Probability aware fault-injection approach for SER estimation

The Soft-Error Rate (SER) estimation is used to predict how electronic systems will respond to the transient electrical pulses induced by the ionizing radiation. SER estimation by radiation test is an accurate method, but it is expensive and requires the real device. Traditional simulation methods incorporate logical, temporal and electrical masking effects while injecting faults at the output of the device's functional elements. Nevertheless, they do not consider the probability of the ionizing radiation to produce a transient fault at the output of each class of functional element. On the other hand, studies in the stochastic computing domain deal with a probabilistic fault-injection approach. Since many concomitant faults among the elements may occur, the fault probability of each element is treated independently. This leads to the use of one Pseudo-Random Number Generator (PRNG) and a probability comparator for each functional element. However, the analysis of a single fault is usually enough for SER estimation. In this context, this work presents a different approach for probability-aware fault-injection, in which a weighted distribution of faults is defined considering the relative fault probability of each functional element. This approach enables the use of just one PRNG and a decoder for the entire device, instead of a pair 'PRNG-comparator' per element, leading to a significant reduction in logic blocks consumption. For the example analyzed in this study, the use of relative fault probability decreases the number of logic blocks from 875 (adopting independent fault probability) to 495.