DESCRIPTION :
The deployment of specialized hardware accelerators has become a cornerstone of modern computing platforms. Devices ranging from graphics and tensor processing units (GPUs and TPUs) to emerging processing-in-memory (PIM) architectures are now used to efficiently execute increasingly data-intensive workloads, particularly for large-scale AI models whose parameters reach billions to trillions. Unlike traditional CPU-centric designs, modern accelerators provide massively parallel compute capabilities and reduce data movement overheads, enabling onboard processing for domains such as autonomous robotics, remote sensing, and complex signal processing.
This trend is particularly visible in safety- and mission-critical domains. Small satellites increasingly integrate COTS SoCs with AI accelerators for on-board inference and radar processing, while automotive, avionics, and medical systems rely on accelerators for perception, decision-making, and classification. In these contexts, strict correctness and timing constraints, combined with harsh conditions such as radiation, voltage and temperature variations, and long operational lifetimes, impose strong reliability and security requirements on compute-dense accelerators.
As these workloads become both memory-bound and specialized accelerator-dominated, evaluating how faults propagate through the hardware and into algorithmic layers, potentially corrupting neural network outputs, control decisions, or mission-critical computations, has become a central research challenge. Understanding these failure modes is essential not only for safety but also for security, as similar fault mechanisms can be exploited to launch fault attacks or cause integrity violations in accelerator-based computing systems.
In this context, this postdoctoral position is centered on driving decisive advances in the reliability and security of modern hardware accelerators, with the expectation of producing robust, high-impact results that shape both research and practice in safety-critical computing systems. The postdoctoral researcher will work closely with the team and benefit from strong guidance. The candidate is expected to contribute to the conception and writing of innovative, forward-looking research projects, offering a valuable opportunity to help shape the field's research agenda.
Recent research directions within the team focus on reliability analysis and fault protection for Artificial Intelligence (AI) inference. Specifically, we investigate radiation experiments [1,3,4], cross-layer approaches [1,10], strategies for minimizing fault-injection time [8,9], and selective protection methods [1,5,8]. Additionally, we are exploring the relationship between approximate computing and reliability [6,7,11,12,13].
The postdoctoral researcher will be encouraged to take an active role in one or more of these research directions, advancing them while building a unifying research line that connects them.
Team reference publications
[1] "Cross-Layer Reliability Evaluation and Efficient Hardening of Large Vision Transformers Models", Lucas Roquet, Fernando Fernandes dos Santos, Paolo Rech, Marcello Traiola, Olivier Sentieys, Angeliki Kritikakou, Design Automation Conference (DAC), 2024
[2] "Improving Deep Neural Network Reliability via Transient-Fault-Aware Design and Training", Fernando Fernandes dos Santos, Niccolò Cavagnero, Marco Ciccone, Giuseppe Averta, Angeliki Kritikakou, Olivier Sentieys, Paolo Rech, Tatiana Tommasi, IEEE Transactions on Emerging Topics in Computing, 2024
[3] "Impact of Radiation-Induced Effects on Embedded GPUs Executing Large Machine Learning Models", Bruno Loureiro Coelho, Fernando Fernandes dos Santos, Matteo Saveriano, Gregory Allen, Andrew Daniel, Steven Guertin, Sergeh Vartania, Edward Wyrwas, Christopher Frost, Paolo Rech, IEEE Transactions on Nuclear Science, 2025
[4] "Impact of High-Level-Synthesis on Reliability of Artificial Neural Network Hardware Accelerators", Marcello Traiola, Fernando Fernandes dos Santos, Paolo Rech, Carlo Cazzaniga, Olivier Sentieys, Angeliki Kritikakou, IEEE Transactions on Nuclear Science, 2024,
[5] "HTAG-eNN: Hardening Technique with AND Gates for Embedded Neural Networks", Wilfread Guilleme, Angeliki Kritikakou, Youri Helen, Cédric Killian, Daniel Chillet, Design Automation Conference (DAC), Jun 2024, The postdoc will contribute to the development of models, methodologies, and tools to understand and improve the reliability and security of accelerator-based computing platforms. Possible directions include (but are not limited to):
* evaluation of transient, permanent, and aging-related faults in accelerators
* cross-layer analysis of fault propagation from hardware to algorithm
* hardening strategies at the architecture and algorithm levels
* secure-by-design accelerator architectures (e.g., isolation, memory protection)
* interactions between fault tolerance and security
* defenses against fault attacks, malicious perturbations, or integrity violations
* reliability evaluation for emerging accelerators (e.g., TPUs, PIM, nonvolatile memories, RISC-V)
* modeling of failure modes under radiation, voltage/temperature stress, or long-duration operation
Both theoretical and experimental contributions are welcome, depending on the candidate's expertise., This is a research-focused academic position, not an engineering or support role. Expected duties include:
* conducting original research and publishing in leading venues
* supervising and mentoring PhD students and Master's interns
* participating in European and National research projects, including:
* research contributions in collaborative work packages
* project reporting and deliverables
* contribution to project design and writing for future calls
interaction with industrial and academic partners
presenting results at international conferences and consortium meetings
The position offers opportunities to work with simulation frameworks, accelerator testbeds, and radiation-testing facilities, depending on the research direction.
Code d'emploi : Chargé de Recherches (h/f)
Domaine professionnel actuel : Scientifiques
Temps partiel / Temps plein : Plein temps
Type de contrat : Contrat à durée indéterminée (CDI)
Compétences : Intelligence Artificielle, Plates-Formes Informatiques, Réseaux de Neurones Artificiels, ETL, Tolérance aux Pannes, Boîte Noire Transactionnelle, Machine Learning, Reduced Instruction Set Computing, Traitement de Signal, Graphics Processing Unit (GPU), Fiabilité des Systèmes, Hardware Acceleration, Prise de Décision, Fiabilité, Esprit d'Équipe, Innovation, Algorithmes, Architecture, Conception Architecturale, Systèmes Automatisés, Industrie Automobile, Avionique, Organisation d'Événements, Expérimentation, Diagnostique des Causes de Défaillance, Gestion de Projet, Industrie Nucléaire, Recherche Post-Doctorale, Télédétection, Conception et Réalisation en Robotique, Communication Scientifique, Transformateurs (Électrique), Analyse des Besoins de Sécurité, Simulations, Voltage, Vie des Systèmes Critiques, Architecture Matérielle, Coaching, Périmètre de Projet, Commercial Off-The-Shelf, Publication / Edition, Applications des Règles et Consignes de Sécurité
Courriel :
dos-santos@inria.fr
Téléphone :
0299847100
Type d'annonceur : Employeur direct