Giordano, Marco Ferraro, Rudy Magno, Michele Danzeca, Salvatore General Purpose and Neural Network Approach for Benchmarking Microcontrollers Under Radiation 1-8 In this work a testing methodology for micro-controllers exposed to radiation is proposed. General purpose benchmarks are reviewed to provide a mean of testing all the macro-areas of a microcontroller, and a neural network benchmark is introduced as a representative class of novel computing algorithms for IoT devices. Metrics from literature are reviewed and a new metric, the Mean Energy per Unit Workload Between Failure, is introduced. It combines computing performance and energy consumption in a single unit, making it specifically useful to benchmark battery-operated edge nodes. A method to analyse reset causes is also introduced, giving important insights into failure mechanisms and potential patterns. The testing strategy has been validated on a representative class of four Cortex M0+ and Cortex M4 microcontrollers irradiated under a 200MeV proton beam with different fluences. Results from the irradiation campaign are presented and commented on to validate the benchmarks and metrics discussed. eng Protons; Performance evaluation; Radiation effects; Particle beams; Microcontrollers; Neural networks; Failure analysis; Benchmark testing; Edge computing; 2021

Imported from Invenio.