Repair worklog : Old rusty HP 3458A

[Kleinstein]

Looks like a Problem with R4 / R5 or possibbly extensive bias if U165.
The 10 V measured will be compared to a combination of the +12 V and the -12 V. So it's no so supprising that the meter reading is drifting more than the -12 V reference.

The good thing is, that it might be possibly to use external stable resistors instead of R4/R5.

Are there decouping caps direktly at the OPs ? Looks strange with resistors at the power inputs, but no caps.
 Is it possible that there is some oscillaton or excessive ringing ? This could also cause some aperent drift.

[TiN]

Dr.Frank
I have two A3 PCBs, original (let's call it OLD) one which came with meter, with U180 ceramic current steering ASIC throwing ERROR 114 : rundown convergence.
And second A3 PCB (let's call it NEW, while it's not technically new) which I got later, with U180 ceramic ASIC with this drift issue.

Now what I did - swapped U180 ASICs between boards. That's how I confirmed that U180 is cause of drift, not external components. Data and graphs are from old A3 PCB with "new" U180 ASIC. Now, there is a little chance that both boards have faulty parts which cause same drift, but that would be a long stretch assumption. If that would be U165, then it should have at least different speed of drift between two boards, but no, it's exactly same, about 1ppm per hour.

Yes, there is a chance that those internal resistors get damaged/mismatched. That is why I want to hook resistors from "bad OLD" ASIC to see if issue change. I plan to glue old ASIC in deadbug style right on top of ASIC in the board, so thermally everything will be same.

[TiN]

Yes, there are caps in close proximity, I just omit them on LTSpice sim for clarity.

Here's actual schematics from CLIP:



I measured pin 2 on each opamp too, in case that can help? I can add plots too.

[Kleinstein]

I don't see a big advantage of using the old Chip just for the resistors. I think 2 identical resistors (or mayby later a dual resistor module (maybe LTC5400)) should be just as good. These two resistors only need to match each other.

[Dr. Frank]

Dr.Frank


Yes, there is a chance that those internal resistors get damaged/mismatched. That is why I want to hook resistors from "bad OLD" ASIC to see if issue change. I plan to glue old ASIC in deadbug style right on top of ASIC in the board, so thermally everything will be same.

If you read the hp journal, about the A/D network , the resistors have to be matched for equal T.C. to guarantee short term stability, between ACALs.

If your +12V reference runs monotonically in one direction, obviously independent from temperature changes, then this is probably a constantly running degradation process of one of the two 10k resistors.
Something like constant corrosion, if for example the protection film on top has pinholes.

That's really strange.

I also do not know anything about the resistor technology, HP had used.
They have also their FineLine arrays, since the early 80ties, also unknown to me, what they use.

I assume, that could be something like TaN technology, as it has to be superior over Thin Film technology ( Ni/Si/Cr system).

Anyhow, maybe you can identify the root cause by bodging 2 external  ~10k Vishay resistors.

Frank

PS: Another idea: if you can really exclude the external components, then you may observe this continuous drift directly by measuring R4 or R5. Hopefully, the about 1.0 test voltage from the DMM used is enough to initiate possible diffusion processes, otherwise  1mA constant current and voltage measurement by DMM is preferred.

Such corrosion processes will always increase the resistance.

As -12V is increasing by absolute value, R5 would be most suspect.

[TiN]

Yes, I remember HPJ. Let's see If behaviour changes, I'll look into proper resistors and networks for more permanent solution.
Thanks for suggestions, appreciate.

[TiN]

Bodged up with copper wires...



Forgive the crude worker 

Changes in these REF voltages indeed impact readings. I just move cable to scan card a little, and readings change 8ppm instantly.
Set 2002 to read only -12VREF, fixed scan card at one relay only to reduce interference. Will leave it sampling log over weekend, will see. Last 40 samples reading is ~0.264ppm so far, -12VREF stays ~0.302ppm.

[Jwalling]

Bodged up with copper wires...



Forgive the crude worker 

You, sir, are NUTS!

[Kleinstein]

Dead bug style to it's best.

[ManateeMafia]

I sense more pcb daughter boards for the 3458A. I hope this can prove there is a solution for your particular issue. Of course, there is plan C....

[TiN]

Well, initial results does not look good.

There is plan D too, ditch ratiometric references, and just put three LTZs in there to make +12V, -12V and +5V references 

[ManateeMafia]

Yes. If you need encouragement, I don't think there will be a shortage of people to give the thumbs up.

I agree that the Keithley models are much easier to service.

[TiN]

I agree that the Keithley models are much easier to service.

I was sure about opposite just 2 month ago!  But yea, with Keithley's my experience was "it either works normally, or does not work at all". 

Issue with independent references is, that when any of them drift differently to others - it will magnify error.
I'll try with another opamps and without scancard wires tomorrow. Maybe my go-to LTC2057's instead of LT1001's?

[plesa]

You are defining  term Volt-Nut
I'm glad that Keithley used the of the shelf parts instead of custom made unobtanium ASICs.

[TiN]

It shoot thru the floor overnight.



-77ppm, record! Decay seem to speed up
Could be resistors bad in old ASIC too?

EDIT: Bodged up LT5400A-3 instead of -12sum 10K resistors, replaced U165,U151 with LTC2057 choppers.
Let's see...

[TiN]

LT1001 have offset current 0.4nA typ, 3.8nA max (+25°C) with input bias current +/-0.7nA typ, +/-4nA max (+25°C). En is about 11-18nV/SQRT, specced 300nVpkpk, GBW 0.8MHz.
LTC2057 have offset current 0.06nA typ (+25C), 0.46nA max (-40 to +85°C) with input bias current +/-0.3nA max (+25C), +/-3.5nA max (-40 to +85°C). En is about 11nV/SQRT, no 1/f noise, specced 200nVpkpk, GBW 1.5MHz

Also 2057 have total supply voltage at 36V, which means those resistors at opamp power supply pins can be removed. I left them in place for now (not sure if ASIC current balance circuit would like higher amplitude, but likely be allright, as it's charge balance ADC topology, not voltage.), but even then voltage at opamp pins is higher now, about -16.5 and +17.3VDC, due lower opamp supply current.



Voltage is creeping up, likely due to heat (blue line on graph, TEMP? data). You can see control meter - K2001 - also went up about 1ppm due warmer temp (3458A sits on top of k2001).
PPM scale for 3458A reading (red line) is green scale, went up 0.6ppm from power on state , sorry for misleading coloring
Purple line is +5VREF (wire from -12VREF came out, did not bother to reassemble everything again).

As a side note, I'm impressed with 3245A+LTZ performance so far, seem to be really stable thing, likely beyond my current measuring capabilities.

[TiN]

Drifted up 7ppm today, 5V reference went a deep dive (hundreds of ppms).

Now replaced 24K+10K from ASIC with external resistors (10k fluke wire wound 3ppm/K + R2R 25.5K&2*226K 25ppm/K (don't have any shmick 24K resistors here))

Will it help? Only way to find out is to let it warmup for another six hours....

What puzzles me  - if it's something caused resistors go bad in ASIC, why resistors used in +12VREF are okay and that reference is stable on both ASICs?

[TiN]

+16ppm overnight. But now we talking some business.



Recap for configuration:

+12VREF - resistors from old bad error 114 ASIC
-12VREF - pair of resistors from LT5400A-3 network
+5VREF - Fluke 10K 3ppm/K resistor + RNC50J2552BS 25.5K 25ppm/K with two RNC50J2263BS 226K 25ppm/K in series to get 24.1K

Now, what you see, and why it is good?

[Kleinstein]

The 5 V are still not very good - but this was expected and does not reay matter. There seems to be no influence of that.
So maybe this is already good enough.

The +12 V and DMM reading show quite exactly a oposite trend. The -12 V very much paralles the ADC reading. So the ADC itself, comparing the 12 V to the input seem to work very good.

So the only part still not working good is the 7 V to 12 V step - but this does not need to be done in the ASIC. Due to the ACAL function of the DMM this does not even have to be long term stable - so just to resistors with well matched TC will do the job.

There a a few strange steps in the voltages - maybe the reference module itself is not that perfect.

[TiN]

I'll go to those steps later. You can see updated realtime datalog.

What I understood from graph in post above , are indeed two things:
Having 5V formed by external resistors (even bad as 25ppm/K) now got 5V to behave with close coupling to temperature. Before we had only constant drift, which was way larger that any tempco deviations. Knowing this simplifies issue just to matter of dollars for stable resistor network.

Now, we go back original LTspice simplified circuit I shown before. -12v and +5v refs are derived from +12vref, remeber? And now since we have -12v and +5v "stabilized" almost all of that drift (~30ppm) comes from...ASICs resistors 6.0K and 4.14K, which we still have used from old deadbugged U180.

This morning I ditched ASIC allthogether and replaced 6K with 10K Fluke mica wirewound (3ppm/K) paralleled with 19K Vishay VHP (0.3ppm/K max) and 4.14K replaced with 5K Fluke wirewound mica (5ppm/K). This gives a tad higher 12.45, -12.45 and +5.15v ref levels, but should be no worries.

Anyone into business to get us 6-resistor network for ultimate performance?
I ordered 10k+10k hermetic vishay network from ebay to replace LT5400 (its absolute tempco is not so great). Will see how that goes.

All data is with NPLC10 btw. If levels stay stable (within 2ppm), we will switch for more accurate NPLC100 testing with proper resistor  wiring, single copper thermal anchor and no 40cm scancard wires hanging in the air all over the joint.

I might rewind few of 10k/50k fluke resistors to make 6.5k, 4k and 24k resistors. I have few 250k ones, but I don't think my soldering skills up to task to make resistor network with 0.01mm manganin wire...

[VK5RC]

Hi TiN, I am not sure if you saw this thread of Andreas some time ago, a contributor Edwin Pettis added an interesting perspective of stability of resistors (selecting the ones with the TempCo you want).
https://www.eevblog.com/forum/projects/t-c-measurements-on-precision-resistors/

[TiN]

Sure I saw it And here with this ADC rare example when both abs. Tempco and matching are needed. I'm not sure if Edwin will be up to 6-resistor network in few pieces qty.

[Kleinstein]

The requirements for the resistors are not that high here. Especially there should be no need for all 6 resistors to be matched. This is 3 pairs of resistors, that can be treated seperately. The main issue is to have very good TC matching, especially with the 10K/10K pair. The absolute values are not at all critical and only the 1:1 ratio for the 10 K ones should be close. I see no problem if the others are a few percent off.
Also long term drift should not be an issue - the thing is to have less than 0.1 ppm over a day or so. It would not mater to have somthing like 20 ppm/year drift, as a ACAL should readjust for new vaues.

The other question is where the 5 V is actually needed for.  My guess is, it might be used for Ohms  - but the diagram will tell us. Depending on this the 5 V may not even be critical at all.

[TiN]

I meant tempco and tempco matching. Resistance value is not that important, yes. Hopefully VPG hermetic network should do well.

[Theboel]

Sure I saw it And here with this ADC rare example when both abs. Tempco and matching are needed. I'm not sure if Edwin will be up to 6-resistor network in few pieces qty.

There is nothing wrong to ask