Keithley 2000 gain repair

[Galaxyrise]

I started to get the itch to augment my meter collection with a 6.5 bench meter after hearing people here talk about getting good used gear from eBay.  But after watching the prices for a bit, I came to thinking that if I bought a broken meter instead, not only was I buying a meter, I was buying hours of entertainment!   Still, I was cautious and waited for something with a problem I thought I would be capable of fixing, and eventually found a Keithley 2000 which was failing self test 200.2.  The service manual describes this as the as 200.1, but with a different resistor network switched in.  I figured I could probably handle troubleshooting resistors!

The unit was amazingly clean inside.  I feel guilty for leaving probe marks all over the place, hehe.  I was able to piece together the schematic for the part of the board in question.  As you can see, the 100 gain section is only producing a 90 gain, and the 1079 ohm resistor seems to be the likely candidate for being out of spec as a 1k in its place would produce the correct gain.

I took a picture of that part of the board, but sadly my camera (and photo skills) are not up to the task of taking pictures where you can read the part numbers.  U166 is an AD711, U129 is a DG211DY, R309 is labelled 1k00 and R310 is labelled 9k09. 

I did some quick algebra and computed that a roughly 14k resistor in parallel with the 1079 would bring it down to 1000, which would fix the gain ratio.  So I wrapped a 10k and a 4.7k together and held them in place... voila! Passes the self test

Some questions for you all:

• What function does R311 serve? (or R308 for that matter.)
• I haven't modified the board yet, so to get those resistance values, I measured various points on the board and then set up systems of non-linear equations and guessed values in excel until I got close to my measurements.  Is there a better way?
• What causes a portion of a resistor pack to change like that?  The simplest repair would be to solder a 14k between pins 9 and 6 of the resistor pack, but can I trust the 1079 resistor to be stable now?  Or am I better off bypassing the resistor pack entirely? Getting stable resistors in a 99:1 ratio seems pretty tough, but I suspect 100:1 would do.

(schematic done with www.circuitlab.com)

[KJDS]

If you draw out the schematic with the op-amp facing to the right and just the X10 components in place, I suspect you'd understand the circuit much easier.

[free_electron]

the reason for those ressitor packs is that these are fine tuned using laser during production. having them on the same substrate also keeps the temperature coeficcient pulling in the same direction so it increases overall performance of the meter.

the drawback : if one resistor goes poof its game over. that resistor pack is custom.
the reason for the pack going wonky ? probably electromigration.

there is several posts on the web where that pack failed...
they should have used glass passivated as opposed to that sot package... but the sot is cheaper. the glass passivateds are in thte 2001 and 2002 , not the 2000 ...

[Galaxyrise]

Searching the net once I found the defective part is a good idea! Obvious, even, in retrospect.  I didn't find many hits, but I did find one here on eevblog where a good R271 is measured out.  Very nice!  It confirms my conclusion that 9-6 was supposed to be 1k. 

electromigration... interesting!  Reading up on that, I'm certainly inclined to bypass the 1079 completely.

[Rick]

What is the tolerance of that TF-245? 1%, 0.1% or 0.01% ?
I got that resistor network from the same company as Costas and its values match Costas' findings. So it seems fine.
I did not have time to try a repair though. If it turns out this is the only problem, it will be great.
But if I replace it with precision resistors I would like to know the tolerance. Or may be the tempco is more important than the tolerance, is it ?

[free_electron]

its not so much the absolute values but the matching between them. and the fact that they all drift with the same percentage under temperature.

[plesa]

Resistor network which is 1:1 replacement for K2000/2015/2001/2002 networks are
http://www.bitechnologies.com/pdfs/660.pdf
Some of them have Mouser on stock.

[Galaxyrise]

Resistor network which is 1:1 replacement for K2000/2015/2001/2002 networks are
http://www.bitechnologies.com/pdfs/660.pdf

I don't think we're talking about the same resistor network.  The TF-245 resistor pack was measured by costas to have the following network:

[Galaxyrise]

Followup: I bought a Caddock 1k-9M resistor network to replace the 1k:99k divider.  Since the actual resistor pack was done with a 1k, a 9k, and a 90k, I opted to replace all three of those with the ones in the Caddock.  I cut the resistor pack pins off with a side cutter and then wired the Caddock to the corresponding pads. And now all self tests pass, yay! I attached a picture of the resulting Frankenstein's monster I'd like to secure the resistor with something more than just the bodge wires, but they seem adequate for now.

As expected, the ohms measurements need recalibrating now.  But it seems that I can't just calibrate resistance, I have to do all of DC measurement.  So that's a bunch of ranges to recalibrate, most of which are working fine right now.  And it also calls for a 100V dc, which I don't have a source for right now. So... now I have my next project! But 100V is way above anything I've designed with before... 

@KJDS: I think my question about the extra resistors was answered by Dave's Fundamental Friday series on opamps.  You don't see those extras in the basic non-inverting descriptions.
@free_electron: Thanks a lot for the info!