We have compared the data retention of irradiated commercial NAND flash memories with that of unirradiated controls. Under some circumstances, radiation exposure has a significant effect on the retention of flash memories. Introduction. Previously, we compared the endurance reliability of irradiated and unirradiated NAND flash memories, and found no significant difference [1]. An endurance failure would have required enough radiation-induced defects in the tunnel oxide to induce a significant shift in V T . But the unhardened commercial parts used as test samples could be irradiated only to modest doses before they failed for other reasons. It turned out that there would not have been enough radiation-induced defects to cause an endurance failure, until after radiation had caused other failures. However, there was a concern that retention might be more sensitive to radiation, because a retention failure requires only a very low current leakage path. Such a leakage path really only requires a very small number of radiation-induced defects, if they are properly aligned. Therefore, we have compared the retention failure rates for irradiated and unirradiated samples. We find that, under some circumstances, radiation exposure can cause a significant increase in the rate of retention failures in flash memories. We note, however, that flash manufacturers typically specify the endurance and retention characteristics of their products, assuming that error correction software will be used. However, in this test, we did not use error correction, because we were trying to characterize the underlying technology, and not the effectiveness of the error correction. It is likely that, if we had used error correction, the parts would still have met all their reliability specifications.
Description of Samples.The samples used in this study are 8G NAND flash memories from Samsung Semiconductor (part number K9F8G08U0A, LDC xxx), and from Micron Semiconductor (part number MT29F8G08ABABA, LDC ###). Both have 4K blocks, with 64 pages per block, with 4Kx8 page organization, plus 64 redundant columns. Each uses a nominal 3.3 V power supply (2.7-3.6 V, full range). The Samsung parts are intended to operate over the industrial temperature range, -40 to +85º C, while the Micron parts are intended for the commercial temperature range, 0-70°C. All NAND flash products typically have a few bad blocks, which have to be screened out. For both manufacturers, the specification is <80 of the 4096 blocks will be bad, but in our experience, a single digit number is more typical. Both the write (Programming) and Erase operations proceed by Fowler-Nordheim tunneling of electrons through the tunnel oxide. Fowler-Nordheim (F-N) injection requires very high fields, and the operation of a charge pump circuit to step up the power supply voltage. F-N injection also introduces damage into the tunnel oxide, contributing to wear-out. It is for this reason, that manufacturers typically guarantee flash memory only for 10 5 (P/E) cycles. This stress-induced...