as all detected events of this type were corrected.

The evaluation of the device dynamic response shows that by enabling the cache memories, it is possible to gain in performance of the application without compromising reliability, as all the detected errors produced in data and instruction cache memories were corrected by the parity protection. The non-correctable errors that occurred in the different dynamic tests were produced by bit-flips in general purpose registers since registers do not implement any protection mechanism.

Results show that the predicted application error-rate is reasonably close to the measured one. Consequently, despite the complexity of the many-core processor, this work supports the relevance of the use of the CEU approach to predict the error-rate of applications implemented in such devices.

This work is part of the Vanessa Vargas's PhD thesis defended in April 2017. This PhD thesis was advised by Raoul Velazco (TIMA) and Jean-François Méhaut. Four of the authors worked for the Kalray company. The experiments under radiation were performed on the GENEPI2 platform of the LSPC laboratory. This work was also published in the IEEE Transactions on Nuclear Sciences [10]. It was partially funded by the STIC-Amsud EnergySFE project 8.4.2.