Professor Paolo Rech received the 2022 Best Paper Award from the IEEE Nuclear and Plasma Sciences Society for the paper "Selective Hardening for Neural Networks in FPGAs."
The study, published in the “IEEE Transactions on Nuclear Science” in 2019, was carried out in collaboration with the Universidade Federal do Rio Grande do Sul in Brazil, Brigham Young University in the United States, and the Rutherford Appleton Laboratory in the United Kingdom.
The study shows how it is possible to increase the reliability of neural networks - a promising solution for automatizing vehicles in the automotive, military, and aerospace markets -, thanks to the potential of programmable circuits such as the FPGAs (Field-Programmable Gate-Arrays). The latter are inexpensive, flexible, and have low power consumption. Nevertheless, they are also known to be susceptible to radiation-induced errors.
Prof. Rech and his colleagues have conducted experiments with accelerated neutron beams, thus identifying the most critical parts of the circuit and hardening them, without expensive or unnecessary modifications.
Read the paper here.