Empirical mathematical model of microprocessor sensitivity and early prediction to proton and neutron radiation-induced soft errors

Abstract

A mathematical model is described to predict microprocessor fault tolerance under radiation. The model is empirically trained by combining data from simulated fault-injection campaigns and radiation experiments, both with protons (at the National Center of Accelerators (CNA) facilities, Seville, Spain) and neutrons [at the Los Alamos Neutron Science Center (LANSCE) Weapons Neutron Research Facility at Los Alamos, USA]. The sensitivity to soft errors of different blocks of commercial processors is identified to estimate the reliability of a set of programs that had previously been optimized, hardened, or both. The results showed a standard error under 0.1, in the case of the Advanced RISC Machines (ARM) processor, and 0.12, in the case of the MSP430 microcontroller.