Precise and accurate measurement of high value resistors

[stevebyan]

As an example, let’s say I want to restore an old HP-410B VTVM. It has two 50 Megohm 1% resistors in its input voltage divider. I need to measure them to within 0.33% or even better to within 0.1% in order to determine whether they are within tolerance or if instead they must be replaced. What instruments can I use to make this measurement? The only choices I know of are the Guildline 9520 Teraohmmeter and its successors, the General Radio 1666 DC Resistance Bridge, and the Guildline 9920 Current Comparator Bridge and its successors.

I’m particularly interested in instruments that are affordable by a hobbyist. Are there other possibilities? I’d like to cover up to a couple hundred Megohms with 0.1% accuracy.

[jbb]

I could be horribly mistaken, but would a Source Measure Unit (SMU) work? They can apply a pretty high test voltage in combination with a sensitive current range.

[Kleinstein]

The high value resistors are used in the higher voltage ranges. It should be enough to check if these range work OK and thus the divider. This is testing the resistor ratio for the divider, but this may be even better than just a resistor measurement.

[stevebyan]

So I then need an electrometer, something with an input impedance of 5 Gohms or more, with 0.1% accuracy, right? So a Fluke differential voltmeter, I guess.

Edit: looks like my HP 3456A could do it if I stick to less than 10 Volts input.

[Wallace Gasiewicz]

Is it more important that the 50 Meg Ohm resistors be entirely accurate or more important that they are the same value?   
Do you have equipment to measure One Meg?

[mawyatt]

Since the resistors are for a voltage divider why not use the Ratio Function of the HP34401A or other DMMs? Set a convenient Ref Voltage and measure this and divider output, thus the voltage division ratio as shown in HP34401A manual.

See page 44 of HP34401A users manual.

https://engineering.purdue.edu/~aae520/hp34401manual.pdf

Best,

[Alex Nikitin]

It depends on the value. Up to 100M even a decent 6.5 digit multimeter, like Keysight 34465A, will give you 0.3% (1 year) accuracy and if you have a reference resistor with, say, 0.1% tolerance, you could improve on it. Above 100M things became increasingly difficult with the value increase. In my home lab (using Keithley 263) I can measure up to 100M with better than 0.2% accuracy, up to 1G better than 0.3% and up to 100G better than 1% (and can compare resistors to a much better accuracy, however my reference resistors for high values are accurate to 0.07% for 100M, 0.1% for 1G, 0.25% for 10G and 0.4% for 100G). My Keithley 617 electrometer can measure up to 200M with 0.3% accuracy (1 year) and up to 100 TOhm with about 3% accuracy (using V/I method).

Cheers

Alex

P.S. - I've just tried a direct measurement of 100M from K263 by HP3456A, the displayed value is 98.84M for the 99.007M reference , so about -0.25% error (specifications for 100M range is 2%) .  Might be OK for comparative measurements though, however for my HP3456A, I know that the input leakage current is under 10pA, might not be the case with a different unit.

[stevebyan]

Is it more important that the 50 Meg Ohm resistors be entirely accurate or more important that they are the same value?   
Do you have equipment to measure One Meg?

They need to be accurate, as they are in series with other values that make up the range divider.

Yes, my HP 3456A should be adequately accurate for measurements below one megohm.

Edit to add: I do have a working GR 1666 that as near as I can tell is still accurate, and a pair of Guildline 9520’s that I hope to get working. But I’m interested in other options.

[stevebyan]

I wonder how HP did it back in the 1950’s, without a 34465A or 34401A? And without the GR 1666 or Guildline 9520 or 9920?

[Wallace Gasiewicz]

I think you meant to say that the 3456 is accurate enough below One Gig Ohm.   
It should be fine for this measurement.     

Otherwise put a One Meg Ohm resistor in parallel  and use one Meg  scale.

[stevebyan]

I think you meant to say that the 3456 is accurate enough below One Gig Ohm.   
It should be fine for this measurement.     

Otherwise put a One Meg Ohm resistor in parallel  and use one Meg  scale.

No, at 90 days the accuracy spec for the HP 3456A on the 100 Mohm range is 1.8% plus one count. The 1 Gohm range is 16% plus one count :-(

100 Megohms in parallel with 1 Megohm on the 1 Megohm range might get me there, I’ll have to run the numbers.

[Vgkid]

*snip*

They would have used a guarded wheatstone bridge. That is  what everyone was using.

[EC8010]

I've hit this one before. The problem with any electrometer is that it uses high value resistors internally, so they have to be thick film, and that means they have a significant voltage coefficient, yet (in a Keithley) they might have any where from 1V to 200V across them. In short, an electrometer is only accurate to about 1%. I needed some accurate 100M resistors, so I measured a load of 10M 1% metal film resistors individually on a 34410A, added the numbers together and added a trim resistor to get an accurate 100M. Yes, eleven resistors in series. I looked, but 0.1% tolerance isn't available in 10M unless you go to metal foil, in which case you can have 0.02%. At a price. If you don't mind buying 10M 0.02% metal foil, you could use that as a way of measuring the 10M 1%, either by making a bridge or using the ratio function of a meter. Whatever you do, it will be tricky and huarding will probably be necessary. And screening the measurement to minimise hum.

[Alex Nikitin]

I've hit this one before. The problem with any electrometer is that it uses high value resistors internally, so they have to be thick film, and that means they have a significant voltage coefficient, yet (in a Keithley) they might have any where from 1V to 200V across them. In short, an electrometer is only accurate to about 1%.

That is not quite correct, the maximum voltage on the measuring resistor in Keithley 617 is 20V, so a possible additional error from the VC even as high as 20ppm/V , would be less than 400ppm or 0.04%.

Cheers

Alex

[stevebyan]

They would have used a guarded wheatstone bridge. That is  what everyone was using.

Which guarded Wheatstone bridge? The General Radio 544-B Megohm bridge is only accurate to 4% at high resistances, and the newer 1644-A Megohm bridge is only accurate to 1% in my range of interest.

Did Leeds and Northrup make something suitable?

[EC8010]

That is not quite correct, the maximum voltage on the measuring resistor in Keithley 617 is 20V, so a possible additional error from the VC even as high as 20ppm/V , would be less than 400ppm or 0.04%.

But if you use a 6517, it can go to 200V, so (by your calculation) it introduces a far less tolerable 0.4% error. When I first encountered it, I thought that extending the electrometer's voltage swing to +/-200V was a very clever solution to the range switching problem. Now, I'm not so sure.

Back to the OP's problem, a very specific measurement needs to be made. Make a bridge from scratch and ensure that its terminals are guarded. That would be achieved by using an old-fashioned tag strip having its terminals on ceramic insulators fitted in a steel strip.

[TimFox]

The long-gone Keithley 515B bridge was rated for 0.05% accuracy from 100 k\$\Omega\$ to 100 M\$\Omega\$ and 0.1% accuracy from 100 M\$\Omega\$ to 1 G\$\Omega\$.
https://download.tek.com/manual/515B(Model515).pdf

[EC8010]

I've just looked up the Keithley 515. Wow. On the plus side, with a size like that it should be cheap if one ever turns up. But post and packing won't be.

[stevebyan]

The long-gone Keithley 515B bridge was rated for 0.05% accuracy from 100 k\$\Omega\$ to 100 M\$\Omega\$ and 0.1% accuracy from 100 M\$\Omega\$ to 1 G\$\Omega\$.
https://download.tek.com/manual/515B(Model515).pdf

Awesome! Thanks! I'll keep an eye out for one of those!

[stevebyan]

My Keithley 617 electrometer can measure up to 200M with 0.3% accuracy (1 year) and up to 100 TOhm with about 3% accuracy (using V/I method).

Alas the Keysight 34465A and the Keithley 263 and 617 are out of my bottom-feeding price range. A broken Keithley 417 might be affordable, I'll go see if the specs are good enough.

[Conrad Hoffman]

I know the GR 716-C capacitance bridge can also measure high value resistors between 300 kohms and 80 kMohms at 1 kHz. No idea of the accuracy and it's probably more of a comparison, but just FYI. See page 6 and 7 of the manual under Special Measurements.

[Alex Nikitin]

You can always build an accurate MOhm Meter  . You would need one 10M 0.01% resistor (Mouser p/n 684-USF340-10M) and an accurate voltmeter (say, your HP3456A) to measure the input and output voltages. The ADA4530 is also available from Mouser and can be replaced by a cheap LMC662, however you would lose some accuracy unless you add an offset adjustment. With a 10M reference resistor and a stable 1V input you should be able to measure resistors from 1M to 100M with better than 0.05% accuracy without any adjustments. You would need to use a correct build method, lifting the inverting input node (Rx1) into the air or on a PTFE support (a good PTFE isolated BNC socket will do the job), and using a very low leakage capacitor (TDK NP0 ceramic 1nF or Polystyrene) for C1 .

Cheers

Alex

P.S. even the input voltage needs to be stable but doesn't need to be accurate, if you use the Ratio mode on the HP3456A for the input and the output voltages. If you reduce the input voltage to 100mV, the range would extend up to 1Gohm with a somewhat reduced accuracy.

[trobbins]

I went the multiple series resistor path to configure 900M and 9G calibration resistors for full-scale analog meter setting of a 1960's megohmmeter.  I had a box of 68M 5% VR68, so sorted them using the same meter, although I did stop short of preparing a 90G calibration resistor.

The accuracy of the summed reference resistor should align with the measurement accuracy of the base parts, so that then returns to how confident you can make a measurement of a base part, and how well you can construct a string of parts with low leakage, and make measurements at a known temperature.  Then compare measurements of the 50M parts with the reference.

[Vgkid]

They would have used a guarded wheatstone bridge. That is  what everyone was using.

Which guarded Wheatstone bridge? The General Radio 544-B Megohm bridge is only accurate to 4% at high resistances, and the newer 1644-A Megohm bridge is only accurate to 1% in my range of interest.

Did Leeds and Northrop make something suitable?

So L&N made the 4232 Guarded wheatstone bridges
https://www.eevblog.com/forum/metrology/ln-4232-wheatstone-bridge/
Here is the 4737
https://www.eevblog.com/forum/metrology/leeds-northrup-4737-a20-tear-down/new/#new
In addition to the Keithley 515 , L&N also made a megohm bridge as early as the mid 60's...
Since I found this as a neat topic I will post a bunch of documents.
Also add Bob Pease "What's all this Tee Network stuff, anyhow?"
and Guildline 9336/9337 manuals / IET Labs SRC-100T manual
There is another one I'm missing as well.

[ch_scr]

[...] I needed some accurate 100M resistors, so I measured a load of 10M 1% metal film resistors individually on a 34410A, added the numbers together and added a trim resistor to get an accurate 100M. Yes, eleven resistors in series. [...]

I've made a python script for just this application. It takes the list of your measured samples and the target value,
and it'll automagically tell you which ones will make a string that's exactly on point. Or rather, how close it gets. It takes surprisingly little overhead to find a "real close" combination.
The idea is, if the stock has a Gaussian spread, the right combination of "odd values" ends up stacking exactly right. The script tries them all for you.

Functions:

Code: [Select]

string_optimizer.py -h
usage: string_optimizer.py [-h] [-i] [-p [R_P [R_P ...]]] Target S/P N R [R ...]

Tool to optimize resistor series string or parallel connection for lowest initial error

positional arguments:
  Target                target value of the series/parallel connection to be optimized for (e.g., 10.198523)
  S/P                   choose between (S)eries or (P)arallel connection
  N                     number of resistors to be selected for connection
  R                     resistor values, space separated

optional arguments:
  -h, --help            show this help message and exit
  -i, --iter            show each iteration that finds a new best value
  -p [R_P [R_P ...]], --prio [R_P [R_P ...]]
                        specify resistor(s) values that will be assumed fixedExample for a target of 100 Meg, series connection, 10 resistors to be selected and 12 measured values

Code: [Select]

string_optimizer.py 100.00 S 10 9.999 10.010 10.123 10.001 9.890 9.780 9.991 10.1 9.8999 10.01 9.987 9.897

Residual error 80.00000 ppm
With these 10 in series:
(9.999, 10.01, 10.123, 10.001, 9.89, 9.991, 10.1, 10.01, 9.987, 9.897)