T.C. measurements on precision resistors

[MisterDiodes]

The general  head's up on LTC24xx and other LT switched cap parts - If you use these for measuring resistors (even for relative value shifts) you have to really pay attention to how you drive those inputs.  When you take the time to actually test these 24 / 32 bit ADC's on accurate & calibrated sources, you find out in a hurry where the datasheet is a bit optimistic.  With further conversations with LT and more product testing we've found that what the datasheet doesn't tell you when you're chasing PPM - in a nutshell:

A) Keep the source impedance <<< lower than the first input switch Ron (which is usually around 5k).  Which means your source impedance wants to really be <<500 ohms for lower gain error.  Keep source impedance <<50 ohms for closer total INL error datasheet spec.  This also reduces that input protection diode leakage effect (which applies a TC effect error to your data); measured leakage values can tend to run around 20~30%  higher than datasheet values across full temp span.  Otherwise your data will be shifted in gain / offset or INL.

In other words, these are made to use an input buffer in almost all cases, unless your signal source is very low impedance to begin with.

B) Pay really really close attention to slapping capacitors or RC filters on these ADC inputs or Vrefs pins.  It seems  counter-intuitive, but those can degrade INL and cause offset in a hurry if you're not careful.

C) Remember that the input pins and Vref inputs will have an AC current component driven onto your signals, be aware.  If you're after PPM level accuracy, this can have a observable effect, so watch out for that.

However you're testing resistors, it's always a good idea to test & verify your method or device data against a known good resistance measuring system (which will always be a resistance bridge, either manual or automatic...NOT a DMM) AND your resistors under test should be at a typical BIAS load and TYPICAL thermal flow + mechanical mounting condition for whatever application you're testing.

If you don't have a resistance bridge available: that's why I suggested running LTZ ratio resistors ON an LTZ - That's the most accurate test for that application.

By the way -   

If you really want to test resistors correctly, here's a nice kit for a fraction of the "new" price.  If you haven't seen how a precision resistor oil bath works, here's an example:

https://www.ebay.com/itm/Measurements-International-Automatic-DC-Resistance-Bridge-Complete-System/282828252605?

You get some nice goodies in there like the Hart oil bath to maintain your references, an SR-104 (which is 10k, not 10M as the description reads), a 3458a, a few Low EMF switch scanners, automatic resistor bridge ($$$$$), etc.

You even get Windows 98!!  WooHoo!!

I'd offer seller ~~$70k to get the conversation rolling, and you might score yourself a VERY sweet deal.

[Echo88]

*TiN  looks into his wallet* "hmmmm" 

[hwj-d]

...

By the way -   

If you really want to test resistors correctly, here's a nice kit for a fraction of the "new" price.  If you haven't seen how a precision resistor oil bath works, here's an example:

https://www.ebay.com/itm/Measurements-International-Automatic-DC-Resistance-Bridge-Complete-System/282828252605?

...

  

[Andreas]

Hmm,

a lot of off topic here in the thread aka: how do I scare the younger readers?
(instead of contributing measurements).

- remember: I am doing always relative (not absolute) measurements.
  So in first order only DNL and gain stability counts. (no bulky bridge necessary).

- I am also not afraid of the 1pF/inch stray capacitance when connecting directly to the ADC.
  (there is enough room up to 1 nF at 1K source impedance).

- every baby knows that resistors on a chip have around +/-30% tolerance,
  and resistance doubles with 100 deg C temperature increase.
  So calling for 50 Ohms source resistance is completely bullshit:
  the chip would not work at higher temperatures with such a desing flaw.

So lets go back on topic.

First ratio measurement on Texas Components VHD200#1 (#1 best guess: is marked with 1 dot on the top). DateCode 1804
Red line: ratio deviation calculated from raw voltage against 12.5 : 1 ratio.
around 5 ppm ratio drift over 30 deg C span.

Green: linear approximation of the ratio drift.

blue: hysteresis  = sliding average ( red - green) 21 values with y-scale on the right side.

resulting value is -0.19 ppm/K linear ratio drift.
hysteresis contains also a small resistor drift and timing effects from thermal time constants between resistor and NTC.
(so effectively no significant hysteresis).

2nd picture from today is more like Jasons setup with 0.2 K / minute (5 minutes per deg C) and 12 deg span from 21-33 deg C.
no surprise here with -0.18 ppm/K linear ratio drift.

with best regards

Andreas

[MisterDiodes]

  So calling for 50 Ohms source resistance is completely bullshit:
  the chip would not work at higher temperatures with such a desing flaw.


with best regards

Andreas

As I said -These ADC's work with higher source impedance, but not really at all at full INL spec if you are chasing low PPM...Which is what your wanting even for "relative" tests.  They are not a bad part at all - far from it.  If they are used correctly.

Talk it over with LT if you need to. Or just measure for yourself on an -accurate- setup.  We use a few cal'd 3458a's to monitor actual input signal and Vref while reading the LTC2400 output.  Then add some source resistance (with C and L if you want) and watch what happens on what the ADC is telling you. Whoops.  CURSES, Reality!!  You've smashed up Theory again!!

Yes, you begin to see sub ppm-level shifts at over 50~150 ohms source resistance (at least in our tests across several part production dates), so better to keep source impedance as low as possible.  This is why LT recommends an input buffer amp in almost every application, especially if you need PPM performance.  The Vref input should also have a low impedance source driving it also for max performance - as with any ADC.

AND look at the how the input switch resistance + input diodes leakage (TWO of those remember) and that input capacitor changes with temp.  And look at what happens when you try to get an accurate measure of the CHANGE of resistance on a PWW when you're injecting ~153kHz current into the test. Small numbers I agree...but you are making judgements on very small data changes too.

Yes, it still looks like its working well even with 500 Ohms source resistance, but not when you're looking for small ppm changes in your signal across small thermal events...now the ADC is adding some of those changes for you.  Look closer and you'll see what I mean.

At 1k source resistance I think your INL/Gain / Offset degradation will have a real effect on your measure results - relative or not.  Because you're trying to exract ppm-level data out of your relative tests.

Like I said - the datasheet is a bit optimistic.  Not too bad as long as everything stays at room temp. but you still want a lower source impedance than what you might be using.

No matter HOW you measure them: Ratio resistors are best tested at real bias and a real thermal flow situation for best result.  As much as possible at least.  Your resistor test setup will give maybe some some general (but maybe not accurate) idea of what's happening as resistance changes, but running LTZ ratio resistors on a real LTZ circuit will provide a much more -accurate- and realistic test result.

Have fun!

[ap]

If you really want to test resistors correctly, here's a nice kit for a fraction of the "new" price.  If you haven't seen how a precision resistor oil bath works, here's an example:

https://www.ebay.com/itm/Measurements-International-Automatic-DC-Resistance-Bridge-Complete-System/282828252605?

For those few of you who are not in a position to spend 105k in resistance calibraton gear, how about that:

https://www.ebay.com/itm/GUILDLINE-TYPE-9975-CURRENT-COMPARATOR-RESISTANCE-BRIDGE-W-AMPLIFIER-9460/232224177086?hash=item3611a3bbbe:g:wjkAAOSw241YkOF5

Slightly lower cost version DCC bridge with still remarkable properties (sub ppm uncertainty for 10:1 transfer), if it works, yet more manual handling but comes Windows-free. Sometimes a real diva. Highly recommendable for ultra precision low resistance calibration (up to appr. 100k). Got mine shipped overseas and was (finally) lucky, only limited damage due to the nanovolt amplifier being loose inside and bouncing arround during flight, hitting against the metal can voltage regulators above it. 10 bucks for NOS parts fixed it.

[amspire]

Hardware like the Guildline is fabulous quality, but if you are prepared to do the work yourself, you can make very high performing dividers at a fraction of the cost.

I did make myself a Hamon 10:1 divider years ago, and it was no problem adjusting to within 0.1ppm without any expensive equipment. All you have to do is add a second adjustable 2:1 divider. The 2:1 divider does not need be great resistors at all. You only need resistors that can stay within 0.1ppm for 1 minute. When the calibration is finished, both the 2:1 and 10:1 divider are spot on. The only problem I had was that I used film resistors so the adjustment lasted about 10 seconds at 0.1ppm. With the right kind of resistors and a good construction, around the 1ppm error level shouldn't be hard at all.

[ap]

You cannot get the accuracy of a 9975 with a hammon divider, thats why I use the 9975. Also keep in mind, it is not just the hammon divider transfer accuracy, it is also the transfer accuracy of the meter used that contributes.

[amspire]

You cannot get the accuracy of a 9975 with a hammon divider, thats why I use the 9975.

Why not? Fluke manage 0.2ppm with this method.

Also keep in mind, it is not just the Hammon divider transfer accuracy, it is also the transfer accuracy of the meter used that contributes.

The only meter needed to calibrate the Hamon divider is a microvolt null meter. It is not that hard to make a null meter with a 0.1 microvolt accuracy. You can even eliminate thermal voltages by having the divider powered by a voltage that reverses at perhaps 30 Hz. The null meter can be run as an AC meter for the final adjustment.

Then you can use the divider to transfer the calibration of a 10V reference to the 100V range and the 1 volt range as long as you have a stable voltage source.

Don't get me wrong - having Guildline sort out all the difficulties for you makes your life much easier. The chances of an error are greatly reduced. If you cannot afford $2K+, there are other ways to get the same result.

[TiN]

Mmmm, 0.1 microvolt accuracy is easy? Please, do tell. Perhaps you confuse accuracy and resolution.

[amspire]

Mmmm, 0.1 microvolt accuracy is easy? Please, do tell. Perhaps you confuse accuracy and resolution.

The accuracy needed is 0uV  +/- 0.1 uv and that can be using AC with a bandpass filter if getting 0.1uV with DC is a problem. The resolution is hopefully less then 0.1uV.

[Dr. Frank]

Mmmm, 0.1 microvolt accuracy is easy? Please, do tell. Perhaps you confuse accuracy and resolution.

The accuracy needed is 0uV  +/- 0.1 uv and that can be using AC with a bandpass filter if getting 0.1uV with DC is a problem. The resolution is hopefully less then 0.1uV.

It's getting off-topic here, but I want to point towards a more appropriate description of Hamon type dividers, i.e. the manual for the FLUKE 752A, chapter 3-26 error analysis.

This standard divider uses stable 20V (from a 5440B) for excitation of the Wheatstone Bridge, and a Null reading of <= 0.5µV then is sufficient for about 0.1ppm alignment accuracy of the 10:1 divider, and about 0.2ppm for the 100:1 divider. (Additional errors apply)
Therefore, the Null detector has to have a resolution of about 0.1µV, but it also needs to be quiet (noise free) enough, that's the correct and much more important criterion.

The bias current also needs to be as low as possible, below 1pA.

Both criterions are not at all easy to meet.
You might successfully use a high grade DVM like the 3458A.
Better results are achieved by means of the classic FLUKE 845AB/AR.

If you don't use stable precision resistors for the Hamon divider and the Wheatstone bridge, the whole effort is useless, because otherwise the whole divider drifts too much during calibration mode, already.

I've used my DIY Hamon divider to successfully cross-check the 5442A and 3458A ranges at <1ppm accuracy level. Anyhow, I don't over-estimate the achievable ratio uncertainty.

For resistor ratios, additional current cancellation circuitry is necessary.. I assume that's what this Guildline box also includes.

Frank

[amspire]

This standard divider uses stable 20V (from a 5440B) for excitation of the Wheatstone Bridge, and a Null reading of 0.5µV then is sufficient for about 0.1ppm accuracy of the 10:1 divider, and about 0.2ppm for the 100:1 divider.
Therefore, the Null detector has to have a resolution of about 0.1µV, but it also needs to be quiet (noise free) enough, that's the correct and much more important criterion.

The bias current also needs to be as low as possible, below 1pA.

Both criterions are not at all easy to meet.
You might successfully use a high grade DVM like the 3458A.
Better results are achieved by means of the classic FLUKE 845AB/AR.

A couple of things on the bias current. The null meter does not need to have a high impedance - 100K is probably fine. 1M is fabulous. Even 10K is usable but the sensitivity will be down. That helps a bit. Secondly you can feed the Hamon divider and 2:1 divider while calibrating with slow AC instead of DC and use capacitor coupling. This reduces the leakage current to the capacitor's current. Third, for this kind of purpose, I would make a small battery powered galvanometer rather then go through the extreme lengths HP has to go to in floating the 3458A inputs. For the AC source, I would probably use something like a centre tapped mains transformer - provided the windings matched well.

If you don't use stable precision resistors for the Hamon divider and the Wheatstone bridge, the whole effort is useless, because otherwise the whole divider drifts too much during calibration mode, already.

Yes - you have to have resistors that you have proven stable over the amount of time you are measuring and that have no voltage coefficient - so precision wirewound resistors are perfect. You could possibly slow down thermal effects by putting all the Hamon resistors in an oil bath and then insulating it. You have to check the resistors for drift with maximum applied voltage. The ideal would be to have all the resistors used in the divider to be made with the same wire. If you made the divider with 10 identical resistors, you could actually pick the single low resistor that was closest to the average coefficients of the other 9 resistors. This will minimize drift when a high voltage is applied to the divider.

As I said, the other 2:1 divider used in calibrating the Hamon divider often only has to be stable for a minute. The more stable, the less time you will waste since if the 2:1 divider is drifting, you have to keep trimming it while you are adjusting the Hamon divider. The only point you get zero volts difference is when both dividers are spot on and you can tell which divider is off.

A do-it-yourself approach will use a heap of time since you have to double and triple check everything. You are going to calibrate, do the divide by 10 and then recheck the calibration to make sure there is no drift.

Last thing is the possible errors due to poor construction. When I was testing my crude metal film Hamon divider in a plastic box, I actually got a successful if short lived null at 0.01 ppm error, but the circuit was so sensitive that I had to leave the room and watch from the door to check the match. It seemed to be picking up body thermal radiation. That was my best guess. It was really fascinating to realise I could make a circuit sensitive enough to detect a nearby body with standard 0.1% metal film resistors.

[ap]

The above statements refer actually to two topics:
- precision measurement  and transfer of resistance values (Measurements International and Guildline DCC bridges, as well as SR1010 or Hammon dividers); referring to the thread topic
- Voltage division using Hamon dividers (Fluke 752A)

The Fluke 752A voltage divider is specified with sub ppm accuracy, true. (BTW, the linput bias current of the 3458A, at least the ones in my lab, are above what is usable with the 752A; it draws bias current also at Uin=0V, a Keithley 155 or the like is needed). The 752A is not intended for resistance transfer, given it is only available in one resistance setup. You could use it as part of a bridge (generating a 1:10 voltage), and the other part being one standard resistor and one adjustable resistor, then nulling the bridge voltage. That however is not what resistance transfer is. You want to transfer two fixed resistors, one being known. A 720A could be used here, with adjustable voltage on the one bridge side, but when you do the measurement uncertainty math here, you will find you are in the 1-2ppm pange.

The transfer of resistance using 10 similar resistors (100:1 transfer), e.g. with the IET labs SR1010 series is less accurate.  Besides its transfer accuracy (1ppm + 0.1uOhm) there are two measurements necessary, comparing the standard and the DUT with the SR1010 resistance. As a consequence, sqrt 2 times the transfer accuracy of each of teh two measurements of the ohms meter used adds. So the transfer is 2ppm-ish accurate, at best.

The Measuerments International and the Guildline 9975 DCC bridge are specified at about 0.2ppm and below for a 10:1 resistor transfer, depending on the resistance decades and type used. By far the best way to transfer resistances, thats why they are the default solution in National Labs and highly accurate cal labs.

Of course you can argue, 2ppm is already very precise (and it is), why shoot for 0.2ppm? But these guys (and others) need to derive decade ranges of resistance from one known value (quantum hall standard), and uncertainties pile up when you do multiple transfers.

[amspire]

The transfer of resistance using 10 similar resistors (100:1 transfer), e.g. with the IET labs SR1010 series is less accurate.  Besides its transfer accuracy (1ppm + 0.1uOhm) there are two measurements necessary, comparing the standard and the DUT with the SR1010 resistance. As a consequence, sqrt 2 times the transfer accuracy of each of teh two measurements of the ohms meter used adds. So the transfer is 2ppm-ish accurate, at best.

The SR1010/MT is built a little differently from the SR1010. It can do a 0.1ppm transfer at 100:1 and 10:1. The temperature coefficient however is 1ppm/C so to get the 0.1ppm transfer, you have to watch the room temperature variation and the time taken for the transfer.

The SR1010/MT was available in 1968, but I cannot see any mention of it now. The major difference between the two is that after 2 years, the SR1010 fixed resistors can be different by as much as 100ppm. The SR1010/MT resistors are a little more stable and can be adjusted to match each other within 0.1ppm over the short term. The 100ppm drift only causes significant error with the 10:1 ratio basically because you are cheating when you do a 10:1 transfer - you are comparing 10 resistors to 9 resistors.

The major reason for the SR1010's 1ppm rating for seems to be that when the SR1010 was designed in 1968 or earlier), they decided to rate the resistors for about a 1ppm stability - it was plenty back then. The SR1010/MT has slightly better resistors, but they rated it as 10 times better. It also has a case leakage specification of 10¹² ohms that the SR1010 is lacking.  I think the big factor is the SR1010/MT resistors can be adjusted to exactly match the SR104 10K standard and when you are matching two identical values, it is easy to get a 0.1ppm transfer accuracy for the SR1010/MT.

Basically, a standard 2 year old 10K SR1010 used with a SR104 can end up with an error in the transfers of absolute worse case 10ppm but more typically much less then 5ppm. This big error is  solely due to the cheat they use for making 10K resistance with a 10K transfer box and it is now not necessary. A 3458A should be able to reduce a transfer error to much less then 1ppm in total error even with the SR1010. The accuracy of the 3458A does not matter, only the resolution, the linearity and the variations of measurement over a period of a few minutes.

To sum it up, no matter what it costs, the SR1010 is not brilliant. Definitely not state-of-the-art. Resistors that can drift apart by 100ppm worst case in two years is not stellar performance. It is basically a solid tool when a few ppm is good enough if it is used the way the manual suggested in 1968. I think a cheaper DIY Hamon resistive transfer box (and probably the SR1010) with modern measurement options can do better in 2018.

[MisterDiodes]

Back to a comment on the Vishay guesstimate / ballpark measure above of ~0.19ppm ratio TC..

That's not a spectacular result as far as the cost / benefit ratio of the Vishay Magical Datasheet devices go, since you're looking at what?: - 3X or 4X cost over PWW?

I've got several Pettis ratio resistors here at 13k/1k that do that or better over 15 to 40C, say 0.14 to 0.22ppm ratio TC.  Some Reidon's and GR's  that are slightly worse at 0.18 to0 .25, etc.  They have to be good  thermal buddies, but it does work.  All measured on real cal'd equipment and in real LTZ circuits.

I've even got some matched pairs of 5k dividers that Pettis made that are running at or below 0.1 ppm ratio TC in a tightly coupled thermal setup. (I think around $12~$13 per PAIR in qty).  Those are typically better numbers when you have matched ratio resistor pair values.

All of these work MUCH better than LTC5400's by the way, and at much less noise.

But none of that matters in the end circuit:  The point is that in the right application, QUALITY PWW work fine, and there is absolutely NO REASON to spend 2X, 3X, 4X or more for a Vishay Magical Datasheets (especially for LTZ use) -  Because at DC thru Audio freq you will never see any real increase in ratio performance, and just about zero cost to benefit ratio.

Now for higher freq AC applications, or if you're constrained for space and you aren't worried about increased noise - yes foils have a use, of course.  Foils are not really required for general LTZ Vref apps though - and remember the LTZ is very forgiving even on the absolute resistor values.

[Andreas]

Hello,

I am lagging a bit with the evaluation of the measurement values.
Here are results from Jasons 120R VHP202Z resistors. (all with datecode 1804).
Measurement setup is against my temperature stabilized Z201 1K #3 resistor.
All ratiometric evaluated from the 5V reference of one of my ADCs.

#1 (one dot on the top) 27.02.2018
Box: 1.10 ppm/K  (including noise)
LMS: -0.95 ppm/K @ 25 deg C (but very linear T.C.)
Box LMS 0.95 ppm/K (without noise)

#2 (2 dots on the top) 28.02.2018
Box: 1.12 ppm/K
LMS: -0.96 ppm/K @ 25 deg C (linear)
Box LMS: 0.96 ppm/K

#3 (3 dots on the top) 01.03.2018
Box: 0.47 ppm/K
LMS: -0.31 ppm/K @ 25 deg C (curvy)
Box LMS: 0.35 ppm/K

#3 shows visible larger T.C. on lower temperatures.
Jason measured only the 25-35 deg C part of this.

with best regards

Andreas

[cellularmitosis]

#1 (one dot on the top) 27.02.2018
Box: 1.10 ppm/K  (including noise)
LMS: -0.95 ppm/K @ 25 deg C (but very linear T.C.)
Box LMS 0.95 ppm/K (without noise)

#2 (2 dots on the top) 28.02.2018
Box: 1.12 ppm/K
LMS: -0.96 ppm/K @ 25 deg C (linear)
Box LMS: 0.96 ppm/K

#3 (3 dots on the top) 01.03.2018
Box: 0.47 ppm/K
LMS: -0.31 ppm/K @ 25 deg C (curvy)
Box LMS: 0.35 ppm/K

VHP202Z, 120R, #1, #2, and #3
Results: roughly -0.9, -1.0 and -0.4 ppm/C.

Woohoo!  Glad to see such close agreement between two very different measurement setups!

[babysitter]

But who is right?


Just joking. Notable agreement.

Das

[MisterDiodes]

...OR both results are less than accurate.  Mount the ratio resistor pair so they are at bias and proper thermal flow and sometimes you see a different story.  We've seen that many, many  times.

For instance testing a resistor pair flapping in the breeze you'd swear the TC ratio was positive, then mount 'em in a real circuit on a real board in a real enclosure and suddenly the ratio TC of the resistor pair has gone negative.  Or maybe it doesn't change.

The bottom line: Always test in a real circuit for the most accurate measure result, because you don't know anything for sure until then.

[branadic]

If we had a third independant measurement, with a third independant setup, giving the same results, would you still doubt them?
And isn't it true to characterise a component, even though it might behave somewhat different within a given circuit? Is the component supplier characterizing the component within your circuit? No, he's characterizing the component as it is and chanes are it might behave different in it's final application, don't you agree?

-branadic-

[mimmus78]

If we had a third independant measurement, with a third independant setup, giving the same results, would you still doubt them?
And isn't it true to characterise a component, even though it might behave somewhat different within a given circuit? Is the component supplier characterizing the component within your circuit? No, he's characterizing the component as it is and chanes are it might behave different in it's final application, don't you agree?

-branadic-

If you want, I can do the third run.

Inviato dal mio ONEPLUS A5010 utilizzando Tapatalk

[TiN]

If we had a third independant measurement, with a third independant setup, giving the same results, would you still doubt them?

Even 10 measurements with incorrect data on 10 different setups would not make up for 1 correct measurement .
Component supplier also provides test conditions for all measurement parameters (at least good one try to), so it's up to customer to decipher to see if component fit their requirements in their exact circuit or not. Serious customers also do inbound characterization on their own.

[branadic]

Even 10 measurements with incorrect data on 10 different setups would not make up for 1 correct measurement .

So what do you want to say? That data measured by Andreas are wrong? That you do better then Andreas? Don't get your statement Illya. 

-branadic-

[TiN]

All I want to say that amount of measurements does not matter , it is the accuracy of these measurements and specific setup/circuit requirements is what matters. That is what makes metrology valuable and necessary, not writing down bunch of digits and celebrating when one averaged something agree with other averaged something.

I'm puzzled why the fact that each measurement, especially down in ppm-area, affected by many variables, including temperature, voltage and current bias , way how leads attached, measurement system parameters and even mechanical constrains? That is what MisterDiodes pointed out. Electronics design is not just sum off all characteristics of each individual component, but it's how they work together.