Logo image
Contribution à la fiabilisation d’une intelligence artificielle embarquée dédiée au traitement d’image en environnement radiatif spatial
 

Contribution à la fiabilisation d’une intelligence artificielle embarquée dédiée au traitement d’image en environnement radiatif spatial

Salma Achaq
Doctoral, École Doctorale Information, Structures, Systèmes
16/12/2025
Reliability of embedded systems Artificial neural networks Radiation effects Fiabilité des systèmes embarqués Réseaux de neurones artificiels Effets des radiations
The growing integration of embedded AI in space applications is raising new reliability concerns, especially in edge-AI hardware that can be exposed to space radiation environments. Radiation-induced Single-Event Effects (SEE), including Single-Event Upset (SEU) and Single-Event Functional Interrupt (SEFI) can induce various types of failures in these systems, potentially disrupting their functionality. This thesis aims to assess the reliability of an image-processing embedded AI system, implemented on the AMD Versal programmable SoC (one of the latest generations of programmable devices), for its use in space radiation environments. The necessary background to understand this work is first provided. The main blocks of the target device, particularly the Adaptive Intelligent Engines (AIEs) and the associated design workflow are described in detail. We next present the architecture of the AIE benchmarks we developed for being tested under laser. The experimental setup, including hardware, software, and test facility, is then presented, followed by a description of the testing methodology. Results demonstrating the laser capability to produce SEU in the 7nm FinFET node are reported followed by results on AIEs resources running our benchmarks. The work concludes with a discussion of the correlation between laser test data, simulation data and experimental heavy-ion data along with recommendations to enhance both the benchmark design and testing conditions.

(1)

url
Find in HAL
1
Logo image