TY - JOUR
T1 - How Reduced Data Precision and Degree of Parallelism Impact the Reliability of Convolutional Neural Networks on FPGAs
AU - Libano, F.
AU - Rech, P.
AU - Neuman, B.
AU - Leavitt, J.
AU - Wirthlin, M.
AU - Brunhaver, J.
N1 - Funding Information:
Manuscript received November 20, 2020; revised December 27, 2020; accepted January 7, 2021. Date of publication January 11, 2021; date of current version May 20, 2021. This work was supported in part by the Department of Energy of the United States, in part by the CAPES Foundation of the Ministry of Education, and in part by the CNPq Research Council of the Ministry of Science and Technology.
Publisher Copyright:
© 1963-2012 IEEE.
PY - 2021/5
Y1 - 2021/5
N2 - Convolutional neural networks (CNNs) are becoming attractive alternatives to traditional image-processing algorithms in self-driving vehicles for automotive, military, and aerospace applications. The high computational demand of state-of-the-art CNN architectures requires the use of hardware acceleration on parallel devices. Field-programmable gate arrays (FPGAs) offer a great level of design flexibility, low power consumption, and are relatively low cost, which make them very good candidates for efficiently accelerating neural networks. Unfortunately, the configuration memories of SRAM-based FPGAs are sensitive to radiation-induced errors, which can compromise the circuit implemented on the programmable fabric and the overall reliability of the system. Through neutron beam experiments, we evaluate how lossless quantization processes and subsequent data precision reduction impact the area, performance, radiation sensitivity, and failure rate of neural networks on FPGAs. Our results show that an 8-bit integer design can deliver over six times more fault-free executions than a 32-bit floating-point implementation. Moreover, we discuss the tradeoffs associated with varying degrees of parallelism in a neural network accelerator. We show that, although increased parallelism increases radiation sensitivity, the performance gains generally outweigh it in terms of global failure rate.
AB - Convolutional neural networks (CNNs) are becoming attractive alternatives to traditional image-processing algorithms in self-driving vehicles for automotive, military, and aerospace applications. The high computational demand of state-of-the-art CNN architectures requires the use of hardware acceleration on parallel devices. Field-programmable gate arrays (FPGAs) offer a great level of design flexibility, low power consumption, and are relatively low cost, which make them very good candidates for efficiently accelerating neural networks. Unfortunately, the configuration memories of SRAM-based FPGAs are sensitive to radiation-induced errors, which can compromise the circuit implemented on the programmable fabric and the overall reliability of the system. Through neutron beam experiments, we evaluate how lossless quantization processes and subsequent data precision reduction impact the area, performance, radiation sensitivity, and failure rate of neural networks on FPGAs. Our results show that an 8-bit integer design can deliver over six times more fault-free executions than a 32-bit floating-point implementation. Moreover, we discuss the tradeoffs associated with varying degrees of parallelism in a neural network accelerator. We show that, although increased parallelism increases radiation sensitivity, the performance gains generally outweigh it in terms of global failure rate.
KW - Field-programmable gate array (FPGA)
KW - neural networks
KW - parallelism
KW - reduced precision
KW - reliability
UR - http://www.scopus.com/inward/record.url?scp=85099534734&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85099534734&partnerID=8YFLogxK
U2 - 10.1109/TNS.2021.3050707
DO - 10.1109/TNS.2021.3050707
M3 - Article
AN - SCOPUS:85099534734
VL - 68
SP - 865
EP - 872
JO - IEEE Transactions on Nuclear Science
JF - IEEE Transactions on Nuclear Science
SN - 0018-9499
IS - 5
M1 - 9319148
ER -