Having been given, some time ago, 2 8/L core memory stacks to test and then sell, I set
about and fired up my 8/L. This 8/L was bought some years ago for around 500 USD, in known
working condition. Try this today....
The memory tests revelead that one stack was nearly OK, the other was seriously sick.
The nearly OK one was 99.9% OK, but bit 0 @ address 0 was stuck. I intended to live with
that, until it was pointed out on classiccmp that address 0 is very essential indeed....
I decided to take the plunge and tried to repair the memories.
The sick module was checked out first. Result : around 20 of the select diodes were either
shorted, or open ciruit. Further tests show that the coremats themselves were also sick :
resistence of sense and inhibit wires varied widely ( normal values are around 19 ohms vor
inhibit, 24 ohms for sense wires ). Select wires were all OK.
So the stack, made by Dataram, was opened by cutting through 128 wires and separating the
upper diode board from the stack. Then the microscope was fired up and I produced some
pictures :
An overview of the coremat :
ftp://jdreesen.dyndns.org/ftp/Cores/coremat.jpg
Strange to find was a whole bunch of very small cupper wire remains, which where below
some lacquer, which means they must have been there since the beginning :
ftp://jdreesen.dyndns.org/ftp/Cores/wire_remains.jpg
A whole bunch of repairs, covered by some gunk was found. These also were made during
production. They were present only on the sense and inhibit wires :
ftp://jdreesen.dyndns.org/ftp/Cores/old_repair.jpg
Wait a minute : only on the sense and inhibit wires? which where also the wires with the
variables resistance ?
So I took out the smallest tip SMD iron I could find and started resoldering these old
repairs. See
ftp://jdreesen.dyndns.org/ftp/Cores/new_repair.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/new_repair2.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/new_repair_3.jpg
And yes, with each repair one more sense or inhibit wire was OK. Yes ! Good !
Until, with one of the last repairs, the inhibit wire broke, some few mm inside the
coremat...
ftp://jdreesen.dyndns.org/ftp/Cores/broken_inhibit.jpg
In this picture you can seen the green select wire in the middle, which no longer has an
accompaning inhibit wire.
The other core lines still have the inhibit wire, in parallel with the select line.
Major bugger, since the inhibit line is the lowest of the four wires in each core....
I attempted repair by trying to pull out the whole inihibit wire from that line of cores,
only to have it break in several more places....
Endgame for this particular stack.....
... and some weeks later i did open the other stack, made by Fabritek.
First impression was much better than the Dataram stack, no loose copper wire segments,
the pic shows it all :
ftp://jdreesen.dyndns.org/ftp/Cores/coremat_ft.jpg
Also far fewer repairs on this one, and much better organized : instead of just soldering
the wire-ends together, the wires are soldered together in some spare PCB holes that are
sprinkled about the area for that purpose.
ftp://jdreesen.dyndns.org/ftp/Cores/wire_hole.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/wire_hole2.jpg
The reason of the stack failure was soon apparant, see :
ftp://jdreesen.dyndns.org/ftp/Cores/core_broken_ov_ft.jpg
In even greater detail the broken core can be seen here :
ftp://jdreesen.dyndns.org/ftp/Cores/core_broken_ft.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/core_broken2_ft.jpg
I strongly suspect mechanical stresses are biggest in the corner, which I assume is why
this particulare core was broken. But since this core is situated in the coremat corner,
surely a repair must be possible ?
I started with cleaning out the remains of the wires that connect the coremat-pcb with the
diodeboard.
And yes it had to happen : a small shift of the soldering iron, and one select line was
broken...
ftp://jdreesen.dyndns.org/ftp/Cores/omg.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/omg2.jpg
Luckily it was outside the stack, so the wire could be repaired
ftp://jdreesen.dyndns.org/ftp/Cores/omg_repaired.jpg
Then I took some individual cores from the Dataram mat, here they are, with a small piece
of flatcable for size comparison :
ftp://jdreesen.dyndns.org/ftp/Cores/cores.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/cores2.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/cores3.jpg
This is the overal location of the broken core repair, the mat edge, with x/y select and
sense/inhibit wires.
ftp://jdreesen.dyndns.org/ftp/Cores/pcb_edge.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/pcb_edge_2.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/pcb_edge_3_ft.jpg
So first the X select line was opened, a core was inserted, and the select line soldered
in again. See :
ftp://jdreesen.dyndns.org/ftp/Cores/inhibit2.jpg
Next line was the inhibit line, the picture also clearly shows the old, broken, core
ftp://jdreesen.dyndns.org/ftp/Cores/inhinit_mounted.jpg
And the sense and Y select wires were routed through :
ftp://jdreesen.dyndns.org/ftp/Cores/repair_full.jpg
ftp://jdreesen.dyndns.org/ftp/Cores/repair_full_close.jpg
Select, sense and inhibit wires where then measured and found to be OK. I was astonished
to find that the Cu-wires on this stack were considerably less brittle than the ones of
the Dataram stack, although visually they are the same.
Pending a company move, the whole setup was now set away for a few weeks.
In the new lab I set about reassembling the stack.
Which is when disaster struck ... The new magnifying glass setup was unstable and landed
where it could cause maximal damage : the coremat itself, with around 40 cores
broken........
Major major bummer, and really only myself to blame. Not only was the stack now lost, but
I also now cannot produce evidence that the core repair worked.
So in conclusion : can core memory stacks be repaired ?
Potentially yes, but it is every bit as difficult as you would expect, and only certains
failure modes are repairable. In the 8L stack an additional difficulty is the way the
stack is build up with 128 wires to be snipped, 128 holes to be cleaned and to be rewired,
all without damaging the cores.
So i now have 2 stacks, one Fabritek, one Dataram, both opened. Mechanical differences
prevent building one good stack from the two damaged ones.
The Fabritek has two known good coremat-pcb's, the Dataram has two potentially good
coremat PCB's.
I have spent rather too much time on this, and am open to offers on these stacks, in the
condition described above.
If someone wants to sell their known-defective 8L / 8I Dataram/Fabritek corestacks, I
would also be interested.
Jos Dreesen