Loading
Life sciences · Preprint
arXiv · September 4, 2026
Early or partial results. Treat as a signal, not a conclusion.
This is the first system-level proton-irradiation characterization of an open-source Tensil ML accelerator on a Zynq UltraScale+ SoC executing ResNet-20 inference. The unmitigated hardware exhibited 7 workload interruptions, 2 silent output-corruption events including a 39-input stuck-class anomaly, and a spatial correlation with a 4 cm beam but not a 2 cm SoC-centered beam; however, dose–run-order confounding prevents attribution to LPDDR4 or other specific components.
Single-arm observational proton-irradiation exposure characterization. Tensil open-source ML accelerator on Zynq UltraScale+ SoC; no human subjects. Setting: laboratory proton-irradiation facility (unspecified location). Eligibility: unmitigated hardware baseline characterization.. Intervention: Proton irradiation at 20–58 MeV with fluence 4.29 × 10¹⁰ p/cm² under two beam-size geometries (4 cm nominal and 2 cm SoC-centered).. Compared with: Comparison between 4 cm beam (9 anomalies) and 2 cm SoC-centered beam (0 anomalies); no unirradiated control baseline reported..
Proton fluence delivered: 4.29 × 10¹⁰ p/cm² across 20–58 MeV beam energy range Workload interruptions: 7 events requiring 2 process restarts, 4 board resets, and 1 power cycle Silent output corruption: 2 events, including one returning absent CIFAR-10 class for 39 consecutive inputs at normal inference cadence
Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.
This work is not a clinical study and has no direct clinical application. It is relevant to aerospace and space-systems engineers designing radiation-hardened neural-network inference systems; it documents failure modes and silent corruption that must be addressed via software hardening and end-to-end content checks before deployment.
First empirical radiation-response baseline for an open-source NN accelerator under proton irradiation, documenting system-level failures and silent corruption without mitigation strategies or controlled dose–response analysis.
As stated by the source record.
Quoted from the source exactly as published.
This work is not a clinical study and has no direct clinical application. It is relevant to aerospace and space-systems engineers designing radiation-hardened neural-network inference systems; it documents failure modes and silent corruption that must be addressed via software hardening and end-to-end content checks before deployment.
Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.
As spaceborne computing systems increasingly rely on neural network (NN) accelerators, the opacity of commercial, black-box architectures severely restricts the development of verifiable radiation mitigation strategies. Open-source, register-transfer level (RTL)-accessible accelerators resolve this limitation by enabling user-defined instrumentation, yet few have empirical radiation-response baselines. This work establishes a foundational system-level proton-irradiation baseline for an unmitigated open-source Tensil NN accelerator deployed on a Zynq UltraScale+ SoC executing ResNet-20 inference. Under 20 to 58 MeV proton irradiation, we delivered $4.29 \times 10^{10}$ p/cm$^{2}$ within monitored operational windows. Seven workload interruptions required two restarts of the notebook process, four reboots or board resets, and one power-cycle sequence. Two output-corruption events returned incorrect CIFAR-10 classes without loss of service. In the longer event, the accelerator returned a class absent from the ten-image CIFAR-10 pool for 39 consecutive inputs at normal cadence. The process remained alive, while the kernel log, limited memory test, and sampled power showed no anomaly. Observation of the stuck-class sequence ended with scheduled bitstream reconfiguration. All nine onsets occurred under the nominal 4 cm beam, which exposed the SoC, LPDDR4, and additional board circuitry; none occurred under the 2 cm SoC-centered field. This pattern shows a field association but does not establish LPDDR4 as the cause because field size was confounded with run order and dose. Linux-managed accelerators require end-to-end content checks and recovery that reaches the state in which corruption can persist. This baseline documents availability loss and silent output corruption, supporting future software hardening of COTS FPGA-SoCs for neural-network inference in space systems.
Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.