Effective fault simulation of GPU’s permanent faults for reliability estimation of CNNs

Guerrero-Balaguera, Juan-David; Sierra, Robert Limas; Reorda, Matteo Sonza

doi:10.1109/iolts56730.2022.9897823

Cited by 4 publications

(7 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, the evaluation accuracy directly depends on the definition of the error functions from hardware faults. Moreover, the analyses might be limited to data-path structures in a GPU (e.g., register files, memories, and functional units), and more recently some controllers [27].…”

Section: A Methods For Reliability Assessment Of Ai Acceleratorsmentioning

confidence: 99%

Reliability and Security of AI Hardware

Gnad,

Gotthard,

Krautter

et al. 2024

2024 IEEE European Test Symposium (ETS)

View full text Add to dashboard Cite

Section: A Methods For Reliability Assessment Of Ai Acceleratorsmentioning

confidence: 99%

Reliability and Security of AI Hardware

Gnad,

Gotthard,

Krautter

et al. 2024

2024 IEEE European Test Symposium (ETS)

View full text Add to dashboard Cite

“…Unfortunately, the accuracy of the method directly depends on the targeted units (mostly data path units) and the available error models to represent faults. In literature, some works [32] addressed the software-based characterization of large workloads (CNNs) in GPUs when errors affected functional units under software error models limited to random bit-flips. Authors in [33] analyzed the impact of errors in CNNs and developed error models to represent corruptions on applications, but neglecting the fine-grain micro-architecture of the underlying hardware.…”

Section: A Motivation and Related Workmentioning

confidence: 99%

“…Authors in [33] analyzed the impact of errors in CNNs and developed error models to represent corruptions on applications, but neglecting the fine-grain micro-architecture of the underlying hardware. In [34], the authors explored a hybrid strategy to represent software errors from faults in GPU controllers. Unfortunately, their analyses were limited to a few structures with considerable evaluation times.…”

Section: A Motivation and Related Workmentioning

confidence: 99%

“…We adapt a Hardware-Injection through Program Transformation (HIPT) framework [37] (Nvbit [38]) to instrument the code of parallel programs for GPUs and inject software errors as permanent faults from the TCU's hardware. Our tool, called NvBITPERFI [34], represents errors by instrumenting the application's assembly code controlling the TCU cores (e.g., MMA) with custom functions that apply the previously identified scalar and spatial error patterns. The scalar error patterns are injected to corrupt output values to mimic the equivalent effect of a faulty TCU.…”

Section: B Assessing the Reliability Of In-chip Gpu Acceleratorsmentioning

confidence: 99%

See 1 more Smart Citation

Special Session: Reliability Assessment Recipes for DNN Accelerators

Ahmadilivani,

Bosio,

Deveautour

et al. 2024

2024 IEEE 42nd VLSI Test Symposium (VTS)

Self Cite

View full text Add to dashboard Cite

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

show abstract

“…Due to the ever-increasing trend towards having ML models embedded in edge safety-critical systems, researchers have started to investigate the impact of radiation-induced soft errors on the reliability of underlying models considering simulation (e.g., [22][23] [24][25] [26][27] [28][29]), emulation (e.g., [6][14] [30]) and radiation tests (see Table I). While high-level simulation and emulation approaches are highly useful to conduct early soft error assessment and eliminate not suitable options, final systems' configurations must be evaluated through radiation tests.…”

Section: Related Work In Machine Learning Soft Error Assessment and M...mentioning

confidence: 99%

A Lightweight Mitigation Technique for Resource- Constrained Devices Executing DNN Inference Models Under Neutron Radiation

Gava

Hanneman

Abich

et al. 2023

IEEE Trans. Nucl. Sci.

View full text Add to dashboard Cite

Deep neural network (DNN) models are being deployed in safety-critical embedded devices for object identification, recognition, and even trajectory prediction. Optimised versions of such models, in particular the convolutional ones, are becoming increasingly common in resource-constrained edge-computing devices (e.g., sensors, drones), which typically rely on reduced memory footprint, low power budget and low-performance microprocessors. DNN models are prone to radiation-induced soft errors, and tackling their occurrence in resource-constrained devices is a mandatory and substantial challenge. While traditional replication-based soft error mitigation techniques will likely account for a reasonable performance penalty, hardware solutions are even more costly. To undertake this almost contradictory challenge, this work evaluates the efficiency of a lightweight software-based mitigation technique, called Register Allocation Technique (RAT), when applied to a convolutional neural network (CNN) model running on two commercial Arm microprocessors (i.e., Cortex-M4 and M7) under the effects of neutron radiation. Gathered results obtained from two neutron radiation campaigns suggest that RAT can reduce the number of critical faults in the CNN model running on both Arm Cortex-M microprocessors. Results also suggest that the SDC FIT rate of the RAT-hardened CNN model can be reduced in up to 83% with a runtime overhead of 32%.

show abstract

Effective fault simulation of GPU’s permanent faults for reliability estimation of CNNs

Cited by 4 publications

References 16 publications

Reliability and Security of AI Hardware

Reliability and Security of AI Hardware

Special Session: Reliability Assessment Recipes for DNN Accelerators

A Lightweight Mitigation Technique for Resource- Constrained Devices Executing DNN Inference Models Under Neutron Radiation

Contact Info

Product

Resources

About