I'm reasonably happy with the quality of the baseline data collected
so far and I think I can now proceed towards actually trying some
ideas to make the corruption go away for good.

I'll try next to add a 4.7 kOhm pull-up to FLASH_WE_N/WE# (write
enable) and see if this has any effect on NOR corruption. If it
doesn't, then I'll also add a pull-up to FLASH_CE_N/CE0 (chip
enable).

These pull-ups were discussed here:
http://lists.milkymist.org/pipermail/devel-milkymist.org/2011-October/001902.html

The statistical analysis so far allows us to estimate what amount of
testing should be sufficient to encounter NOR corruption with a
certain probability.

Or in other words, we can calculate how many corruption-free cycles
are needed to be able to claim with a certain confidence that NOR
corruption has vanished:

Test cycles     Probability of not encountering NOR corruption
                (even though it is still there)
  100           82%
 1150           10%
 2300            1%
 3450            0.1%
10000            0.0000002%

(This assumes exponential distribution, lambda = 1/500.)

I'm still not sure whether there may also be some environmental
parameter (e.g., temperature) that influences how often NOR
corruption happens, if at all. Since controling and monitoring
temperature would be harder than I had expected, I'll ignore this
potential issue for now, and consider revisiting it later, when we
have a likely fix.

- Werner
_______________________________________________
http://lists.milkymist.org/listinfo.cgi/devel-milkymist.org
IRC: #milkymist@Freenode

Reply via email to