Special Session: Reliability Assessment Recipes for DNN Accelerators

Mohammad Hasan Ahmadilivani (1) , Alberto Bosio (2) , Bastien Deveautour (2) , Fernando Fernandes dos Santos (3) , Juan David Guerrero Balaguera (4) , Maksim Jenihhin (1) , Angeliki Kritikakou (3) , Robert Limas Sierra (4) , Salvatore Pappalardo (5) , Jaan Raik (1) , Josie Esteban Rodriguez Condia (4) , Matteo Sonza Reorda (4) , Mahdi Taheri (1) , Marcello Traiola (3)

1 TTÜ - Tallinn University of Technology
2 INL - CSH - INL - Conception de Systèmes Hétérogènes
3 TARAN - Architectures matérielles spécialisées pour l’ère post loi-de-Moore
4 Polito - Politecnico di Torino = Polytechnic of Turin
5 INL - Institut des Nanotechnologies de Lyon

Reliability assessment is mandatory to guarantee the correct behavior of DNN hardware accelerators in safety-critical applications. While fault injection stands out as a well-established, practical and robust method for reliability assessment, it is still a very time-consuming process. This paper contributes with three recipes for optimizing the efficiency of the reliability assessment: a) hybrid analytical and hierarchical FI-based reliability assessment for systolic-array-based DNN accelerators; b) mixing techniques for the reliability assessment of in-chip AI accelerators in GPUs; c) reliability assessment of DNN hardware accelerators through physical fault injection. The experimental results demonstrate the efficiency of the proposed methods applied to their target DNN HW accelerator platforms.

Mots clés: deep neural networks; approximate computing; fault simulation; error emulation; reliability; resiliency assessment

Domaines: Architectures Matérielles [cs.AR]

Mohammad Hasan Ahmadilivani, Alberto Bosio, Bastien Deveautour, Fernando Fernandes dos Santos, Juan David Guerrero Balaguera, et al.. Special Session: Reliability Assessment Recipes for DNN Accelerators. VTS 2024 - IEEE VLSI Test Symposium, Apr 2024, Tempe AZ USA, United States. pp.131788 - 131828, ⟨10.1109/access.2022.3229767⟩. ⟨hal-04572731⟩