Unstable resistance 6.5 digit DMM

[cowasaki]

I've never had a high accuracy DMM before the Keighley 2100.  Just hooked it up to a 100 ohm resistor using a four-wire probe set and I'm getting 98.981x where the x is basically all over the place occasionally taking the 1 with it.  The four-wire probe set is from a pair of Kelvin probes which I bought off ebay re-wired with 2 cores going to each probe with 1 to each part of the probe.

The cable is some nice silicon stuff I bought for making up banana plugs and thought I would use it for these clamp things.

Am I missing something obvious?  Do I need to do something special with the four-wire set?  shielded cable?  short cable? This is just out of interest, I don't actually need this level of accuracy at the moment......

[GopherT]

You are hoping to get a reading that is essentially 100.000X where the last digit "x" is your drifting result.

I think your resistor is more at fault than your meter. Any tiny temp change will cause that last digit to drift unless you have a very low temp coefficient resistor.  You are kind of asking for less than 1ppm and even radiant heat from your hand being close then far from the resistor can give a slight change to the temp of the resistor and, therefore, the resistance of the unit.

[Kleinstein]

The meter can measure at different rates. For very fast readings there can be higher noise than normal.

Another point is to make sure the meter is actually in 4 wire mode. The 2 wire mode may be a little less stable.

For 100 Ohms, shielded cables are usually not needed.

How good is the resistor used for the test ?  With a relatively cheap 100 ppm/K resistor, the 0.1 mOhms digits correspond to some 10 mK temperature change.  So with a cheap resistor it could be just temperature variations.

[cowasaki]

You are hoping to get a reading that is essentially 100.000X where the last digit "x" is your drifting result.

I think your resistor is more at fault than your meter. Any tiny temp change will cause that last digit to drift unless you have a very low temp coefficient resistor.  You are kind of asking for less than 1ppm and even radiant heat from your hand being close then far from the resistor can give a slight change to the temp of the resistor and, therefore, the resistance of the unit.

I was just a resistor I had sat on the bench it's a 10w 100ohm resistor!  To be fair it's probably 5% tolerance BUT it does the same if the probes are clipped to each other........

[HalFET]

I've never had a high accuracy DMM before the Keighley 2100.  Just hooked it up to a 100 ohm resistor using a four-wire probe set and I'm getting 98.981x where the x is basically all over the place occasionally taking the 1 with it.  The four-wire probe set is from a pair of Kelvin probes which I bought off ebay re-wired with 2 cores going to each probe with 1 to each part of the probe.

The cable is some nice silicon stuff I bought for making up banana plugs and thought I would use it for these clamp things.

Am I missing something obvious?  Do I need to do something special with the four-wire set?  shielded cable?  short cable? This is just out of interest, I don't actually need this level of accuracy at the moment......

That's not unexpected to be honest. Even with a purpose-built setup (i.e. Keithley 2182A nanovolt meter with a dedicated current source) sub-milliohm measurements are non-trivial matters, quite a few things start to matter. Check the Keithley Low Level Measurements Handbook for all the things that start to interfere.

But to get to your specific case: I'm not familiar with the specs of the 2100, but given the specs of other Keithley 6.5 digit meters it's probably operating at its lower limit. So while you might have a resolution of 0.1 mOhm, the measurement noise or uncertainty might be higher than that. Did you set PLCC to 10 and turn on averaging/filtering? Additionally, how long had it been running when you performed that measurement? Keithley has the habit of defining measurement performance after a warm-up period of one hour, anything less and your odds of reaching that performance are somewhat diminished. Also, did you use perfectly symmetric cables?

[GopherT]

Even the copper wire probes you are using suffer from temperature coefficient effects.

The Temperature Coefficient of Copper (near room temperature) is +0.393 percent per degree C.


A 30cm piece of 20 gauge copper will change by about 40 microohms per degree Celsius (I.e. 0.000040 ohms). So your test leads, in your hand, will be jumping around quite a bit when measuring to 6 or more digits.

[GopherT]

Even the copper wire probes you are using suffer from temperature coefficient effects.

The Temperature Coefficient of Copper (near room temperature) is +0.393 percent per degree C.


A 30cm piece of 20 gauge copper will change by about 40 microohms per degree Celsius (I.e. 0.000040 ohms). So your test leads, in your hand, will be jumping around quite a bit when measuring to 6 or more digits.

To clarify, a 4 wire measurement even a 1mm distance of your "sense" probes across your current source probes will cause a measurable drift in resistance with a one degree temp change at the gap between sense probes.

[jancumps]

...
The four-wire probe set is from a pair of Kelvin probes which I bought off ebay re-wired with 2 cores going to each probe with 1 to each part of the probe.
...

Can you draw what you did. 4 wire probes are expected to only connect at the resistor.

[cowasaki]

...
The four-wire probe set is from a pair of Kelvin probes which I bought off ebay re-wired with 2 cores going to each probe with 1 to each part of the probe.
...

Can you draw what you did. 4 wire probes are expected to only connect at the resistor.

Yes.  Attached is a picture.  There are 2 approx 30cm pieces of 2 core, multi-strand copper wire with a pair of banana plugs at one end and the other end going to a pair of Kelvin clips with one core attached to each part of the clip. 

Is there a better way?  Should the wire be shorter?  Should I attach both sides to one part of the clip?

As I said, this was more out of interest.....

[floobydust]

You also need an electrically quiet workbench area for low level measurements.
Keep the mains power cords and outlets away from your test leads and the resistor.
You have a mains power cord nearby your leads which is a no-no. Unshielded 4-wire cables?
Also keep cordless and cell phones, WiFi, SMPS AC adapters, CCFL and fluorescent lights away.

[MrFox]

Using the highest averaging filters helps a lot with low resistance measurement, if you don't mind the slower read.

[jancumps]

on my meter, when using 4 wire measurement and shortcutting everything in one point at the tips, I get 0.005 ohm. the 5 is uncertain to one digit, everything beyond that is not usable. When I clamp the sense probes each approx. a centimeter away from that touch point I measure 0.03 ohm.
If you fixed your wires at the end of thecrocodile clamps, the kelvin functionality stops where the jaws pivot. That’s the first point of contact and closes the sense cycle. the remaining resistance of the crocodile clip, including contact resistance, will not be offset by the kelvin mechanism.

[cowasaki]

Thanks.  I will try making a second set using shorter shielded cable and try it away from other stuff.

[HalFET]

Thanks.  I will try making a second set using shorter shielded cable and try it away from other stuff.

Before you try that, have you even tried measuring a calibration short?

[cowasaki]

Thanks.  I will try making a second set using shorter shielded cable and try it away from other stuff.

Before you try that, have you even tried measuring a calibration short?

The leads are shorted out in the picture and I have set the DMM to zero
On that basis. If that is not what you meant then I’m not sure what you mean.

[HalFET]

The leads are shorted out in the picture and I have set the DMM to zero
On that basis. If that is not what you meant then I’m not sure what you mean.

Your leads may very well be the problem in the first place! I think you do not quite understand the difficulty of the measurement you're attempting to have the device perform.

Assuming it follows the same ranges as the Keithley 2001, you're operating in the 200 Ohm range with a 100 µOhm resolution and a test current of 0.98 mA. This means that your meter is now measuring something in the range of 98 nV. (100 µOhm x 0.98 mA). Is it doable to measure this? Surprisingly (to some), yes. But thermal voltages are a significant issue when operating in this range: http://download.flukecal.com/pub/literature/p18-21.pdf

But there's an even bigger hurdle, it's the uncertainty of the measurement circuit and the reference itself. Since I'm now particularly lazy and do not feel like doing error calculation math on a Monday morning, lets go by the dumbo approach to taking into account measurement accuracy. For example, the Keithley 2001 does about 50 ppm accuracy in its 200 Ohm range. So lets say we stick your 100 ohm resistor in there, 100 Ohm * 50 / 10^6 = 5 milliOhm. In other words, in a single measurement when performing no trickery (i.e. averaging) you simply aren't going to get a reliable number in the first place. It'll show you a number, but what it actually means is a whole different story. Now there are ways to get around these limitations, but realise that this is a dangerous path that risks getting you sucked down into the depths of metrology and soon you end up with a stack of 7.5/8.5 digit DMMs and calibrators lining your bench.

But even then, if you can, gauge the performance of the multimeter using a calibration short. They are meant to quantify the effect of the panel/interface/sockets/... during calibrations and directly short the multimeter pins to each other. An example of one: https://eu.flukecal.com/products/accessories/test-leads-probes-and-clips/884x-short (Keysight makes better, and pricier, ones though.) And while these things might look like a simple PCB, the materials used are probably quite well selected for the intended application.