Analyzing the soft error reliability of convolutional neural networks on graphics processing unit
There have been an extensive use of Convolutional Neural Networks (CNNs) in safety critical applications. Presently, GPUs are most prominent and dominated DNN accelerators to increase the execution speed of CNN models to improve their performance as well as the La...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | en |
| Published: |
IOP Publishing
2021
|
| Subjects: | |
| Online Access: | https://umpir.ump.edu.my/id/eprint/30807/1/Analyzing%20the%20soft%20error%20reliability%20of%20convolutional%20neural%20networks.pdf https://umpir.ump.edu.my/id/eprint/30807/ https://doi.org/10.1088/1742-6596/1933/1/012045 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | There have been an extensive use of Convolutional Neural Networks (CNNs) in safety critical applications. Presently, GPUs are most prominent and dominated DNN accelerators to increase the execution speed of CNN models to improve their performance as well as the Latency. However, GPUs are prone to soft errors. These errors can impact the behaviors of the GPU dramatically. Thus, the generated fault may corrupt data values or logic operations and cause errors, such as Silent Data Corruption(SDC). unfortunately, soft errors propagate from the physical level (GPUs) to the application level (CNN model). This paper analyzes the reliability of the AlexNet model to identify which partof the model more vulnerableto the soft error. To achieve this, we injected the AlexNet run on top of NVIDIA’s GPU, using the SASSIFI fault injector as the major evaluator tool. The experiments demonstrate the most unreliable kernelsare Im2col and Add_biasand the SDC reduced from 14.5% to0.00%and 2.0% to0.10%in Im2col and Add_bias respectively. |
|---|