Nand technology has some issues, which I find it a bit difficult to understand. It is said in nand devices that manufacture guaranty that a page will have no more than X (4/8/16,etc according to nand device) failed bit.

  1. Is it that the manufacture can really guarantee that there shall never be more than this amount of failed bits is a single page ?
  2. Is it that the failed ecc happens just randomly, after some time ?
  3. Is it that re-program the failed ecc page, shall recover its flip bits to the correct value ?

Thank you, Ran

  1. No, it's all a game of probabilities. You just define a probability (e.g. \$1-10^{-14}\$) that all pages fulfill that criterion, the manufacturer models the stochastic process that leads to errors and makes observations of that process, and you so get to a probability that indeed your criterion is fulfilled.
  2. There's different kind of errors. But yes, spontaneous bit flips can happen for thermal, particle / ionizing radiation, or current leakage reasons. Other errors are not spontaneous, but always immediately manifest. Note that a flipped bit doesn't mean the ECC must fail: Error checks are only safe against certain patterns of bit errors, and if multiple bits within a block (or page, or whatever you call it) fail, the ECC might not "notice" it. It's, again, a stochastic thing.
  3. Only if
    1. some recovery mechanism recovered the original data correctly; again, patterns, probability, and
    2. the reason the page failed because a bit flipped randomly. Of course, if a bit cell is simply stuck to a value, then there's nothing rewriting could solve.
| improve this answer | |
  • \$\begingroup\$ I don't see the probability value in datasheet, so for example for a device such as MT29F4G08ABBDAH4, which is 4-biit ecc, the probability should be really low as you've described (e.g. 10^-14) ? \$\endgroup\$ – ransh Dec 21 '17 at 9:30
  • \$\begingroup\$ and another question, does oob (the additional info per page) is guaranteed to be correct ? Thanks! \$\endgroup\$ – ransh Dec 21 '17 at 10:24
  • \$\begingroup\$ you'll have to take a bitter pill: there's no certainty outside of math. Every device, every aspect of every device, just has an error probability. In good cases, this probability is close to 0. \$\endgroup\$ – Marcus Müller Dec 21 '17 at 12:08
  • \$\begingroup\$ That I understand, I just don't see why they don't write the probablity in datasheet. Myabe they mean that 5% there won't be error above 4-bit per page ? 5% is much more than 10^-14 \$\endgroup\$ – ransh Dec 21 '17 at 12:10
  • \$\begingroup\$ The y will say whatever they can, in the end, financially afford to disappoint a customer. When you buy memory modules for places where "nearly all work" isn't good enough, you call them, and get more expensive chips with a specific probability. \$\endgroup\$ – Marcus Müller Dec 21 '17 at 14:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.