The Impact of Allocating Weights and Biases in SRAM Parity Checking Areas on the Soft Error Reliability of CNN Inference Models
This paper promotes a novel semi-automated compiler-based technique that allocates CNN weights and bias to SRAM parity checking areas. Radiation results suggest that it is possible to reduce critical SDC events by up to 90%.
Authors
- Rodrigo Possamai Bastos (ORCID: https://orcid.org/0000-0002-9964-0424)
- Jonas Gava (ORCID: https://orcid.org/0000-0001-7113-6448)
- Luiz Henrique Laurini (ORCID: https://orcid.org/0009-0008-9562-1446)
Institutions
- Institut polytechnique de Grenoble (FR)
- Centre National de la Recherche Scientifique (FR)
- Universidade Federal do Rio Grande do Sul (BR)
- Loughborough University (GB)
- Techniques of Informatics and Microelectronics for Integrated Systems Architecture (FR)
- Pontifícia Universidade Católica do Rio Grande do Sul (BR)
- Université Grenoble Alpes (FR)
Publication Details
- Published
- 2024-09-16
- DOI
- https://doi.org/10.1109/radecs61970.2024.11298584
- Primary Topic
- Radiation Effects in Electronics
- Type
- article
- Field-Weighted Citation Impact
- 0.00