Best-Fit Techniques to Estimate SBU/MCU Cross Sections From Radiation-Ground Tests in Memories

This article studies the probability distribution for the expected number of bitflips per round of reading in radiation-ground tests on a memory device where only single-bit upsets (SBUs) and multiple-cell upsets (MCUs) occur. This distribution is used to estimate the soft error cross sections in actual experiments by means of two best-fit approaches: one based on the gradient descent (GD) algorithm and the other on genetic algorithms (GAs). Besides, it is investigated how this mathematical study is suitable to detect possible variations in the soft error rate (SER) due to different reasons, such as variations in the radiation flux. Finally, the inherent stochastic characteristics of the experiments are used to provide tools to detect forgery in experiment data.