Mystery re: MCM66128L20

Tony Duell ard at
Tue Apr 10 16:57:20 CDT 2007

> I am looking for information about this part; I have something of a 
> mystery on my hands.
> At issue is a 512K IBM memory card that is part of the PC/XT 370 
> option (from 1984). 
> The card I have has 18 stacked chips comprising 512K of RAM;  the top 
> chip of one of the pairs was broken off.

That's the first thing that puzzles me. It was my understanding that 
these were 128Kbit chips, so 18 of them would give you 256K bytes (+ 
parity) of memory. Can you confirm, please, that 18 of them really are 512K

> The top chips are stamped MCM66128L20;  the bottom chips have no part number.

Mu 'guesswork', based on the PC/AT Techref, is that these are a pair of 
64K*1 DRAMs, both with slightly odd pinouts compared to a 4164. As with 
the 4164, each 64K chip has 15 active connections and one N/C, when the 
chips are piggybacked, the address lines, data in and out, write enable, 
and I think CAS/ are linked together, there are sparate RAS/ lines for 
each of the 2 chips, obtained by having the RAS and N/C pins the opposite 
way round on the 2 packages.

> I could only find the part through a broker (and I paid $25 for it).   
> I can't find a spec sheet but expected to receive a stacked part (most 
> likely, two 64K chips stacked to make a "128K" chip. 
> When the new chips arrived (after I signed munitions documents), they 
> weren't stacked pairs.  That left me to wonder whether the new parts are 
> a later rev that incorporates all the circuits into one package or 
> whether (more troublesome) the bottom chip of the pair is now a 
> complete unmarked mystery. 
> The chips were soldered together, not welded.  So . . .  I soldered 
> one of the new chips onto the top of the bottom chip.

Have you tried just a new chip in place of a piggybacked module? I wonder 
if the new chips are actually 128K bits -- that is, they're really the 2 
old chips of the pigyback on one die, in one package. If you piggybacked 
a new chip onto half an old pair, you'd end up with 2 chips 'in parallel' 
for half of the memory space, that might not matter (if you never read an 
un-wirtten-to location, there will be no contention becasue the same 
value will have been written to both chips)


More information about the cctech mailing list