Friday, September 12, 2014

Core memory issue - methodical search for causes and some improvement - plus other progress on restoration project


My time away gave me a chance to work out a set of tests that will draw ever shrinking boundaries around misbehaving sections of the machine until I an isolate the part or parts that are at fault. I realized that some additional failures in one component could cause the failures spread across independent compartments, core stacks, 4K halves and cards. For example, any failure in selection or addressing could be activating more than one section of core at the same time.

Looking up the pins in gates B and D that are on either end of the connecting cables
Before I applied power again, I did a set of continuity tests for signals that should be traveling over the cables linking the B gate of the 1131 with the D memory gate. For example, the sense output of a a compartment of core for a given bit travels over the cable and into the B register card in B gate. If I check the input pin on the B register card and the output pin on the memory sense card for the bit, I should have continuity.

Continuity test of signal flow between B gate and D gate
Any open circuit would deliver a constant value of 1 for the bit and indicates a problem somewhere between the pair of pins. It can be a problem with the backplane traces, wirewrap leads, the cable connectors or the cable itself. Zooming in on the path would identify which of those failures occurred.

I tested sense bit 3, which was working correctly when I last had the machine operating, also bit 5 which is stuck on and bit 11 which is part of the low halfword that is now read as all ones. Bit 3 had connectivity as expected. However, so did bits 5 and 11. I turned to the inhibit lines from the B register into the memory and those also had continuity for bits 3, 5 and 11.

There was still a question of continuity for key timing and control signals, plus the memory address bit signals that run between B and D gates. Those came next. I checked the control and timing signals plus address bits 2, 3 and 15, all of which had good continuity.

Next up is to scope the signals on a good and a bad bit - looking at what comes out of both sides of the B registers for bits 3, 5 and 11. If these are not inverts of each other, I have bad B register cards. If the waveforms are unusual compared to the write cycle signal, I have a problem deeper in the B gate.

Probes added to watch signals on the oscilloscope
CE switches set to Storage Load function
I powered up, put the system in CE Storage Load mode which would cycle through memory repeatedly storing the value I set in the front panel switches. However, it goes up to address 0x00FF then sits with that displayed, not responding to the immediate stop button nor reset. Reset won't work while the machine is running, thus the issue to debug involves what is wrong that won't allow Imm Stop to work but will stop if I move the rotary mode switch up to SI temporarily.

Rotary mode switch
Until I can fix the funny problem with moving beyond IAR of 255 decimal, I won't have the steadily repeating signal that allows me to scope this on the Tektronix 7000 series machine. I see that Storage Write continues to cycle on and off when the machine is hung, which is another clue. The logic of the Imm Stop key is that it will only activate when the machine reaches T5 or X5 cycle, but when it is hung up it is obviously not advancing through that state. However, scoping the T cycle signals show it does move through the steps.

I started at the core of the machine, observing the oscillator, phase A and phase B signals, clock advance pulses, and then several of the T cycle signals (e.g. T5 and T7). All is occurring as it should with those signals.

I didn't yet see why the Imm Stop was not working, but did more investigations. After scoping signals involved with the function, I came to a NAND gate that was not activating although both its inputs were clearly going high. This was on card B-A1C6 which was a hex NAND gate card - 5800000 - of which there are quite a few in the 1130 system.

An SLT card is roughly like one small scale integration IC - a 74xx series chip - although slightly more logic sits on one than on the early 74xx series chips since the 7400 is a quad 2 input NAND gate chip while the 5800000 card has six gates aboard taking only a fraction of a standard single card.

I was ready to pull and swap the cards at B-A1C6 and B-A1M7 when I noticed that the C6 card was slightly askew. Touching it, I realized it wasn't seated properly in the socket. After snapping it into place, I tested the machine which now properly implemented the IMM Stop function.

Similarly, I scoped some lines until I determined that a particular card in the B-B1 compartment was blocking the incrementing of IAR addresses from bit 8 up to bit 7 (words in the 1130 start with the high order bit numbered 0 (32768) and the low order bit position (the ones place) is bit 15. That card was also a bit askew, less obvious until I wiggled it gently, but once snapped into place, the machine was now processing the CE Storage Load through all 16K words.

Bit switches to set the word value to be stored in memory during Storage Load
Two problems down and one to go - that the low half word being entered into memory is reading as xFF but the switches were set to x00. I stopped at this point, with the sun setting, closed up the garage door and went inside to set up my next set of scope points. So far, this technique is working well and will let me shoot problems with a steady state to a signal. Those that are not continuous will need the storage scope, my Tek 465, but I am not yet at that point.

Diagnosing the problems is easy for me because the process of building a replica 1130, matching the real 1130 gate by gate from the ALDs, has given me a very solid understanding of all the signals and how the machine operates at a deep level. I know that the I register (IAR) is incremented by special gates built into the register, not using the adder circuits, and remember where in the ALDs the circuits are documented.

The only details I don't know about the machine are the specific SLT cards on which a given gate is implemented and other details that are specific to SLT technology, since I implemented the logic using fpga instead. Thus, it is a pretty fast learning curve to go from knowing the logical 1130 hardware to knowing the physical instantiation by card, compartment, gate and pins.

The next tests will continue on the B gate side, tracking down the case for the xFF values. Once I find and resolve that issue, I will have validated the correctness of what I send to memory for writing.  If all is good on the B gate side of the value for writing, I then will look at the received bits coming back, to see whether the 1 and 5 bit problem is gone or is still occurring. All of this will occur tomorrow.


I brought back a rear cover, the small cover that goes in front of the blank card hopper, and a spare rear leg, which are the three missing or damaged items on my machine. I have already installed both covers including the new sound deadening foam for the small cover. At some point I will install the replacement rear leg.

Replacement cover for blank card hopper, dirty and color mismatch
When I last left, backspace didn't seem to be working but it just took a bit of patient exhortations before the last of the solidified grease was worked out of certain parts. Now it works well. I am not satisfied yet with the operation of the pressure rollers that register a card in the read station and eject it all the way out into the stacker area. This too is undoubtedly a lingering grease issue.

There is a fair degree of gray aluminum surface corrosion that I want to remove if practical. A product called Nevr-Dull was recommended by someone, which I bought and applied to the stacker card weight and some other areas of the machine. It does remove some corrosion, but it takes quite a bit of effort to really make progress.

Surface corrosion on aluminum parts
I wipe the wadded cloth and solution on a section of aluminum, then wipe off the black crud that is released and polish it a bit. That seems to take off a microscopically thin layer at a time, as I have to repeat this many times to see noticeable improvement in the condition. Nothing I worked on is fully cleaned yet, but I invested a lot of time and sweat.


The 1132 printer signal cable, the one that appeared cut by a loose connector clamp but was found to be intact upon dis-assembly and inspection, was reassembled with a nice tight cable clamp on the outer sheath of the cable.


When I was removing the system connector from the disk drive SLT card box, necessary in order to operate the drive independently with some CE switches placed on the rear of the drive mechanism, the connector came apart.

Connector card at top left with clips visible, missing plastic grid
SLT connectors have a black plastic grid that snaps into the motherboard socket, with metal clips that fit over the PCB copper finger and in turn press onto the side of the pin from the backplane that sticks up into the socket. My PCB and clips came out of the plastic grid, leaving the grid inserted into the backplane slot.

Black plastic grid barely visible in top right backplane slot
I will have to carefully pull the plastic grid out of the slot, then insert the PCB plus clips into the grid to return it to a working connector.


Lukas Tschudi sent me the replacement part for the tape that activates the dual color ribbon feature, which I received in the mail today although I haven't yet installed it. The 1053 uses a small fabric ribbon that has black ink on the top half of its ribbon height and red ink on the bottom. The feature lifts the ribbon holder to move the red part of the ribbon in front of the typeball when shifted, otherwise it sits in a rest position that uses the black portion.

Tape for ribbon color shift mechanism - too short to fit my typewriter
Alas, I have the wrong tape. Lukas asked which of the two sizes of tape I needed, based on the platen width. My width matched the larger but the Illustrated Parts Catalog for the typewriter has only one part listed, the number of the shorter. I went with this part number, but that gave me the incorrect part.


I picked up another tube of the special Mobil grease that is the best substitute for IBM #23 grease, better than the original because it won't dry out and glue parts together the way that the IBM original grease has done over the course of decades.

Mobil Aircraft Grease #28
My can of Nevr-Dull is ready, whenever I am ready again to slave over the keypunch metal parts just to restore a like new sheen to them. It won't be in the next few days as I have had enough polishing and wiping fun for the week.

Metal polish and corrosion remover - one microinch at a time apparently

No comments:

Post a Comment