Andy Holt wrote:
There are several web sites that give the details
on how Hamming Codes
(the
ECC normally used for memory) work. However, to correct some
misinformation
above:
7 check bits are only sufficient for 57 data bits - you need 8 for 64 (or
anything up to
120). This is to detect all 1- or 2- bit errors and to correct all 1-bit
errors (including those in the check bits themselves). With minor
reshuffling of the codes, if necessary, it is also possible to ensure
that
the two commonest multi-bit failures (all 0s, all 1s) are considered
to be
2-bit errors and thus no attempt to "correct" would happen.
Andy
Also at one time with small Dynamic memory ( how one defines small is a
minor point today ) one could do error checking on refesh cycles thus
keeping bit errors down to a minium.