Can We Effectively Correct Errors in Computer Memory?

[Loren Booda]

Computer memory has steadily grown larger, although more important may be guarding against environmental errors for its posterity.

If we save three copies of memory rather than just one, to what degree (in % of memory bits) could subsequent errors (at most one out of three per position among memories) be corrected effectively?

Also, does my PC have such a backup?

 

[0xDEADBEEF]

No your PC does not have such a backup. There is a trick that is a little more elegant, where there are so called parity bits. For every byte of 8 bits there is a ninth bit that contains the logical xor of all the bits. If a bit gets damaged then this is detected immediately and the memory location can be marked as bad. This has been removed a long time ago for cost reasons. So three times the memory is out of the question.

Your scheme will correct bit flips unless two occur in the same triplet. So you define some failure rate that you are willing to accept and calculate what probability of a single bit flip you can allow for, such that the chance of two in one triplet are that high.

There are much more elegant error correction schemes. Look for "cyclic redundancy check", and "Cross-interleaved Reed-Solomon coding" (this is done on CDs)

 

[russ_watters]

Anything important is typically stored on your hard drive, so no such backup is needed. Afterall, what use is a volatile backup when a power failure can wipe it out?

 

[mgb_phys]

See "error corrected ECC memory"
As you get more memory the number of errors goes up (1 in a million events affect 4billion bytes quite often) and as the feature size gets smaller the chance of an error goes up (less energy is needed in a cosmic ray/radioactive decay to flip a bit)

http://en.wikipedia.org/wiki/Dynamic_random_access_memory#Errors_and_error_correction