Marta Bagatin1, Simone Gerardin1, Alessandra Bertoldo1, Alessandro Paccagnella1, Véronique Ferlet-Cavrois2, Alessandra Costantino2, Michele Muschitiello2, Pierre-Xiao Wang3
1 University of Padova. Department of Information Engineering, Padova, Italy
2 ESA ESTEC, Noordwijk, The Netherlands
3 3D PLUS, Bus Cedex, France
The impact of sample-to-sample variability on the number of bit errors induced by total ionizing dose in 25-nm Single Level Cell (SLC) NAND Flash memories was experimentally studied using gamma rays from a Co60 source. Low total ionizing doses were considered, in such a way that degradation of the peripheral circuity was negligible and the errors were mainly due to charge loss from the floating gate cells, and to a much smaller extent to charge trapping in the oxides surrounding the storage element. Two production lots, for a total of more than 1 TBit of cells, were analyzed from a statistical point of view and the results were fitted with different distributions, in order to find the best match. The lognormal and Birnbaum-Saunders functions appeared to offer the best fit. However, physical modeling shows that neither of them is strictly justifiable and that a more complex form is needed to correctly capture the initial Gaussian distribution of cell threshold voltage and the shift induced by total dose. Implications of the choice of the statistical distribution on the number of samples needed to correctly estimate the number of floating gate errors as a function of dose are discussed.