Implementation and Reliability Evaluation of a RISC-V Vector Extension Unit - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

Implementation and Reliability Evaluation of a RISC-V Vector Extension Unit

Résumé

The space environment challenges the reliable functioning of electronic systems due to radiation exposure and extreme temperatures. To mitigate potential failures, fault tolerance techniques are crucial in these systems. With the implementation of fault tolerance techniques, the performance of the system is affected, often impairing data processing. One way to accelerate data processing is to use vector instructions, which allow operating on a data vector with only one instruction. Therefore, we developed a Vector Extension Unit that supports a subset of the RISC-V Vector Extension for HARV, a RISC-V-based processor. We evaluated the impact of using the Vector Extension instructions over the scalar instructions. We assessed the reliability of the core by simulation based on fault injection campaigns with both baseline and hardened HARV. The vector unit offers substantial performance acceleration of up to 8.1 times when compared to scalar operations, and, despite the hardware overhead, it enhanced the reliability of the executions from 93.9% to 97.9% for the baseline HARV and from 97.9% to 99.3% for the hardened HARV.
Fichier principal
Vignette du fichier
2023___DTFS___Implementation_and_Evaluation_of_a_Vector_extension_in_RISC_V_processor___HAL_Version.pdf (426.48 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04266888 , version 1 (31-10-2023)

Identifiants

Citer

Carolina Imianosky, Douglas Santos, Douglas Melo, Felipe Viel, Luigi Dilillo. Implementation and Reliability Evaluation of a RISC-V Vector Extension Unit. DFT 2023 - 36th IEEE International Symposium on Defect and Fault Tolerance in VLSI and Nanotechnology Systems, Oct 2023, Juan-Les-Pins, France. pp.1-6, ⟨10.1109/DFT59622.2023.10313569⟩. ⟨hal-04266888⟩
42 Consultations
110 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More