Partially-bad chips; was: repairing early HP calcs
frustum at pacbell.net
Mon Dec 19 22:23:43 CST 2005
Chuck Guzis wrote:
> On 12/20/2005 at 12:15 AM ard at p850ug1.demon.co.uk wrote:
> Back in the days when DRAM was precious, Intel had some "8Kx1" DRAMs that
> were really "half good" 16K parts. You used the "-x" digit to determine
> which half to use. I don't know if these were in general circulation, but
> the sales engineers were passing them out to customers working on designs.
> I may still have one or two kicking around that I found actually had 16K
> worth of usable bits, providing they weren't run too fast.
> And there was bubble memory with a flaw map.
> Aren't some (or all) high-density DRAMs now made with extra rows and colums
> to improve yields?
I recall reading during one of the DRAM crunches (1986 or 1987) that tandy
bought their own $M tester to sort rejects that they bought cheap, doing the
same thing that you describe intel doing.
Redundancy in DRAM, row and column, has been in use for 20 years at least. At
the ASIC level, largish SRAMs typically also have redundant rows and columns
(largish meaning more than a few kbits).
It is also common knowledge that 3d chips, with their massive parallelism, also
practice large granularity logic reundancy, eg the perfect chips produce 16
pixels per clock and ones with defects produce 12 pixels per clock.
Sure, bubble memories had defect lists, but so do hard disks.
More information about the cctalk