Analyzing and Increasing the Reliability of Convolutional Neural Networks on GPUs
The DOI matches its Crossref record; 68 of the 68 references it cites were found in a registry or an index.
Abstract
Graphics processing units (GPUs) are playing a critical role in convolutional neural networks (CNNs) for image detection. As GPU-enabled CNNs move into safety-critical environments, reliability is becoming a growing concern. In this paper, we evaluate and propose strategies to improve the reliability of object detection algorithms, as run on three NVIDIA GPU architectures. We consider three algorithms: 1) you only look once; 2) a faster region-based CNN (Faster R-CNN); and 3) a residual network, exposing live hardware to neutron beams. We complement our beam experiments with fault injection to better characterize fault propagation in CNNs. We show that a single fault occurring in a GPU tends to propagate to multiple active threads, significantly reducing the reliability of a CNN. Moreover, relying on error correcting codes dramatically reduces the number of silent data corruptions (SDCs), but does not reduce the number of critical errors (i.e., errors that could potentially impact safety-critical applications). Based on observations on how faults propagate on GPU architectures, we propose effective strategies to improve CNN reliability. We also consider the benefits of using an algorithm-based fault-tolerance technique for matrix multiplication, which can correct more than 87% of the critical SDCs in a CNN, while redesigning maxpool layers of the CNN to detect up to 98% of critical SDCs.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“as commented in [9], SASSIFI is the only tool working with the NVIDIA proprietary libraries generally employed when implementing CNNs.”
From Fast and Accurate Error Simulation for CNNs Against Soft Errors (2023) · cites this paper as [9]
In the citing paper
“radiation test offers the most adherent solution to cause soft errors, however, apart from the highly expensive equipment and the complex setup, there is little to no controllability on the injection process, and the observability is only on the primary outputs. As a result, it is suitable for final validation and blackbox campaigns, not for in-depth analysis and investigations [9], [36].”
From Fast and Accurate Error Simulation for CNNs Against Soft Errors (2023) · cites this paper as [9]
Citations
Cites
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2017) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2577031 · Cited by 73,472
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2015) · arXiv · doi:10.48550/arxiv.1506.01497 · Cited by 70,812
Gradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 1×
“Specifically, neural networks for object detection have the peculiarity of using convolution to reduce the information to be processed [15], a class of computations where reliability has not been studied deeply to date.”
BackgroundYou Only Look Once: Unified, Real-Time Object Detection
Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.91 · Cited by 53,803
An introduction to ROC analysis
Fawcett, T. (2006) · Pattern Recognition Letters · doi:10.1016/j.patrec.2005.10.010 · Cited by 21,879 · cited 2×
“5 to compare the corrupted output of a CNN with the radiation-free output, following the T J tradeoff discussion presented in Fawcett et al. [47], which is valid, independent of the source of detection imprecision (intrinsic algorithm detection imprecision or radiation-induced corruption).”
MethodThe pascal visual object classes (VOC) challenge
Everingham, M., Van Gool, L., Williams, C. K. I., Winn, J., & Zisserman, A. (2010) · International Journal of Computer Vision · doi:10.1007/s11263-009-0275-4 · Cited by 19,599 · cited 1×
“We use the PASCAL VOC 2012 [48] and the Caltech Pedes-trian datasets [49] as input for our study.”
MethodCaffe: Convolutional Architecture for Fast Feature Embedding
Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadarrama, S., & Darrell, T. (2014) · Proceedings of the 22nd ACM international conference on Multimedia · doi:10.1145/2647868.2654889 · Cited by 11,077 · cited 2×
“Convolutional neural networks (CNNs) have, in fact, been deployed in self-driving cars to dynamically identify/classify objects [2], [3].”
BackgroundMethodIdentity mappings in deep residual networks
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-46493-0_38 · Cited by 10,741
Identity Mappings in Deep Residual Networks
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · arXiv · doi:10.48550/arxiv.1603.05027 · Cited by 10,665 · cited 2×
“This process is performed repeatedly for all layers in the network, allowing us to use deeper network structures (e.g., 50 to 1000 layers) without impacting precision, and without increasing the algorithmic complexity [45].”
MethodIn-Datacenter Performance Analysis of a Tensor Processing Unit
Jouppi, N. P., Young, C., Patil, N., Patterson, D., Agrawal, G., Bajwa, R., Bates, S., Bhatia, S., Boden, N., Borchers, A., et al. (2017) · Proceedings of the 44th Annual International Symposium on Computer Architecture · doi:10.1145/3079856.3080246 · Cited by 4,537 · cited 1×
“As GPUs are highly efficient for accelerating matrix multiplication, they are a great target for CNNs. Specific processors can be designed for DNN, achieving good performance and low power consumption [42], [43].”
Background
Research works printed but not linked: 4
- sridharan, 2009, The effect of input data on program vulnerability, Workshop on Silicon Errors in Logic-System Effects, 1
- 10.1109/TNS.2014.2301768 · DOI
- 2018, Log database for IEEE transaction on reliability, Github
- collobert, 0, BigLearn NIPS Workshop
Cited by
Items 101–200 of 223
Error Resilience in Deep Neural Networks Using Neuron Gradient Statistics
Amarnath, C., Mejri, M., Ma, K., & Chatterjee, A. (2023) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2023.3335144 · Cited by 11
“Radiation study of soft error vulnerability in CNNs was used in the design of ABFT methods in [31].”
MethodTechniques for detecting and masking faults in semantic segmentation applications
Burel, S., Evans, A., & Anghel, L. (2024) · Microelectronics Reliability · doi:10.1016/j.microrel.2024.115397 · Cited by 11
An Approximation-based Fault Detection Scheme for Image Processing Applications
Biasielli, M., Cassano, L., & Miele, A. (2020) · Design, Automation, and Test in Europe · doi:10.23919/date48585.2020.9116425 · Cited by 8
“The idea of analysing the usability of the overall result of an image processing application has been considered in recent publications to assess the robustness of machine learning applications for image processing [11], [12].”
BackgroundThermal Neutrons: a Possible Threat for Supercomputers and Safety Critical Applications
Oliveira, D., Blanchard, S. P., DeBardeleben, N., dos Santos, F. F., Dávila, G. P., Navaux, P. O. A., Cazzaniga, C., Frost, C. D., Baumann, R., & Rech, P. (2020) · IEEE European Test Symposium (ETS) · doi:10.1109/ets48528.2020.9131597 · Cited by 11
“detection [22].”
Self-citationMILR: Mathematically Induced Layer Recovery for Plaintext Space Error Correction of CNNs
Ponader, J., Thomas, K., Kundu, S. K., & Solihin, Y. (2021) · International Conference on Dependable Systems & Networks/Proceedings · doi:10.1109/dsn48987.2021.00024 · Cited by 9
Improving DNN Fault Tolerance in Semantic Segmentation Applications
BurelT, S., EvansT, A., & Anghel, L. (2022) · Proceedings · doi:10.1109/dft56152.2022.9962354 · Cited by 10
“In dos Santos [23] they propose the Reliable MaxPooling technique that is based on a threshold for the MaxPooling layers arbitrarily set at 10x the maximum value observed during training.”
MethodResultA machine-learning-guided framework for fault-tolerant DNNs
Traiola, M., Kritikakou, A. S., & Sentieys, O. (2023) · Design, Automation, and Test in Europe · doi:10.23919/date56975.2023.10137033 · Cited by 8
“The sources of faults are various, e.g., escaped post-fabrication faults, silicon ageing, and environmental stress, such as heat, humidity, vibration, and radiation [1].”
Bolchini, C., Cassano, L., & Miele, A. (2024) · Computer Science Review · doi:10.1016/j.cosrev.2024.100682 · Cited by 25
“…Time (FIT), Architecture Vulnerability Factor (AVF), Program Vulnerability Factor (PVF), Kernel Vulnerability Factor (KVF) or the SDC rate (e.g., [39]) and those who adopt an application-aware metric, more closely related to the specific and special context, such as usable / not usable (e.g.,…”
BackgroundMethodImproving Deep Neural Network Reliability via Transient-Fault-Aware Design and Training
dos Santos, F. F., Cavagnero, N., Ciccone, M., Averta, G., Kritikakou, A. S., Sentieys, O., Rech, P., & Tommasi, T. (2025) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2024.3520672 · Cited by 16
“Software-based hardening solutions include Algorithm-Based Fault Tolerance (ABFT) [10], [11], [25], [27], [32], and filters for identifying propagation of incorrect values [14], [24], [32].”
InfluentialSelf-citationVéstias, M. P., Duarte, R. P., de Sousa, J. T., & Neto, H. C. (2022) · ACM Transactions on Reconfigurable Technology and Systems · doi:10.1145/3546182 · Cited by 8
“The preferred computing platforms are therefore GPUs (Graphics Processing Units) [26], FPGAs (Field-Programmable Gate Arrays) [6], and ASICs (Application Specific Integrated Circuits) [20].”
BackgroundAn efficient structure to improve the reliability of deep neural networks on ARMs
Liu, Z., & Yang, X. (2022) · Microelectronics Reliability · doi:10.1016/j.microrel.2022.114729 · Cited by 7
Hsiao, Y.-S., Wan, Z., Jia, T., Ghosal, R., Mahmoud, A., Raychowdhury, A., Brooks, D., Wei, G.-Y., & Reddi, V. J. (2023) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2023.3332293 · Cited by 19
“Prior works generally tend to focus on error resilience of the perception stage [1], [14], [55].”
BackgroundHighly Efficient Self-checking Matrix Multiplication on Tiled AMX Accelerators
Mummidi, C. S., Ferreira, V. C., Srinivasan, S., & Kundu, S. K. (2024) · ACM Transactions on Architecture and Code Optimization · doi:10.1145/3633332 · Cited by 12
“Santos et al. studied errors due to radiations in memory and computing resources of GPUs and analyzed the reliability of CNNs [17].”
BackgroundNeutrons Sensitivity of Deep Reinforcement Learning Policies on EdgeAI Accelerators
Bodmann, P. R., Saveriano, M., Kritikakou, A. S., & Rech, P. (2024) · IEEE Transactions on Nuclear Science · doi:10.1109/tns.2024.3387087 · Cited by 11
“When a DNN is executed on a device experiencing a transient fault, its output corruption can be tolerable, if the system behavior is maintained sufficiently correct, or critical, when it causes a system failure [21], [22].”
BackgroundSelf-citationApproximate Fault-Tolerant Neural Network Systems
Traiola, M., Pappalardo, S., Piri, A., Ruospo, A., Deveautour, B., Sánchez, E., Bosio, A., Saeedi, S., Carpegna, A., Göğebakan, A. B., et al. (2024) · IEEE European Test Symposium (ETS) · doi:10.1109/ets61313.2024.10567290 · Cited by 12
“A variety of approaches have been proposed, from the most realistic neutron beam exposure [64]–[68] to combining gate-level fault simulation accuracy with software fault injection’s speed and flexibility [69]–[72].”
BackgroundEffective Fault Effects Evaluation for Permanent Faults in GPUs executing DNNs
Guerrero-Balaguera, J.-D., Condia, J. E. R., & Reorda, M. S. (2025) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3715327 · Cited by 7
“Particularly, the hardware layers physically execute the functionalities outlined in an Instruction Set Architecture (ISA).”
Story of Two GPUs: Characterizing the Resilience of Hopper H100 and Ampere A100 GPUs
Cui, S., Patke, A., Nguyen, T.-H., Ranjan, A., Chen, Z., Cao, P., Bauer, G. H., Bode, B., Martino, C. D., Jha, S., et al. (2025) · International Conference for High Performance Computing, Networking, Storage and Analysis (SC) · doi:10.1145/3712285.3759821 · Cited by 40
Understanding Logical-Shift Error Propagation in Quanvolutional Neural Networks
Vallero, M., Dri, E., Giusto, E., Montrucchio, B., & Rech, P. (2024) · IEEE Transactions on Quantum Engineering · doi:10.1109/tqe.2024.3372880 · Cited by 8
“Despite the fact that extensive research to understand and improve the reliability of traditional neural networks has been triggered already [18], [19], [20], studies about fault propagation in QNNs are still lacking.”
BackgroundMethodInfluentialSelf-citationBrandalero, M., Hernandez, H. G. M., Veleski, M., Kadi, M. A., Rech, P., & Hübner, M. (2020) · IEEE International Symposium on Parallel & Distributed Processing, Workshops and Phd Forum · doi:10.1109/ipdpsw50202.2020.00015 · Cited by 8
“The vast majority of transient faults, then, affect multiple output elements in GPUs [33], [26].”
BackgroundMethodInfluentialSelf-citationEnhancing Fault Tolerance of Neural Networks for Security-Critical ApplicationsPreprint
Alam, M., Bag, A., Roy, D. B., Jap, D., Breier, J., Bhasin, S., & Mukhopadhyay, D. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1902.04560 · Cited by 6
“ter the faults in any of its parameters in the architecture, which promotes its popularity in diverse applications. Fault tolerance of NNs has been extensively studied in the past by many researchers [10,3,6,8,4,5,7,11,12,9]. It has been demonstrated that a typical NN architecture where each neuron computes the weight…”
BackgroundStrategic Resilience Evaluation of Neural Networks Within Autonomous Vehicle Software
Schmedding, A., Schowitz, P., Zhou, X., Lu, Y., Yang, L., Alemzadeh, H., & Evgenia Smirni, A. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-68606-1_3 · Cited by 5
An Intermittent Fault Detection Method for Three- phase PFC Converters Using EMD-CNN
Liu, C., Yang, Z., Xiang, G., Yu, Y., & Tian, J. (2022) · Global Reliability and Prognostics and Health Management Conference (PHM) · doi:10.1109/phm-yantai55411.2022.9941904 · Cited by 5
Concurrent Classifier Error Detection (CCED) in Large Scale Machine Learning Systems
Reviriego, P., Wang, Z., Alonso, Á., Gao, Z., Niknia, F., Liu, S., & Lombardi, F. (2024) · IEEE Transactions on Reliability · doi:10.1109/tr.2024.3355408 · Cited by 5
“Dependability is essential when ML systems are used in safety-critical applications [8], [9], [10].”
Analyzing the Impact of Soft Errors in Deep Neural Networks on GPUsfrom Instruction Level
Ibrahim, Y., Liu, J., Yang, X., Sha, H., Li, P., & Wang, H. (2020) · WSEAS TRANSACTIONS ON SYSTEMS AND CONTROL · doi:10.37394/23203.2020.15.70 · Cited by 5
“intensively analyzed and evaluated the reliability of different DNN algorithms on differ ent GPUs [15].”
BackgroundInfluentialThe Impact of Precision Bitwidth on the Soft Error Reliability of the MobileNet Network
Abich, G., Reis, R. A. L., & Ost, L. C. (2021) · IEEE Latin American Symposium on Circuits and Systems (LASCAS) · doi:10.1109/lascas51355.2021.9667153 · Cited by 8
Unmanned Aerial Vehicle Control Using Hand Gestures and Neural Networks
Nemec, J., & Alba-Flores, R. (2022) · IEEE International IOT, Electronics and Mechatronics Conference (IEMTRONICS) · doi:10.1109/iemtronics55184.2022.9795780 · Cited by 5
Using checksum to improve the reliability of embedded convolutional neural networks
Liu, Z., Deng, Z., & Yang, X. (2022) · Microelectronics Reliability · doi:10.1016/j.microrel.2022.114666 · Cited by 5
Evaluating the impact of Permanent Faults in a GPU running a Deep Neural Network
Guerrero-Balaguera, J.-D., Galasso, L., Sierra, R. L., Sánchez, E., & Reorda, M. S. (2022) · IEEE International Test Conference in Asia (ITC-Asia) · doi:10.1109/itcasia55616.2022.00027 · Cited by 7
“In the reported literature, the HIPT method has been investigated for estimating the reliability of the GPUs with respect to transient faults, only [16]–[19].”
MethodRojas, E., Pérez, D., & Meneses, E. (2024) · Journal of Parallel and Distributed Computing · doi:10.1016/j.jpdc.2024.104879 · Cited by 5
Characterizing Soft-Error Resiliency in Arm's Ethos-U55 Embedded Machine Learning Accelerator
Tyagi, A., Jeyapaul, R., Zhou, C., Whatmough, P. N., & Zhu, Y. (2024) · IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) · doi:10.1109/ispass61541.2024.00019 · Cited by 5
“Other than works on GPUs [25], [40], [41], [85], most of the reliability analysis is carried out by making use of non-commercial ML inference accelerators.”
Understanding and Improving GPUs' Reliability Combining Beam Experiments with Fault Simulation
dos Santos, F. F., Carro, L., & Rech, P. (2023) · IEEE International Test Conference (ITC) · doi:10.1109/itc51656.2023.00034 · Cited by 7
Self-citationCan GPU performance increase faster than the code error rate?
dos Santos, F. F., & Rech, P. (2024) · The Journal of Supercomputing · doi:10.1007/s11227-024-06119-4 · Cited by 4
“On the contrary, If the additional x instructions are executed in parallel with the original sequence, the FIT rate is expected to double (same execution time, same fluence, but doubled the error rate).”
Self-citationImianosky, C., de Mattos, A. M. P., Santos, D. A., Melo, D. R., Kastriotou, M., Cazzaniga, C., & Dilillo, L. (2024) · Electronics · doi:10.3390/electronics13224461 · Cited by 7
Error Resilient Online Reinforcement Learning Using Adaptive Statistical Checks
Amarnath, C., Mejri, M., Isenberg, J., & Chatterjee, A. (2025) · IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems · doi:10.1109/tcad.2025.3529820 · Cited by 6
“Experimental tuning of checksums and radiation studies of soft error vulnerability in smaller DNNs were explored further in [22].”
Analyzing the Soft Error Reliability of Convolutional Neural Networks on Graphics Processing Unit
Adam, K., Mohd, I. I., & Ibrahim, Y. (2021) · Journal of Physics Conference Series · doi:10.1088/1742-6596/1933/1/012045 · Cited by 4
“There are several studies [15][11][16][17][18] that evaluated and analyzed the reliability of CNN models.”
BackgroundProtecting GPU's Microarchitectural Vulnerabilities via Effective Selective Hardening
Condia, J. E. R., Rech, P., dos Santos, F. F., Carrot, L., & Reorda, M. S. (2021) · IEEE International On-Line Testing Symposium (IOLTS) · doi:10.1109/iolts52814.2021.9486703 · Cited by 8
“…has been extensively This work has been partially supported by the European Commission through the Horizon 2020 RESCUE-ETN project under grant 722325 and PERIOD under the grant agreement No 886202. studying GPU reliability through fault injection/simulation [4], [5] or beam experiments [6], [7].”
BackgroundSelf-citationAnalysis and Optimization of Flute Playing and Teaching System Based on Convolutional Neural Network
Xu, C. (2022) · Mathematical Problems in Engineering · doi:10.1155/2022/1846863 · Cited by 4
Soft Error Assessment of CNN Inference Models Running on a RISC-V Processor
Gava, J., Dorneles, G., Reis, R. A. L., Garibotti, R., & Ost, L. C. (2022) · IEEE International Conference on Electronics, Circuits and Systems (ICECS) · doi:10.1109/icecs202256217.2022.9970958 · Cited by 5
“In this direction, authors have been investigated the impact of soft errors on CNN inference models or algorithms running in GPUs [1], accelerators [4], [5], FPGAs [6], [7] or processors [8]–[11].”
BackgroundTowards Reliable AI Applications via Algorithm-Based Fault Tolerance on NVDLA
Saniç, M. T., Guo, C., Leng, J., Guo, M., & Ma, W. (2022) · International Conference on Mobility, Sensing and Networking (MSN) · doi:10.1109/msn57253.2022.00120 · Cited by 8
“A recent study demonstrated that algorithmbased fault tolerance(ABFT) could maintain high protection for GEMMs in convolutional neural networks [34].”
BackgroundAlam, M., Bag, A., Roy, D. B., Jap, D., Breier, J., Bhasin, S., & Mukhopadhyay, D. (2020) · ACM Journal on Emerging Technologies in Computing Systems · doi:10.1145/3409594 · Cited by 4
“Fault tolerance of NNs has been extensively studied in the past by many researchers [dos Santos et al. 2019; Minnix 1992; Murray and Edwards 1993, 1994; Neti et al. 1992; Phatak and Koren 1995; Protzel et al. 1993; Segee and Carter 1991; Tchernev et al. 2005a, 2005b].”
BackgroundResultA cross-layer fault propagation analysis method for edge intelligence systems deployed with DNNs
Liu, T., Fu, Y., Xu, X., & Yan, W. (2021) · Journal of Systems Architecture · doi:10.1016/j.sysarc.2021.102057 · Cited by 6
Deterministic and Statistical Strategies to Protect ANNs against Fault Injection Attacks
Köylü, T. Ç., Reinbrecht, C., Hamdioui, S., & Taouil, M. (2021) · Conference on Privacy, Security and Trust · doi:10.1109/pst52912.2021.9647763 · Cited by 4
Tian, H., Ibrahim, Y., Sharma, M., Belev, G. S., Wen, S.-J., & Chen, L. (2025) · Microelectronics Reliability · doi:10.1016/j.microrel.2025.115929 · Cited by 3
Transient Fault Detection in Tensor Cores for Modern GPUs
Hafezan, M., & Atoofian, E. (2024) · ACM Transactions on Embedded Computing Systems · doi:10.1145/3687483 · Cited by 4
“Radiation testing [ 51 , 54 , 58 ] utilizes neutron beam experiments to evaluate the reliability of GPUs.”
Challenges in Assessing and Improving Deep Neural Networks’ Reliability
dos Santos, F. F., & Rech, P. (2024) · IEEE Design and Test · doi:10.1109/mdat.2024.3486658 · Cited by 3
“A vibrant research has been triggered to improve DNNs’ reliability using software strategies [7], [10], [11].”
Self-citationOzen, E., Ozerdem, O., & Orailoğlu, A. (2025) · IEEE Design and Test · doi:10.1109/mdat.2025.3530147 · Cited by 4
“While [26] explores a range of architectures and distinct models, we point out specifically the YOLO model with the Titan X and Tesla K40 architectures, on which ABFT is validated.”
Effective fault simulation of GPU’s permanent faults for reliability estimation of CNNs
Guerrero-Balaguera, J.-D., Sierra, R. L., & Reorda, M. S. (2022) · IEEE International On-Line Testing Symposium (IOLTS) · doi:10.1109/iolts56730.2022.9897823 · Cited by 8
“So far, SWIFI has been successfully applied to estimate the reliability of Neural Networks in GPU devices under transient faults, only [15], [16].”
MethodExploring the Effects of Silent Data Corruption in Distributed Deep Learning Training
Rojas, E., Perez, D., & Meneses, E. (2022) · International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD) · doi:10.1109/sbac-pad55451.2022.00013 · Cited by 5
“In [12] the authors analyzed the reliability of the GPUs running CNNs.”
BackgroundEvaluating the Impact of Transition Delay Faults in GPUs
Condia, J. E. R., & Reorda, M. S. (2023) · International Conference on VLSI Design · doi:10.1109/vlsid57277.2023.00077 · Cited by 3
“In the first case, the combination of several abstraction levels (i.e., micro-architectural and real experiments) boosts the reliability evaluation on complex applications [7], [8], [3].”
BackgroundSparseFT: Sparsity-aware Fault Tolerance for Reliable CNN Inference on GPUs
Byeon, G., Lee, S., Kim, S., Kim, Y.-J., Nair, P. J., & Hong, S. (2023) · International Conference on Parallel Architectures and Compilation Techniques (PACT) · doi:10.1109/pact58117.2023.00041 · Cited by 3
“In addition to performance and versatility, safety-critical applications also require a high degree of hardware resiliency [1].”
Toca-Díaz, Y., Palacios, R. H., Tejero, R. G., & Valero, A. (2024) · Microprocessors and Microsystems · doi:10.1016/j.micpro.2024.105023 · Cited by 4
ZOR: Zero Overhead Reliability Strategies for AI Accelerators
Vacca, E., Azimi, S., & Sterpone, L. (2024) · IEEE Interregional NEWCAS Conference (NEWCAS) · doi:10.1109/newcas58973.2024.10666350 · Cited by 3
“In contrast, adopting hardware-level fault injection allows correlating DNN errors with the accelerators [9][10].”
MethodAspis: Lightweight Neural Network Protection Against Soft Errors
Schmedding, A., Yang, L., Jog, A., & Smirni, E. (2024) · Proceedings/Proceedings - International Symposium on Software Reliability Engineering · doi:10.1109/issre62328.2024.00036 · Cited by 11
“Aspis is an efficient alternative to these existing solutions such as Structural Coding and Chipkill, specifically for reliable operation in extreme resource-constrained environments such as autonomous vehicles.”
Adam, K., Mohd, I. I., & Ibrahim, Y. (2021) · International journal of electrical and computer engineering systems · doi:10.32985/ijeces.12.2.4 · Cited by 4
“There are several studies in [17][18][19] [20][21] authors evaluated and analyzed the reliability of CNN models.”
BackgroundAnalyzing the instructions vulnerability of dense convolutional network on GPUS
Adam, K., Mohd, I. I., & Ibrahim, Y. (2021) · International Journal of Power Electronics and Drive Systems/International Journal of Electrical and Computer Engineering · doi:10.11591/ijece.v11i5.pp4481-4488 · Cited by 4
“Therefore, when the DNNs is performed in GPUs, their reliability implication is not well understood in the healthcare applications, because the errors propagate from the GPUs to DNNs (i.e., DenseNet201) [7], [12].”
Experimental evaluation of neutron-induced errors on a multicore RISC-V platform
dos Santos, F. F., Kritikakou, A. S., & Sentieys, O. (2022) · IEEE International On-Line Testing Symposium (IOLTS) · doi:10.1109/iolts56730.2022.9897448 · Cited by 2
“Our findings are aligned with previously published works that measured the criticality of the errors on GPUs and TPUs [14], [33], [34].”
BackgroundResultSelf-citationStrategic Input Selection for Deep Neural Networks Reliability Evaluation
Roquet, L., dos Santos, F. F., Kastriotou, M., & Kritikakou, A. S. (2026) · IEEE Transactions on Nuclear Science · doi:10.1109/tns.2026.3675927 · Cited by 2
“GPUs have been widely demonstrated to be highly susceptible to various types of radiation, including atmospheric neutrons [18], protons [6], and heavy ions [7], which can induce transient faults leading to soft errors that severely impact DNNs by degrading their accuracy [18].”
Self-citationConstructing Application-level GPU Error Rate Model with Neutron Irradiation Experiment
Ito, K., Itsuji, H., Uezono, T., Toba, T., Itoh, M., & Hashimoto, M. (2022) · European Conference on Radiation and Its Effects on Components and Systems · doi:10.1109/radecs55911.2022.10412577 · Cited by 2
“Fault injection is performed using SASSIFI [11] and NVBit [9], [12], [13].”
BackgroundMethodComparative analysis of soft-error sensitivity in LU decomposition algorithms on diverse GPUs
León, G., Badía, J. M., Belloch, J. A., Lindoso, A., & Entrena, L. (2024) · The Journal of Supercomputing · doi:10.1007/s11227-024-05925-0 · Cited by 3
“Most recent GPUs also include special units, such as tensor cores, to accelerate key steps in machine learning applications [13].”
InfluentialAnalyzing the Influence of Memory and Workload on the Reliability of GPUs Under Neutron Radiation
León, G., Badía, J. M., Belloch, J. A., García-Valderas, M., Lindoso, A., & Entrena, L. (2024) · IEEE Transactions on Nuclear Science · doi:10.1109/tns.2024.3387490 · Cited by 2
“Furthermore, in recent years, numerous studies have focused on evaluating the fault tolerance of neural networks on such devices, as demonstrated by research included in [26] and [27].”
BackgroundDesign Exploration of Fault-Tolerant Deep Neural Networks Using Posit Number Representation System
Yousefloo, M., & Akbari, O. (2024) · IEEE Transactions on Very Large Scale Integration (VLSI) Systems · doi:10.1109/tvlsi.2024.3396555 · Cited by 2
“In [36], it is shown that a single fault occurring in a GPU tends to propagate to multiple active threads, significantly reducing the reliability of a CNN. More-over, relying on error-correcting codes dramatically reduces the number of silent data corruptions (SDCs), while does not reduce the…”
BackgroundA Hierarchical Assessment Strategy on Soft Error Propagation in Deep Learning Controller
Liu, T., Fu, Y., Zhang, Y., & Shi, B. (2021) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1145/3394885.3431573 · Cited by 3
“The failure-masking ability brought by feature size reduction in RNN agrees with [7, 9, 10], where weight quantization and maxpooling with RELU operation in CNNs are proved to be significant in preventing SDCs from propagating to downstream layers.”
BackgroundAccurate FIT Rate Estimation Through High-Level Software Fault Injection
Bodmann, P. R., Oliveira, D., & Rech, P. (2022) · IEEE Transactions on Nuclear Science · doi:10.1109/tns.2022.3195068 · Cited by 3
Self-citationVulnerability Estimation of DNN Model Parameters with Few Fault Injections
Zhang, Y., Itsuji, H., Uezono, T., Toba, T., & Hashimoto, M. (2022) · IEICE Transactions on Fundamentals of Electronics Communications and Computer Sciences · doi:10.1587/transfun.2022vlp0004 · Cited by 2
“Hardware-level attacks include row-hammer attack [8], laser beam attack [19], radiation attack [2].”
BackgroundSoft Error Induced System Errors in Image Inference Applications with Glow-Compiled CNNs on MCUs
Witkowski, S. M., Anderson, G., & Abramowitz, P. (2024) · IEEE International Reliability Physics Symposium proceedings · doi:10.1109/irps48228.2024.10529335 · Cited by 2
Enhancing the Reliability of Split Computing Deep Neural Networks
Esposito, G., Guerrero-Balaguera, J.-D., Condia, J. E. R., Levorato, M., & Reorda, M. S. (2024) · IEEE International On-Line Testing Symposium (IOLTS) · doi:10.1109/iolts60994.2024.10616071 · Cited by 3
“In [24] it is proposed to redesign the maxpool layer of CNNs so as to alter the fault propagation.”
BackgroundOn the Measurement of Safe Fault Failure Rates in High-Performance Compute Processors
Bramley, R., Huang, Y., Duan, G., Saxena, N. R., & Racunas, P. (2020) · IEEE International Test Conference (ITC) · doi:10.1109/itc44778.2020.9325239 · Cited by 2
“Safe faults or safeness can be studied at the microarchitectural level or at the application level [2] or a mixture of both which is where we focus our work.”
BackgroundFast and Accurate Error Simulation for CNNs Against Soft Errors
Bolchini, C., Cassano, L., Miele, A., & Toschi, A. (2023) · IEEE Transactions on Computers · doi:10.1109/tc.2022.3184274 · Cited by 1
“Some of these tools, such as GPU-Qin [14], require a long profiling activity; moreover, as commented in [9], SASSIFI is the only tool”
BackgroundMethodPower and Performance Costs of Radiation-Hardened ML Inference Models Running on Edge Devices
Abich, G., Silva, A. I., Gava, J., Suzim, A., Reis, R. A. L., & Ost, L. C. (2023) · Symposium on Integrated Circuits and Systems Design · doi:10.1109/sbcci60457.2023.10261657 · Cited by 1
“The authors in [7] evaluate the soft error reliability of ML models executing on Graphic Processing Units (GPUs).”
BackgroundMethodInfluentialGava, J., Sherjil, A., Laurini, L. H., Atukpor, E., Bastos, R. P., Moraes, F., Reis, R. A. L., & Ost, L. C. (2025) · IEEE Transactions on Nuclear Science · doi:10.1109/tns.2025.3585859 · Cited by 1
“In addition, Wang et al. [9], Santos et al. [11], Gava et al. [14], and Libano et al. [18] propose the use of mitigation techniques, such as triple modular redundancy (TMR), to enhance the reliability of different CNN models.”
InfluentialGhavami, B., Shahidzadeh, M., Shannon, L., & Wilton, S. J. E. (2024) · IEEE International On-Line Testing Symposium (IOLTS) · doi:10.1109/iolts60994.2024.10616090 · Cited by 2
“Recently, Ghavami et al. [14] introduced a fine-grained activation function to improve the error resilience of DNNs. Research ([15]) emphasizes the critical need to detect errors in CNN execution to prevent their propagation to fully connected layers.”
BackgroundComparing Application-Level Hardening Techniques for Neural Networks on GPUs
Esposito, G., Guerrero-Balaguera, J.-D., Condia, J. E. R., & Reorda, M. S. (2025) · Electronics · doi:10.3390/electronics14051042 · Cited by 4
Early Reliability Assessment of AI-based Automotive Systems
Hegde, S. S., Selvaraj, D. C., Condia, J. E. R., Amati, N., Chiasserini, C. F., Deflorio, F. P., & Reorda, M. S. (2025) · ACM Transactions on Internet of Things · doi:10.1145/3758327 · Cited by 1
“To this end, it might adopt exhaustive, pseudo-exhaustive, statistical, or custom fault injection campaign approaches [14, 23, 46, 52] to identify the most vulnerable parts of an AI-based application.”
On-Hardware Resilience Analysis of DPUAccelerated CNNs on FPGA-Based Systems
Buccellato, F., Sio, C. D., Azimi, S., & Sterpone, L. (2025) · Euromicro Conference on Digital System Design (DSD) · doi:10.1109/dsd67783.2025.00017 · Cited by 2
“For example, in [19], researchers investigated the susceptibility of NVIDIA Kepler, Maxwell, and Pascal GPUs to bit-flip errors during YOLO, Faster R-CNN, and ResNet executions, uncovering the role of streaming-multiprocessor topology in error rates.”
A specification that supports FPGA devices on the TensorFlow framework
Zhao, H., Kan, H., Wang, Y., Zhao, Q., Su, D., & Huang, G. (2020) · International Conference on Electronic Information Technology and Computer Engineering (EITCE) · doi:10.1145/3443467.3443860 · Cited by 1
“However, due to the fixed architecture of the GPU [5], the instructions natively supported by the hardware are also fixed, while the FPGA is programmable and can flexibly modify the circuit for the algorithm used by itself.”
MethodInfluentialStrategies for Improving the Error Robustness of Convolutional Neural Networks
Morais, A., Barbosa, R., Lourenço, N., Cerveira, F., Lombardi, M., & Madeira, H. (2022) · IEEE International Conference on Software Quality, Reliability and Security (QRS) · doi:10.1109/qrs57517.2022.00092 · Cited by 4
“Moreover, algorithm-based fault tolerance [19], [20] has been applied to neural networks and shown to improve error robustness.”
MethodUnderstanding and Improving GPUs' Reliability Combining Beam Experiments with Fault Simulation
dos Santos, F. F., Carro, L., & Rech, P. (2023) · IEEE European Test Symposium (ETS) · doi:10.1109/ets56758.2023.10174206 · Cited by 1
Self-citationNeural network reliability analysis based on fault injection
Guo, B., Li, J., & Wang, H. (2023) · International Conference on Communication Networks and Machine Learning (CNML) · doi:10.1145/3640912.3640985 · Cited by 3
“The study revealed that CNNs had a higher occurrence of Detected Unrecoverable Error (DUE) compared to SDC. Additionally, Faster R-CNN and ResNet, which performed a higher number of operations, were found to be more susceptible to corruption compared to YOLO [5-6].”
ResultImproving GPU Reliability with Software-Managed Pipeline Parity for Error Detection and Correction
Braga, G. A., Gobatto, L. R., Gonçalves, M. M., & Azambuja, J. R. (2024) · IEEE Latin American Symposium on Circuits and Systems (LASCAS) · doi:10.1109/lascas60203.2024.10538027 · Cited by 1
“However, this efficiency also presents challenges, forcing engineers to implement fault tolerance techniques to ensure reliability in these critical applications [5] [6].”
BackgroundBadía, J. M., León, G., García-Valderas, M., Belloch, J. A., Lindoso, A., & Entrena, L. (2024) · Sustainable Computing Informatics and Systems · doi:10.1016/j.suscom.2024.101049 · Cited by 3
Range Restriction to Harden CNNs Against Hardware Faults: A Broad Empirical Analysis
Bolchini, C., Cassano, L., Miele, A., & Passarello, D. (2024) · Proceedings · doi:10.1109/dft63277.2024.10753525 · Cited by 2
“Other approaches [8], [9] define Algorithm-Based Fault Tolerance techniques specifically tailored to matrix elaborations performed within convolutions to obtain a high robustness at a reduced cost.”
Estimating the Impact of Soft Errors on AI-Based Perception in Automotive
Hegde, S. S., Selvaraj, D. C., Condia, J. E. R., Amati, N., Chiasserini, C. F., Deflorio, F. P., & Reorda, M. S. (2025) · IEEE Latin American Test Symposium (LATS) · doi:10.1109/lats65346.2025.10963959 · Cited by 1
“The static analysis focuses on identifying the most vulnerable parts (e.g., CNN layers) of an AI-based application used in an AD&SAD system resorting to a detailed pseudo-exhaustive fault injection campaign [21], [22] working on static images.”
An Optimal Design Method of Conv2d Operator for TensorFlow Based on FPGA Accelerator
Li, R., Kan, H., Su, D., Wang, Y., Zhao, H., & Tong, P. (2020) · International Conference on Computer Science and Application Engineering (CSAE) · doi:10.1145/3424978.3424987 · Cited by 1
“propose effective strategies to improve CNN reliability [5].”
Effects of thermal neutron radiation on a hardware-implemented machine learning algorithm
Trindade, M. G., Benevenuti, F., Létiche, M., Beaucour, J., Kastensmidt, F. L., & Bastos, R. P. (2020) · Microelectronics Reliability · doi:10.1016/j.microrel.2020.114022 · Cited by 1
“Artificial Neural Networks (ANNs) and Convolutional Neural Networks have been analyzed in [7, 8, 9], while [10] focused on Bayesian Machines and [11] on Binary SVM.”
BackgroundMethodReliability Analysis of Compressed CNNs
Ribes, S., Malek, A., Trancoso, P., & Sourdis, I. (2021) · Chalmers Research (Chalmers University of Technology) · Cited by 2
Quantitative Performance Analysis of BLAS Libraries on GPU Architectures
Öz, I. (2024) · Deu Muhendislik Fakultesi Fen ve Muhendislik · doi:10.21205/deufmd.2024267606 · Cited by 1
2019 Index IEEE Transactions on Reliability Vol. 68
(2019) · IEEE Transactions on Reliability · doi:10.1109/tr.2020.2965786
Impact of Layers Selective Approximation on CNNs Reliability and Performance
Rech, R. L., & Rech, P. (2020) · Proceedings · doi:10.1109/dft50435.2020.9250821
“As already shown in [7], [8], [11], approximation has the side effect of increasing the probability for a perturbation (due to a fault) to induce misdetection.”
BackgroundSelf-citationAnalysis On Deep Learning And Its Types
PRABA, A. L. (2021) · i-manager’s Journal on Software Engineering · doi:10.26634/jse.15.2.18164
Fault Mode Analysis of Neural Network-based Object Detection on GPUs with Neutron Irradiation Test
Zhang, Y., Ito, K., Itsuji, H., Uezono, T., Toba, T., & Hashimoto, M. (2020) · European Conference on Radiation and Its Effects on Components and Systems · doi:10.1109/radecs50773.2020.9857684
“In addition to beam experiments, fault injection that perturbs the weights and inputs of network is also conducted for studying the resiliency of network (e.g. [3], [4]).”
BackgroundMethodGuerrero-Balaguera, J.-D., Condia, J. E. R., & Reorda, M. S. (2022) · IEEE International Conference on Electronics, Circuits and Systems (ICECS) · doi:10.1109/icecs202256217.2022.9971036
“Unfortunately, these parallel devices may suffer of reliability issues [2], [3].”
BackgroundAbich, G., Ost, L. C., & Reis, R. A. L. (2023) · Synthesis lectures on engineering, science, and technology · doi:10.1007/978-3-031-18599-1_3
Cost-Effective Error-Mitigation for High Memory Error Rate of DNN: A Case Study on YOLOv4
Chao, W.-J., & Hsieh, T.-Y. (2023) · IEEE International Test Conference in Asia (ITC-Asia) · doi:10.1109/itc-asia58802.2023.10301172
Compiler-Managed Replication of CUDA Kernels for Reliable Execution of GPGPU Applications
Kaya, E., & Öz, I. (2024) · Journal of Circuits Systems and Computers · doi:10.1142/s0218126624502542
dos Santos, F. F., Traiola, M., & Kritikakou, A. S. (2024) · IEEE International On-Line Testing Symposium (IOLTS) · doi:10.1109/iolts60994.2024.10616094
Self-citationALPRI-FI: A Framework for Early Assessment of Hardware Fault Resiliency of DNN Accelerators
Mahmoud, K., & Nicolici, N. (2024) · Electronics · doi:10.3390/electronics13163243 · Cited by 2
Generalizing Functional Error Correction for Language and Vision-Language Models
Peng, W., Zheng, S., Baluja, M., Xie, T., Jiang, A. A., & Siegel, P. H. (2024) · International Conference on Machine Learning and Applications · doi:10.1109/icmla61862.2024.00110 · Cited by 1
“Fernandes et al. [18] focus on the errors when neural networks are running on a GPU.”
Analysis and Mitigation of Soft-errors in GPU-accelerated Hyperspectral Image Classifiers
Abed, S.-M., Guerrero-Balaguera, J.-D., Condia, J. E. R., Lucia, G. D., Lapegna, M., & Reorda, M. S. (2025) · IEEE International Symposium on Design and Diagnostics of Electronic Circuits & Systems (DDECS) · doi:10.1109/ddecs63720.2025.11006812 · Cited by 1
“The first strategy, ( Physical/beam experiments ), exposes the system to external sources (e.g., radiation or electromagnetic) to induce faults during operation, effectively assessing overall reliability against transient faults (e.g., Single Event Upsets or SEUs) [6].”
A Hardware-Aware Failure-Detection Method for GPU Control-Logic
Itsuji, H., Uezono, T., Toba, T., Ito, K., & Hashimoto, M. (2025) · IEEE Access · doi:10.1109/access.2025.3584759 · Cited by 1
“Selective hardening increases the redundancy of selected data ( e.g. , registers [7], [9], kernels [27], [28]).”
Investigating and Mitigating Critical Faults in Floating-Point and Posit Arithmetic Hardware
Condia, J. E. R., Guerrero-Balaguera, J.-D., Sierra, R. L., & Reorda, M. S. (2025) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2025.3615827
“This section describes the adoption of three coarse-grain selective hardening mechanisms to mitigate large-magnitude errors, which are critical to produce SDCs in complex applications, such as machine learning [42].”
Metadata
- Authors
- Santos, F. F. D., Pimenta, P. F., Lunardi, C., Draghetti, L., Carro, L., Kaeli, D., & Rech, P.
- Affiliations
- Universidade Federal do Rio Grande do Sul (BR)
- Northeastern University (US)
- ORCID
- Venue
- IEEE Transactions on Reliability
- Published
- 2019-06
- Volume
- 68
- Issue
- 2
- Pages
- 663-677
- ISSN
- 0018-9529 (ISSN-L)
- 1558-1721
- OpenAlex
- W2901848761
- Cited by
- 226
- References
- 67
- Citations by year
- 2026: 18 · 2025: 18 · 2024: 41 · 2023: 33 · 2022: 42 · 2021: 34 · 2020: 27 · 2019: 13
- FWCI
- 7.08
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Transactions on Reliability (publishedVersion)
- Institutional Research Information System (Università degli Studi di Trento) (submittedVersion)
- Science and Technology Facilities Council (submittedVersion)
- Topics
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Radiation Effects in Electronics
- Physical Sciences › Computer Science › Computer Networks and Communications › Distributed systems and fault tolerance
- Physical Sciences › Computer Science › Hardware and Architecture › Parallel Computing and Optimization Techniques
- Keywords
- convolutional neural networks, GPU reliability, fault injection, silent data corruption, error-correcting codes, algorithm-based fault tolerance, matrix multiplication, object detection, safety-critical systems, fault propagation, NVIDIA GPUs
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.