H214 core memory single bit error in single location mystery

[thunter0512]

One of my spare H214 core modules for my PDP-11/05 has a for me mysterious one bit fault in a single location.
The location is 035134 where bit 0 is stuck low. The entire rest of core is perfectly fine. I can very easily reproduce the fault from the front panel by depositing 177777 into that address and when I examine it see 177776.

I initially stumbled across this running "MAINDEC-11-DZMMC-A-D_NO_DUAL_ADDRESS_TEST", but a subsequent run of "MAINDEC-11-DZMMD-A-D_BASIC_MEMORY_PATTERNS_TEST" confirmed it that all the rest of core is perfectly fine except bit 0 in that single location which always reads 0.

Is there any rational explanation how this could happen except for a manufacturing fault - e.g. a missing core doughnut or missing wire in a core doughnut.
I would expect that DEC would have run diagnostics before the board was released for sale, so a manufacturing fault is unlikely. Also this board came out of a memory extension chassis, so appears to have been in use.

Another H214 tests perfectly fine in the exact same configuration (so only the H214 was swapped for another). This eliminates the possibility of the error being caused by the G231 or G110.

I know that the H214 should be tested in the original set, but as all other core doughnuts behave as expected, I think it is unlikely that a mismatched G213 & G110 is causing the one bit error in a single location.

Any thoughts, ideas or suggestions?

Am I blind to the obvious?

 

[bobaboba]

Sounds as though a single ferrite has failed/been damaged somehow.

 

[daver2]

+1.

Dave

 

[czunit]

Could be failed. If you don't care much about the board you could try increasing the write voltage or decreasing the sense current levels a bit to see if it's picked up. Are there any marginal bits on the board?

 

[czunit]

As a side note I believe I have a spare H21x board that came out of a pdp10 (so it does have the extra 2-3 mats for 18 bit operation). I think these work, if you're totally stuck we can make a deal whereby if anyone with a real KA10 or KI10 needs the 18 bit board you will swap them for a working 16 bit board they find and send to you.

 

[thunter0512]

Sounds as though a single ferrite has failed/been damaged somehow.

The ferrite doughnuts are actually quite robust compared to the fine copper wire.
The ferrites are embedded in resin. There is no visible damage to the core mat.

I cannot see how a single ferrite could fail.

I will probe the sense amplifier and compare bits 0 and 1.
I will also see if any adjustable settings make a difference.

 

[Bitly]

I cannot see how a single ferrite could fail.

A manufacturing defect coupled with repeated heating/cooling cycles could lead to a stress fracture. A cracked ferrite would shift the response curves enough to show up as a single bit failure.

CW

 

[daver2]

The wires are woven through multiple cores, so a single wire break should show up on multiple cores.

Dave

 

[gwiley]

I'm not an 11/05 expert, but you could probe the output of the sense amplifier and see if the pulse is weak or non-existent when writing and then reading that location. Need to write first so you'll have the core flip when it's read. Could toggle in a simple scope loop program. Also would be worthwhile to check the strobe timing to see how it aligns with the sense amp output pulses.
The pulse would probably look something like this: https://forum.vcfed.org/index.php?threads/restoring-a-pdp-8-l.1239827/post-1283468
It's a PDP-8/L sense amp waveform, but probably looks similar on the 11/05.