The DOI matches its Crossref record; 3 of the 3 references it cites were found in a registry or an index.
Vision Transformer Reliability Evaluation on the Coral Edge TPU
Open accessAbstract
Vision transformers (ViTs) outperform convolutional neural networks (CNNs) in tasks such as image classification, and, despite their high computational complexity, they can still be mapped to low-power EdgeAI accelerators, such as the Coral tensor processing unit (TPU). In this article, through accelerated neutron beam experiments, we study the reliability of six ViTs on the Coral TPU and four microbenchmarks. According to our data, the internal size of attention heads (the main computational block in ViTs) has negligible impact on the failure-in-time (FIT) rate of the model compared to increasing the number of heads in the model; furthermore, our results show that employing convolutions in the patch embedding reduces the FIT rate of the model. Additionally, we decompose ViTs into four basic computational blocks that represent the main operators of the model, showing that, although the transformer layer [with multihead self-attention and multilayer perceptron (MLP)] presents the highest FIT rate, it is actually the patch embedding that is more likely to cause misclassifications. These results can be leveraged to design hardening techniques that improve the resilience of the critical blocks of a ViT, identified in our evaluation while minimizing the additional overhead.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 27
- 2018, Road Vehicles—Functional Safety
- 10.1109/CVPR.2018.00745 · DOI · Paper page
- 10.1109/CVPR52688.2022.01167 · DOI · Paper page
- 10.1109/TNS.2023.3286728 · DOI
- 10.1109/TNS.2022.3142092 · DOI
- Hassani, 2021, Escaping the big data paradigm with compact transformers, arXiv:2104.05704 · arXiv
- Dosovitskiy, An image is worth 16×16 words: Transformers for image recognition at scale, 1
- Li, EfficientFormer: Vision transformers at MobileNet speed, 35, 12934
- 10.1007/978-1-4419-6993-4_3 · DOI
- 10.1109/IOLTS56730.2022.9897813 · DOI
- 10.1109/TDMR.2005.853449 · DOI · Paper page
- 10.1109/TNS.2015.2450997 · DOI
- 10.1145/1815961.1816023 · DOI
- Casey, Recent radiation test results on cots AI edge processing ASICs, 1
- 10.1109/ISCA52012.2021.00010 · DOI · Paper page
- 2023, Google Coral Edge Tpu Explained in Depth
- 10.1109/DSN-S54099.2022.00018 · DOI
- 10.23919/DATE54114.2022.9774600 · DOI
- 10.1109/TNS.2024.3359524 · DOI
- 10.1109/TNS.2024.3387087 · DOI
- 10.23919/EDHPC59100.2023.10396640 · DOI
- 10.1109/TVLSI.2023.3317138 · DOI · Paper page
- 10.1109/IOLTS59296.2023.10224882 · DOI
- 10.1109/ICCV48922.2021.00060 · DOI · Paper page
- 10.1088/1742-6596/1021/1/012037 · DOI
- Hendrycks, 2023, Gaussian error linear units (GELUs), arXiv:1606.08415 · arXiv
- Howard, 2017, MobileNets: Efficient convolutional neural networks for mobile vision applications, arXiv:1704.04861 · arXiv
Cited by
Review of large YOLOv8 and RT-DETR energy efficiency on edge devices for real-time detection
Suchý, I., & Turčaník, M. (2026) · Scientific Reports · doi:10.1038/s41598-026-46453-6 · Cited by 14
Characterization of Sub-20 nm FinFET-Based AI Chips Under Single-Event Effects
Feng, S., Liang, B., Tao, M., Yang, J., Chi, Y., Chen, J., Luo, D., Yu, G., Qiao, G., Wang, X., et al. (2026) · IEEE Transactions on Nuclear Science · doi:10.1109/tns.2026.3665046 · Cited by 1
Understanding Neural Network Fault Tolerance: From Analog Hardware to the GPU
Evans, A., Dampfhoffer, M., Burel, S., dos Santos, F. F., & Lopez, J. M. (2026) · Springer eBooks · doi:10.1007/978-3-032-17948-7_1
The Effect of Heavy Ions on Inference Performance on the IBM NorthPole
Viramontes, F. O., Vergara, V. M., Romero, A. E., Spear, M. E., Slater, W. S., & Quinn, H. (2026) · IEEE Transactions on Nuclear Science · doi:10.1109/tns.2026.3715116
Cherezova, N., Jutman, A., & Jenihhin, M. (2026) · IEEE International On-Line Testing Symposium (IOLTS) · doi:10.1109/iolts69666.2026.11633827
Metadata
- Authors
- Loureiro Coelho, B., Rafael Bodmann, P., Cavagnero, N., Frost, C., & Rech, P.
- Affiliations
- Department of Industrial Engineering, University of Trento, Trento, Italy
- Institute of Informatics, Universidade Federal do Rio Grande do Sul (UFRGS), Porto Alegre, Brazil
- Politecnico di Torino (IT)
- Rutherford Appleton Laboratory (GB)
- University of Trento (IT)
- Venue
- IEEE Transactions on Nuclear Science
- Published
- 2025-04
- Volume
- 72
- Issue
- 4
- Pages
- 1443-1451
- ISSN
- 0018-9499 (ISSN-L)
- 1558-1578
- OpenAlex
- W4405180052
- Cited by
- 5
- References
- 0
- Citations by year
- 2026: 5
- FWCI
- 0.52
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- IEEE Transactions on Nuclear Science (publishedVersion, cc-by)
- PORTO@Iris (submittedVersion, cc-by)
- Science and Technology Facilities Council (submittedVersion, cc-by)
- Funders
- RADNEXT (101008126)
- Ministero dell’Istruzione, dell’Università e della Ricerca (L.232/2016)
- Topics
- Physical Sciences › Engineering › Electrical and Electronic Engineering › Radiation Effects in Electronics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
- Keywords
- vision transformer, edge AI accelerators, reliability analysis, attention heads, patch embedding, multi-head self-attention, multilayer perceptron, hardware fault tolerance, microbenchmarks, model decomposition
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Vision Transformer Reliability Evaluation on the Coral Edge TPU. https://app.cochat.ai/papers/10.1109/tns.2024.3513774
- The paper
- Loureiro Coelho, B., Rafael Bodmann, P., Cavagnero, N., Frost, C., & Rech, P. (2025). Vision Transformer Reliability Evaluation on the Coral Edge TPU. IEEE Transactions on Nuclear Science. https://doi.org/10.1109/tns.2024.3513774
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.