top of page
RADECS25 logo.png

SESSION E Thursday, 2 October

 Radiation Effects in Complex Devices and Systems

Chairs: Luigi Dilillo & Almudena Lindoso

​

Queen Elisabeth Hall

RADECS25 logo.png

E1 09:35 - 09:50

Analysis of In-System Performance of Point-of-Load Power Converters During System-Level Irradiation​

​

Authors: I. Slipukhin(1), R. Garcia(1), F. Saigné(2), J. Boch(3), L. Dilillo(4), D. Prelipcean(1), M. Cecchetto(1), C. Cazzaniga(5), M. Kastriotou(5), S. Fiore(1), J. Lendaro(1)

1. CERN, Switzerland, 2. Université de Montpellier, France, 3. Univ Montpellier, France, 4. IES/University of Montpellier/CNRS, France, 5. STFC, United Kingdom

​

Presenting author: Ivan Slipukhin

​​

With the implementation of in-system component-level observations based on the sampling of critical voltages of the Point-of-Load power converters, the radiation-induced effects causing system-level failures were revealed and investigated during three irradiation campaigns.

RADECS25 logo.png

E2 09:50 - 10:05

Microbeam Testing of an ASIC Radiation-Hardened 32-bit RISC-V Microcontroller​

​

Authors: C. Elash(1), S. Peracchi(2), Z. Li(1), J. Xing(1), H. Lu(1), R. Fung(3), Z. Pastuovic(2), R. Drury(2), G. Martin(4), L. Chen(1)

1. University of Saskatchewan, Canada, 2. ANSTO, Australia, 3. Cisco, USA, 4. Quicklogic Corp, USA

​

Presenting author: Christopher Elash

​​

Microbeam testing was performed on a radiation-hardened ASIC 32-bit RISC-V microcontroller. Results show applied hardening techniques make the core and peripheral circuitry extremely robust. Individual scanning of memory blocks shows SEFI rates are unevenly distributed.

RADECS25 logo.png

E3 10:05 - 10:20

HARV-SoC Evaluation on SmartFusion2 and PolarFire FPGAs Under Neutron and Proton Radiation​

​

Authors: W. Grignani(1), D. Santos(2), M. Kastriotou(3), C. Cazzaniga(3), D. Melo(4), F. Wrobel(5), L. Dilillo(6)

1. IES - University of Montpellier, France, 2. IES, University of Montpellier, CNRS, France, 3. STFC, United Kingdom, 4. Univali, Brazil, 5. Universite Montpellier, France, 6. IES/University of Montpellier/CNRS, France

​

Presenting author: Wesley Grignani

​​

This work evaluates HARV-SoC under neutron irradiation, implementing and analyzing different designs on SmartFusion2 and PolarFire FPGAs. Different architectures are explored to assess fault tolerance and performance trade-offs for dependable applications.

RADECS25 logo.png

E4 11:00 - 11:15

Uncovering Fault Behavior in Arm Cortex-M Microprocessors with cFS Under Neutron Radiation​

​

Authors: L. Gobatto(1), F. Benevenuti(2), L. Laurini(3), B. Cheymol(4), R. Bastos(5), F. Lima kastensmidt(6), J. Azambuja(6)

1. Federal University of Rio Grande do Sul, Brazil, 2. Universidade Federal do Rio Grande do Sul, Brazil, 3. Univ. Grenoble Alpes, CNRS, Grenoble INP, TIMA, France, 4. Univ. Grenoble Alpes, CNRS, IN2P3, Grenoble INP, LPSC, France, 5. Univ. Grenoble Alpes / TIMA laboratory, France, 6. UFRGS, Brazil

​

Presenting author: Fernanda Lima Kastensmidt

​​

This work analyzes how cFS and cache impact the reliability of Arm Cortex-M under neutron irradiation, revealing increased software failures and I/O timeouts, often caused by hardware and software exceptions.

RADECS25 logo.png

E5 11:15 - 11:30

Self-healing architecture for RISC-V soft core processor in SRAM-based FPGAs using Dynamic Partial Reconfiguration​

​

Authors: J. Cano(1), L. Entrena(2), M. Garcia(3), A. Lindoso(4)

1. Universidad Carlos III de Madrid, Spain, 2. Universidad Carlos III, Spain, 3. Universidad Carlos 3, Spain, 4. University Carlos III Madrid, Spain

​

Presenting author: Jorge Cano

​​

This work proposes a RISC-V fault-tolerant processor architecture for SRAM-based FPGAs. The architecture aims to provide high reliability and availability using Dynamic Partial Reconfiguration, and has been validated for a commercial FPGA under proton irradiation.

RADECS25 logo.png

E6 11:30 - 11:45

Reliability Analysis of Neural Network Inference on a Fault-Tolerant RISC-V SoC with Vector Extension

​

Authors: C. Imianosky(1), D. Santos(2), W. Grignani(3), L. Dilillo(4)

1. IES - Université de Montpellier, France, 2. IES, University of Montpellier, CNRS, France, 3. IES - University of Montpellier, France, 4. IES/University of Montpellier/CNRS, France

​

Presenting author: Carolina Imianosky

​​

This work explores the deployment of quantized neural network inference on a fault-tolerant RISC-V SoC, enhancing performance with vectorized convolution and widening instructions. Through fault injection simulations, a reliability evaluation of the SoC is performed.

RADECS25 logo.png

POSTERS

RADECS25 logo.png

PE-1

Rust vs. C: Evaluating Soft Error Sensitivity​

​

Authors: A. Martinez_alvarez(1), D. Gonzalez montesoro(1), R. Bastos(2), S. Cuenca-asensi(1), A. Serrano-cases(1)
1. University of Alicante, Spain, 2. Univ. Grenoble Alpes / TIMA laboratory, France​

Presenting author: Sergio Cuenca-Asensi

​​

This paper evaluates Rust as an alternative to C for developing reliable applications in radiation-prone environments. Despite its memory safety mechanisms, Rust outperforms C in SDC cross-section under radiation, suggesting enhanced reliability in such conditions

RADECS25 logo.png

PE-2

Impact of Atmospheric Neutrons on Semantic Segmentation Correctness​

​

Authors: B. Coelho(1), M. Saveriano(1), C. Frost(2), P. Rech(1)

1. University of Trento, Italy, 2. ISIS Neutron and Muon Facility, United Kingdom

​

Presenting author: Bruno Coelho

​​

We test semantic segmentation models on TPUs and observe tolerable, mild, and critical mispredictions. We correlate errors with raw corrupted output and input image characteristics, showing that even few corrupted pixels can drastically change classifications.

RADECS25 logo.png

PE-3

Comparative Analysis of Software Redundancy on RISC-V architecture: Bare-metal vs Linux​

​

Authors: S. Cuenca-asensi(1), E. Abma romero(1), M. Ijzelendoorn(1), P. Martin_holgado(2), A. Serrano(3), A. Martinez_alvarez(1)

1. University of Alicante, Spain, 2. Centro Nacional de Aceleradores (CNA), Spain, 3. Alicante University, Spain

​

Presenting author: Sergio Cuenca-Asensi

​​

Fault-tolerance techniques were evaluated under proton irradiation on a RISC-V SoC. Results reveal consistent SDC and SEFI cross-sections across software stacks, with higher SEFI sensitivity in Linux, and emphasize cache optimizations' relevance to reliability.

RADECS25 logo.png

PE-4

A Lockstep-based Solution for a Radiation-Resilient RISC-V FPGA-Based Architecture​

​

Authors: H. Closquinet(1), F. Miller(2), L. Noizette(2), Y. Helen(3), P. Girard(4), T. Vayssade(5), A. Virazel(5)

1. LIRMM / Nucletudes, France, 2. Nucletudes, France, 3. DGA, France, 4. CNRS, France, 5. LIRMM, France

​

Presenting author: Hugo Closquinet

​​

This paper presents a RISC-V architecture that implements redundancy by Lockstep. This method ensures an efficient fault detection and correction in harsh environments while optimizing resources usage. Fault injection through emulation has validated this approach.

RADECS25 logo.png

PE-5

AI-Driven Anomaly Detection in Satellite Telemetry for Space Mission Assurance​

​

Authors: F. Buccellato(1), D. Nicolini(1), E. Vacca(1), L. Sterpone(1)

1. Politecnico di Torino, Italy

​

Presenting author: Federico Buccellato

​​

We present a data-driven anomaly detection framework for satellite telemetry, enhancing space mission assurance by identifying faults without labeled data, enabling early detection and increased reliability in dynamic orbital environments.

RADECS25 logo.png

PE-6L

Radiation Heavy-Ion Fault Analysis of URSA: a Reconfigurable Systolic Array for CNN Acceleration in SRAM-based APSoC​

​

Authors: U. Maffazioli(1), F. Limakastensmidt(1), F. Benevenuti(2)

1. UFRGS, Brazil, 2. Universidade Federal do Rio Grande do Sul, Brazil

​

Presenting author: Ulisses Maffazioli

​​

This work presents a fault-tolerant reconfigurable SA for CNN inference in SRAM-based APSoCs. We evaluate trade-offs in performance, area, and reliability. Radiation experiments validate reliability improvements from array sizing, memory configurations, hardware redundancy, and scrubbing.

RADECS25 logo.png

PE-7L

Single Event Upset Effects on Deep Neural Networks: A Model and Dataset-Level Perspective​

​

Authors: P. Trevor(1), D. Loveless(1)

1. Indiana University, USA

​

Presenting author: Peyton Trevor

​

​​This study systematically evaluates single-bit upset resilience in over 47 CNNs fine-tuned on remote sensing data, revealing per-architecture and per-sample vulnerability variation and underscoring the importance of model and input-specific robustness in radiation-prone applications.

RADECS25 logo.png

PE-8L

Reliability Assessment of Fault-Tolerant ZynqNet AI Accelerators in SRAM-Based APSoC​

​

Authors: E. Marañon(1), F. Benevenuti(1), F. Kastensmidt(1)

1. UFRGS, Brazil​

​​

This work presents a reliability and area trade-off analysis of fault-tolerant ZynqNet AI accelerator designs running a 4-layer CNN model trained on the SAT-6 dataset, validated through emulated fault injection and heavy-ion irradiation tests.

bottom of page