I need help diagnosing a HP9100B



#18

I am trying to repair a HP9100B. This is a very complex machine. At least to someone that is used to integrated circuits.

I have a copy of the manual, the service manual and the schematic drawn by Tony.

I have tried running the diagnostic program from the service manual. No way. Then I tried running just the routine from the program that tests the SIN X, COS X, TAN X and ARC functions. When I ran just that
it completed without errors every other time and stopped with an error every other time. Next I single stepped through the routine a few times. What I discovered was that the routine was producing errors on every run but was only signaling an error every other time.

This made me realize that the diagnostic program is of extremely limited utility being as it could experience MANY errors and still run through completely and never indicate any problem whatsoever.

So here are the basic symptoms I have observed so far.

ADDITION SUBTRACTION MULTIPLICATION DIVITION WORK

IN DEGREES SIN X AND ARC SIN X WORKS FROM 58 TO 147

COS X AND TAN X TEND TO GET THE SAME VALUES AS SIN X

IF I USE ARC COS X AND ARC TAN X THEY ACTUALLY WORK
EVEN TO VALUES LARGER THAN 147 AND LESS THAN 58

LOG X FAILS AND LN X FAILS

E TO X WORKS BUT GETS NEGATIVE VALUES

SQRT WORKS

THE REGISTER FUNCTIONS WORK

THE PROGRAM CONTROL FUNCTIONS SEEM TO WORK

Any help any one can give me would be greatly appreciated.

Thanks,

Bruce


#19

The 9100 calculators are astonishingly complex machines that, seen from a modern digital perspective, might as well have been crafted by elves and work by magic.

Really, that's not too far off the mark. These machines pushed so far beyond state-of-the-art at the time it's astonishing they work at all.

One a more prosaic level, individual parts like the static deflection CRT are unique to the machine; even individual transistors were hand-picked for each machine and modern equivalents do not exist (you can substitute some modern parts, but it requires hand-testing response curves.)

I know of three experts on these machines.

Of course Tony Duell is the Grand Master. He hangs out here occasionally. Located in England, Tony does this for a hobby and has decades of experience.

Larry Atherton spent months repairing two eBay-acquired 9100Bs I have. They were stripped, repaired, cleaned, had their power supplies rebuilt, keyboards disassembled and cleaned, defective parts located and replaced, and were repainted with matching epoxy paint. They're now museum-quality pieces. (It wasn't cheap.) You can reach Larry at latherton@motionsoftware.com.

Mark Sims is the one other person I know who has successfully serviced multiple 9100s, but he's a reclusive hermit and doesn't want to talk to you.


#20

Quote:
even individual transistors were hand-picked for each machine and modern equivalents do not exist (you can substitute some modern parts, but it requires hand-testing response curves.)

Hmm, that would usually indicate poor design. But maybe they had to resort to some tricks in order minimize component count.

Let's see if the schematics are on the Museum DVD...


#21

The schematics are not on my copy of the DVD but they are on the site in Australia.

#22

Quote:
Hmm, that would usually indicate poor design. But maybe they had to resort to some tricks in order minimize component count.

I am a software person, not a hardware person, but the hardware people that know these machines describe them as being "...touched by the hand of God."

OK, well, not really. But they should not have been possible given the technology of the time, so I'll forgive them this.

#23

How does Tony feel about being contacted? Or does he still show up here occasionally?


#24

I don't know how Tony feels about being contacted. Larry and I posted questions for some weeks before he joined the conversation. Other people here know him far better and would have a better idea.

So, other people? Can we give out Tony's email address?


#25

I have no objection to being contacted by anyone trying to repair an HP desktop calculator (or for that matter any other device that I might have knowledge of). I may not be able to help, but I am not going to complain.

My e-mail address is pretty well known anyway, I have no objection to you passing it on.


#26

In that case:

Tony Duell <ard@p850ug1.demon.co.uk>

#27

Off the top of my head, this suggests an addressing problem with the main ROM. Some locations are not being addressed properly.

First test : Does it behave in exactly the same way if you exchange the 2 flip-flop boards? The one on the left (next to the core-on-a-rope microcode) is mostly concerned with addressing the ROMs, the one on the right (next to the core memory) is mostly processor registers.

Check for bad connections between the sideboards, the gating board and the ROM assembly. Remove the sideboards, clean the connectors.

Then check the transistors on the main ROM itself (middle board on the underside of the processor). Be very careful working on this board, as you possibly know it's many layered and the ROM is formed by indcutive coupling between the tracks so layout is critical.

After that, check the ROM address decoder transistors (rear board on the underside). I suppose it might be a problem with the ROM data (the 2 boards at the sides of the ROM) or even diodes on the gating board, but I think that's less likely.


#28

i tried swapping the 2 flip-flop boards (Paul suggested that) and checked the diodes while doing so. I could find no difference in the machine after doing so. I will try the rest of your suggestions next.

#29

Sounds like a memory problem to me. Is there a way to do a read/write test on the RAM? Is there a way to verify that the ROM contains the right values?

As others suggested, an error in either of these tests could indicate bad memory (RAM or ROM), or a problem with the addressing circuitry.

I don't know anything about the 9100B. These are general comments that apply to any digital device with these symptoms.

Good luck, and let us know what you find!

Dave


#30

"RAM"? "ROM"? Ah ha ha ha ha!

The 9100 has no "RAM". It has 3072 bits (in the "B" version) of magnetic core memory. You know, really tiny iron donuts strung on wires.

If this has a problem you're pretty much out of luck. Fortunately, aside from physical damage, mag core memory doesn't really "wear out", so problems there are likely in the technically-repairable addressing and support circuits.

The implementation of "ROM" is so bizarre I won't even try to describe it, because although it's been described to me I can't grasp the concepts.


#31

Core memory is Random Access Memory (RAM) as where "Thompson" tubes as opposed to coils of spring steel wire or tubes of mercury which operate like a large shift register and you have to wait for the data you want to access to come around. Core also has the advantage of being non-volatile. On one old machine I serviced the core was so reliable that some of the operators did not have a clue how to reload the machine. they would shut it off at night and turn it on again in the morning and at most reset the machine to get get it going. You could also play tunes on older core unit when they had larger donuts that hung loose on the x,y, and sense wires because they would rattle when accessed.

Edited: 6 July 2012, 1:13 p.m.

#32

Quote:
The 9100 has no "RAM". It has 3072 bits (in the "B" version) of magnetic core memory. You know, really tiny iron donuts strung on wires.

The mag core memory systems with which I'm familiar meet fully any definition of random access that I've seen.

Quote:
...aside from physical damage, mag core memory doesn't really "wear out"...

I'd bet that the mag core RAM and rope ROM in Apollo 10 LM Snoopy's AGC would work today if powered back up...where ever it is now.

Quote:
The implementation of "ROM" is so bizarre I won't even try to describe it, because although it's been described to me I can't grasp the concepts.

Like rope memory?


#33

Quote:
Like rope memory?

The 9100 has two kinds of ROM. One is a 64 word by 29 bit "core rope memory" (AKA "wire braid memory" or "threaded core memory"), and is the lowest level control store. Core rope memory is able to store multiple bits per core, by use of wires passing either through or around the core to represent a zero or one. Each core is used purely as a transformer, rather than storing a single bit via its magnetic hysteresis loop. In the 9100, the core rope memory has 64 wires and 29 cores.

The next higher level of control store is a 512 word by 64 bit (32 Kbit) inductively-coupled 16-layer printed circuit board memory. This is in principle very similar to the IBM "TROS" memory used for the 360/40 microcode, but fabricated as a single PC board.

HP described the 64x29 ROM as being "control sequences" and the 512x64 ROM as being "microcode". This is consistent with HP's use of terminology on their later calculators, but normal industry terminology would call the 64x29 ROM "microcode" and the 512x64 ROM the "program" (which is then responsible for interpreting the RPN user language). As a result of this terminology confusion, some sites claim that the 64x29 ROM is higher level code than the 512x64 ROM, which is incorrect.

#34

A description of the HP 9100 pcb ROM with close-up photo is here:

HP 9100 pcb ROM

--Steve


Possibly Related Threads...
Thread Author Replies Views Last Post
  HP9100B available on TAS Frank Boehm (Germany) 7 263 10-02-2010, 04:20 AM
Last Post: Juergen Keller
  HP9100B parts and information Jools Smyth 0 84 10-18-2003, 02:34 PM
Last Post: Jools Smyth
  HP9100B Repair Thomas Falk 2 149 04-18-2003, 07:04 PM
Last Post: Tony Duell (UK)
  HP9100B Card reader Marcel LERAT 6 241 03-24-2001, 04:12 PM
Last Post: Marcel Lerat

Forum Jump: