Repairing Core Memory

“Abandon all hope, ye who enter here.”  – Dante’s Devine Comedy.

I had been advised by those who know these things, that under no circumstances should you open up the core memory box.  The problem almost certainly is not going to be in there.

Except it was.

I removed the core memory from the machine, complete with the three interface cards and cable looms that connect it to the backplane of the computer.  A series of photos below – Note the R/W and Inhibit currents written on the memory box.

Case with cover removed and core stack revealed:

I first identified the faulty core which had lost continuity in one wire (broken) by checking the continuity between pins across the stack, and identified it with the pink nail polish.

I then split the stack by cutting 256 interconnecting wires to remove the faulty core, and held the spacers in the stack together with loops of wire to replace the original bolts in all four corners.

Faulty core plain containing 4096 ferrite cores:

Note factory join in the sense wire at the bottom of the core mat.  Later another factory join that ran through all 4096 of the cores broke and needed to be repaired.

Photo though Stereo Microscope lens – note Pink blob of nail polish covering a new and replacement joins in the wire, and also the brighter gold replacement wire threaded diagonally from the pink blobs through the core mat:

Note in next photo the new wire threaded between the thirteenth and fourteenth top wire starting on the identical underneath side of the core.  This was not the original broken wire, but one of several that broke during the repair process.

The initial task was to replace the first broken core wire.  It was easy to identify as the break was at the entry to the solder on the edge of the core mat. 

This involved first removing the broken wire completely from the core, and then begin the process of threading the replacement 0.1mm wire back through the 64 ferrite cores to the other side of the mat.

All of this had to be conducted under the lens of a Stereo Microscope, as it would be impossible to perform with lesser magnification.

As mentioned earlier, I used two ‘sharpened’ tweezers to assist with threading the wire through the cores.  First thing to note is that the cores are orientated in a particular direction, so it is important to feed the wire through the core when it is orientated correctly.  Even though it is held in place by other wires, sense and inhibit, it is possible to thread it incorrectly.  

Threading is a combination of pushing and pulling the wire through the cores.  The more you do it, the better you get at performing this task.  Much patience is required. Sometimes the wire wants to sneak under other wires running through the core mat, so you have to turn the core over to the other side to continue the threading process.  This will happen several times on any individual run.

Placing the surrounding wires under some strain is impossible to avoid while threading.  Some of the surrounding wires were brittle from age, and possibly from heat and / or over current at some time in the core’s history.  Consequently I broke at least three other wires during the process, all of which were subsequently replaced, and then factory join in the inhibit wire broke.

This caused me the greatest heart ache as it threads through all 4096 cores, and I couldn’t imagine the time and patience required to complete that task, or envisage how many other wires I would break in that process!  There was nothing for it but to replace a couple of lengths, and then re-join the new wire in two places.

First the good news.  The new 0.1mm wire which is covered by an insulating enamel is easy to join.  Just apply heat with the soldering iron, the enamel insulation melts exposing the copper wire, and it solders easily.

Now the bad news.  The enamel on the 50+ year old original coper wire showed no interest in melting at all – ever.  I was able to test the process on wire I had removed from the original and subsequent broken sections.

Checking around, others had attempted to remove the insulation using a solvent called Dichloromethane which is a registered substance here in New Zealand and difficult to obtain.  I do have a contact in a lab and they were able to provide me with some they had in stock.  Others had used 600 grade sandpaper to rub the ends of the wire to remove the varnish.

In the end, I deployed a combination of both methods.  I soaked the end of the wire with a cotton bud containing the DCM, and then gently applied the sandpaper.  Thankfully this worked sufficiently to allow me to solder the old with the new, and restore continuity.

With this task completed, all that remained was to restore this single core plain back into the stack, re-join the broken 256 connections, install and test.

In a perfect world, you would replace the broken stack links with fresh wire and solder them in place. Unfortunately, the way this stack is constructed, it is not possible to solder in new wires without starting at the top of the stack and working down.  The alternative was to re-join the cut wires using my trusty soldering iron, and doing my best to avoid high resistance or weak joints.

This task was sufficiently difficult to test the patience of any saint.  When heat is applied to the short wire link it melts the solder at the wire’s base and it then flops around threatening to fall out and drop into the core mat beneath it.  When you apply heat to both of the cut ends you double the problem.  I largely overcame this by using a pointed wooden barbeque stick that I could slide into the core assembly underneath the cut wires and above the core mat.  This helped to hold both ends in place while I soldered them together.  At times I had to add a ‘third wire’ as a joining wire to connect the other two together.  It was time consuming and ugly but it worked.

It is the one element of the repair that I’m not completely happy with but it is out of sight, and functional, and yes I can now store data in every memory location.

Restoring the core memory box was simply the reverse of the deinstallation process, being careful to correctly orientate the two adjacent W025 cable loom cards which I’m sure any careful engineer would have labelled Left and Right immediately after removing them from the machine.