Sample-to-Sample Variability of Floating Gate Errors Due to Total Ionizing Dose

We studied sample-to-sample variability in the number of floating-gate errors induced by total-dose in nand Flash memories at doses and in conditions where degradation of the peripheral circuitry is negligible. Experimental data are fitted with statistical methods, using several probability distribution functions commonly used in reliability studies. The lognormal and Birnbaum-Saunders functions provide the best fit to our dataset. The accuracy of using a reduced number of samples for estimating the parameters of these two distributions is evaluated, showing that the lognormal is a more robust choice. Finally, a physical model for sample-to-sample variability of floating-gate errors is presented, which leads to a nonstandard probability distribution, amenable to neither the lognormal nor the Birnbaum-Saunders, but still involving a Gaussian function.