Temperature test box for component characterization

[janaf]

Getting deeper into volt-nuttery, I realize I need a test box for characterizing temperature dependency of components like resistors, zeners, LTFLU...

So far these ideas are just beyond sketching.

I have followed Andres thread on TC measurements of precision resistors, beginnig to end. The basic principle seems very good, doing measure relative known characteristics at stable temperautre, using the same reference voltage for the measurement device / DAC as for the test signal for the DUT.
https://www.eevblog.com/forum/projects/t-c-measurements-on-precision-resistors/msg615711/#msg615711

What I want:
- Very good temperature control / stability for DUT. <0.01C?
- Ability to measure sub?ppm levels of changes in voltage / resistance / current versus temperature.
- Moderate temperature range like 0 to 75C.
- Multiple components tested in parallel?
- Moderate physical size. Like a shoe-box?
- Automated test cycle, PC interfaced.

Have got :
- A peltier unit.
- Some big fat chunks/slabs of (tellurium) copper
- Some DAC boards: LTC2498, LTC2442, ADS1282 & DMMs
- Various PID temperature regulators (12V)
- Precision NTCs and LM35
- Various decent voltage references, 5V, 2.5V...
- Galvanic isolation USB interface
- Lots of Li-XX battery packs

Peltier unit:
http://www.ebay.com/itm/THERMOELETRIC-PELTIER-JUNCTION-COOLER-ASSEMBLY-LWKW-0317-/220865136946?pt=LH_DefaultDomain_0&hash=item336c968932

My plan is to:
- Remove the smaller heat sink from the Peltier untit
- Replace it with a copper slab with holes drilled for temperature sensors & cables
- Add screw-on copper slab "lid" with routed cavity.
- Add foam insulation cap surrounding the copper slabs and peltier unit except outer heat sink.

Build a second shielded box, temperature controlled at moderate temperature, for reference components, DAC boards etc.

All electronics could be run on battery but Peltier has to on mains => 12V converter. The temperature controller sensor needs to interface with Peltier via Isolated SSR / FET / transistor driver.

Comments welcome!

To be continued. 

[janaf]

Reserved for measurement results.

[janaf]

Peltier module arrived.

I've split the module using a kitchen spatula and gentle brute force.

The larger heat sink is approx 125x125 mm (5"x5")
The Peltier is the common 40x40mm size surrounded by two types of quite thick insulation foam.

I will add more foam, out to the full 125x125. With 25mm foam on all sides, the copper slab can be 75x75mm. Leaving 7.5 mm copper walls, I end up with 60x60 "test chamber" which I think is quite enough for my needs.

[splin]

Do you need good long term temperature stability? I'm guessing that absolute temperature accuracy is not too important but if you want to compare results over long periods then the stability of the temperature sensor is important. If it is only to be used to characterise temperature coefficients/sensitivities etc. it doesn't matter but it would be a shame to do a lot of testing over long periods and then have relatively large temperature uncertainties such that you couldn't accurately characterise long term drifts of voltage references, resistors etc.

Naturally sensors with proper drift specifications aren't very common. Whilst Platinum RTDs can be excellent,  common low cost ones aren't especially good (or don't specify the drift) whilst lab grade ones are very expensive and difficult to use accurately.

They aren't easy to find, or cheap ($40+), but the best I've found is the YSI (now owned by Measurment Specialities) 46000 "Super stable" series with a claimed typical thermometric drift < 10mK over 100 months (Tamb = 25C or 70C). They also are available with interchangeablity tolerances down to 50mK ($180!). The 45000 series are cheaper but are specificed as < 50mK drift over 10 months (Tamb = 100C).

More easily obtainable are US Sensor PR series, also interchangeable to .05K; however, whilst they spout a lot of hype about their high stability, they don't actually specify it anywhere that I can find. They probably are excellent but it would be an act of faith to rely on it.

Much cheaper are NXP KTY82 and KTY83 diode based sensors. The latter are specified at 1 ohm drift over 10,000 hours (Tamb =175C), equivalent to 128mK at 25C. The drift should be way less when used at < 70C, but who knows exactly how much?

Recalibrating periodically is another option, but horribly expensive if you can't do it yourself and not particularly easy at temperatures other than at the triple point of water. Personally I'd pay for at least one super stable thermistor and use it to calibrate cheaper working sensors such as the KTY83 or an NTC thernistor.

I wonder what sensors Fluke used in their voltage calibrators?

[babysitter]

Don't use bare 100% PWM with low frequency. During the 0V phase, heat will flow back from where it came, and during the 100% phase, you have maximum heat generation due to U*I.
There are dedicated Peltier controller chips, just look at the linear website.

[janaf]

For now; no need for very accurate / long time stable temperature measurements. LT35 and some US sensor NPTs will do. Later, who knows...

Will not use low frequency PWM for temperature.

[janaf]

In the mean time, a simple aging oven:

• A 100 ohm power resistor 10W in a small thermos
• power at 12v (ie near 100mA)
• Stabilizes at near 110C
• With a proper cork it will reach 120C easy.

I'll add a thermostat or controller just in case. It will run on a regular 12V adapter. As it consumes only a bit more than 1W it can be left on for months, years.



 

[janaf]

Quick and dirty #1

I was asked to do some measurements on wire-wound resistors from Pettis and hermetic foils from Vishay. Here are pics of a quick-and dirty setup I made for this, while I'm waiting for components for a more permanent unit.

• A block of copper, 50x50x18mm, 0.4kg
• Two ceramic 100 ohm heater resistors
• Two thermocouples
• DUT, two resisotrs

The stuff above is stuck on the copper block, using semi-soft, heat resistant silicone rubber.

• The copper block and parts will be wrapped with insulation and placed in a large thermos / Dewar
  One of the thermocouples is connected to a PID regulator, output power is via a DC SSR
• One of the thermocouples is connected to a PID regulator
• Output power is via a DC SSR, 0-30VDC, as needed
• The other thermocouple is for temperature logging
• Resistors will be connected 4-wire to DMMs

Sorry, no cooling, no PT-100 available this time.

Data ETA, this weekend.

[janaf]

Comments please!

I have been doing some thinking about measurements of the ppm differences over temperature. Andreas's thread about temperature characteristics of precision resistors uses a clever circuit, using the REF voltage of the ADC for the resistance measurements.

I found something similar in a MAX197 app note AN-270 (attached), figure 5. The nice thing about these circuits is that they compensate for any drift in the reference voltage, by measuring ratios. Probably nothing new under the sun. The current through the resistor / RTD / DUT is given by I=Vref/R_iset (i.e I=2.5V/25K=0.1mA)

Anyway, I rolled my own, with a Instrumentation amp instead. It's attached as RTD_circuit_2.

An example: in the case of a 1000 ohm resistor with 0.1mA current (reasonable?) there will be a voltage drop of 0.1V over the test resistor. This voltage can be amplified by a factor 10 or a bit more to improve SNR etc. With the components / values in the figures, the voltage will be nominally 2.6V before and "always" 2.5V after the resistor.

However, I realized that if we add a very stable reference resistor see RTD_circuit_3, also 1K, with it's own current loop, we can generate another 2.6V reference point, while we can be sure the 2.5V on the lower side is still 2.5V. Now we can connect the 2.6V of the reference and the 2.6V of the DUT, both to the instrumentation amp, amplify by, say, a factor 1000, and get a very much amplified signal from ppm level changes of the DUT.

As far as I can see, the whole circuit will compensate for any REF voltage changes as well as minor changes to the R_iref resistor, while R_ref has to be rock stable over a measurement cycle.

The price is two op-amps, a reference resistor and a negative power rail.

Comments please!

[Andreas]

Hello Jan,

sorry but I do not understand how the compensation works,
and where you will really attach your measurement instrument.
Did you try to simulate this with LTSPICE?

Further point: does the (temperature dependant) gain of the instrument amplifier change the output?

With best regards

Andreas

[janaf]

You somehow reinvented the wheatstone bridge.
Excellent idea, but nearly 200 years too late!

Damned, I thought I was first  Yes, it's a kind of Wheatstone bridge.

[janaf]

Hello Jan,

sorry but I do not understand how the compensation works,
and where you will really attach your measurement instrument.
Did you try to simulate this with LTSPICE?

Further point: does the (temperature dependant) gain of the instrument amplifier change the output?

With best regards

Andreas

Hello Jan,

sorry but I do not understand how the compensation works,
and where you will really attach your measurement instrument.
Did you try to simulate this with LTSPICE?

Further point: does the (temperature dependant) gain of the instrument amplifier change the output?

With best regards

Andreas

The ADC connects to out, ref and ground.

The measurement is between the known voltage on the positive side of the Ref resistor and the positive side of the DUT. So I'm measuring the variation of the DUT (over temperature / time), not the actual voltage/resistance over it. The big fat assumption is that the rest of the circuit is constant enough to make meaningful measurements.

However, the good part is that the change in voltage over DUT needs to be measured with a few digits accuracy only. Instead of measuring the voltage over the DUT down to single ppm, I'd be measuring the variation; something like 2 digit accuracy and 3 digit resolution should be enough to see the characteristic of the resistor.

As you point out, the gain of the InAmp is a weakness & error source. I must have a highly stable resistor there.

On the positive side; the noise from the voltage reference would be common mode to the InAmp and therefore largely suppressed.   

SPICE: coming...

[babysitter]

Consider making a second (quite) thermostatted environment where you put all the parts that you want stable, e.g. your differential voltmeter amplifier, the thermostat circuit, reference resistors...


Hendrik

[mzzj]

Do you need good long term temperature stability? I'm guessing that absolute temperature accuracy is not too important but if you want to compare results over long periods then the stability of the temperature sensor is important. If it is only to be used to characterise temperature coefficients/sensitivities etc. it doesn't matter but it would be a shame to do a lot of testing over long periods and then have relatively large temperature uncertainties such that you couldn't accurately characterise long term drifts of voltage references, resistors etc.

Naturally sensors with proper drift specifications aren't very common. Whilst Platinum RTDs can be excellent,  common low cost ones aren't especially good (or don't specify the drift) whilst lab grade ones are very expensive and difficult to use accurately.

They aren't easy to find, or cheap ($40+), but the best I've found is the YSI (now owned by Measurment Specialities) 46000 "Super stable" series with a claimed typical thermometric drift < 10mK over 100 months (Tamb = 25C or 70C). They also are available with interchangeablity tolerances down to 50mK ($180!). The 45000 series are cheaper but are specificed as < 50mK drift over 10 months (Tamb = 100C).

More easily obtainable are US Sensor PR series, also interchangeable to .05K; however, whilst they spout a lot of hype about their high stability, they don't actually specify it anywhere that I can find. They probably are excellent but it would be an act of faith to rely on it.

Much cheaper are NXP KTY82 and KTY83 diode based sensors. The latter are specified at 1 ohm drift over 10,000 hours (Tamb =175C), equivalent to 128mK at 25C. The drift should be way less when used at < 70C, but who knows exactly how much?

Recalibrating periodically is another option, but horribly expensive if you can't do it yourself and not particularly easy at temperatures other than at the triple point of water. Personally I'd pay for at least one super stable thermistor and use it to calibrate cheaper working sensors such as the KTY83 or an NTC thernistor.

I wonder what sensors Fluke used in their voltage calibrators?

One of the better options amongst Pt100 sensors would be this http://www.distrelec.de/en/Resistance-thermometer-TDI-P100-3045/p/17668845
Pricing is around same ballpark or cheaper than YSI 4600 series.

[janaf]

Consider making a second (quite) thermostatted environment where you put all the parts that you want stable, e.g. your differential voltmeter amplifier, the thermostat circuit, reference resistors...
Hendrik

Will do!

[blackdog]

Hi, Some remarks from me :-)


I made some months ago a temp sensor calibrator, and klik on the link to see the topic on the Dutch Forum Circuitsonline, use google translate!
http://www.circuitsonline.net/forum/view/123403

If you want good control over the temperature, then it is important that you have a good link between the sensor and heater.
I use the NTC for controling the temperature and de PT1000 between the holes for the D.U.T. goes to the TEK DMM4050 Multimeter for checking of the set temperature.

I hoop this info helps :-)


Kind regarts,
Blackdog

[janaf]

Thanks.

If you go back to post #7 you can see the quick-and-dirty thermal block I'm using to start with, waiting for the real thing.

[janaf]

Tatatata!

I'm doing real time measurements on the forum now!  I have set up a quick and dirty test with two resistors:

• One 1K, 1%, wirewound from Pettis
• One 1K, 1%, VHP101T hermetic foil from Vishay

Look in post #7 for some more details on the setup.

The temperature is set with a PID controller. I wait until the set temperature has been reached, wait at least five minutes, then take an average over one minute.
 
UPDATE: Yo, there was something wicked going on with the first measurements of resistor B. I have updated the attached file, it looks much more consistent now. I can't really explain what happened with the first series.

UPDATE: Finished for today. Well after midnight here. I attach an updated file.

Final plot: I attach the final version, have removed some garbage from the first plot. This is as good as it gets with a quick-and-dirty setup.

• I really need to automate the process and finish the final setup. It took 8+ hours to do it manually. Not realistic to do that except in very special cases.
• I'm quite sure it would be sufficient to 5-7 temperature points in all, to see the general trends.
• I need to do measurements in a temperature controlled of the room, these weren't. At these levels (ppm) the room temperature influences the DMMs (by 1-2ppm/C)
  

Bottom line: With the box method, resistor A (Pettis) has a TCR of +0.7ppm/C while resistor B (VHP101T) has a TCR of -0.24ppm/C over the measured range. Hysteresis was very low or zero for both, levels below measurable in this setup.

PS Taking things apart showed that one of the leads of the Pettis resistor was touching a thermocouple which explains the early "strange" results. After fixing that, the resistor behaved very well, much better than specified.

PS2: The price of the VHP101T is about five times higher than the Pettis resistor.

PS3: As the absolute accuracy of the resistors for the LTZ1000ACH is not important, I specified each at 1% tolerance. Both where delivered better than 0.01%!

According to my DMM: Pettis; 1.00009 kOhm, VHP101T; 1.00002 kOhm

[janaf]

So I'm finished now. It was a good learning experience and gave me ideas for the later, final, setup.

The Pettis resistor performed much better (0.7ppm/C) than specs (5ppm/C?) at this temperature range.
The VHP101T result was near its "typical" specification of 0.2ppm/C with best part of the curve lower than room temperaure.

Comments?

[Dr. Frank]

..
I'm looking for a replacement "master" resistor for my HP3458A.  This is currently a hermetically sealed S102C resistor from VPG [or whatever the production 3458A's received].  If I replace this 40K resistor with a more TCR stable one [and recalibrate], then the resistance function in the DMM will automatically be much better.

If I order a big pile of 80K resistors from Edwin [with his special extra-stability procedures added], there are sure to be a pair of them that [in parallel] will give me a very stable 40K.

 Hi Ken,

I'm also searching for a better 40k resistor for my 3458A, but your statements are incorrect in several aspects.

The 40k needs to have a T.C. as low as 0.1ppm/K (currently specified  ~1.3ppm/K)
This will improve  the "Temperature Coefficient With ACAL" parameter ONLY! (by a factor of 10)

Anyhow, all Ohm ranges will still have a very bad T.C. of 3ppm/K, prohibiting a useful Ohm Transfer Stability specification.

To really improve the Ohm mode for T.C., you would have to replace ALL resistors which are involved in the Ohm measuring function, by low T.C. ones, i.e. BMF types R307 = 300, R308 = 3.00k, R310=10.00k,  and the complete resistor array RP300, containing 30k, 60k, 600k, 6M, 10k, 0.3k, 3k. If I did not forget any, that would make 10 additional low T.C. resistors, see picture.

Additionally, because the timely drift specification is so bad, the 40k would need to be a hermetically sealed, AND oil filled type. VHPxxx types give a typical annual drift of 2ppm/6years.
PWW resistors are not suitable, as these have typical drift rates of 20ppm/year, maybe below 5ppm/year for selected ones.

This 40k resistor is part of the DCI / current ranges, which are also quite unstable, maybe also affecting the ACAL OHM function.
Therefore, I would also like to replace most of the current shunt resistors by <1ppm/K ones.
At least, R 207=40k (0.1ppm/K 2ppm/6yr.) , R206 = 500k, R208 = 4k53, R209 = 634, R210 = 90, R211 = 9, R212 = 1 Ohm, 2W.

Frank

[janaf]

Consider making a second (quite) thermostatted environment where you put all the parts that you want stable, e.g. your differential voltmeter amplifier, the thermostat circuit, reference resistors...

I realize I have the same problem for my DMMs, in case I use those, like I did in the Q&D measurements. There are reference resistors in the DMM that are temperature sensitive. It seems like the measured resistance value changes by 1-2ppm/C. I can compensate by running internal calibrations now and then but don't know how far that helps. I should put a thermostat on the fan of the DMM mainframe. I have seen the 3458A owners have similar problems.

[babysitter]

I know in the meantime I must appear like a DB6NT sales pusher, but let me point to the QH40A (QH60A) crystal heater again to temperature stabilize small things, e.g. reference resistors!

[Dr. Frank]

I realize I have the same problem for my DMMs, in case I use those, like I did in the Q&D measurements. There are reference resistors in the DMM that are temperature sensitive. It seems like the measured resistance value changes by 1-2ppm/C. I can compensate by running internal calibrations now and then but don't know how far that helps. I should put a thermostat on the fan of the DMM mainframe. I have seen the 3458A owners have similar problems.

No wonder, as the Ohm measurement topology is very similar between all the 6 1/2  7 1/2 bench meters and the 3458A, see my description just above!

In the lower grade meters, even worse quality resistors than in the 3458A are used.
Therefore, you won't get these meters stable in Ohm mode...

It's a pity even for the 34470A, as obviously they used an even  better 10k BMF reference resistor than in the 3458A, but as the specification clearly indicates, they don't manage to either transfer its high stability to Ohm modes (due to downgraded ACAL functionality), nor did they use appropriately temperature stable resistors for Ohm mode..

Frank

[janaf]

One problem is that to do a TCAL, you need to know the TCR of the resistor, either by having a very repetitive supply or by measuring the resistors individually.

The best is of course to solve the root of the problem but with my DMMs I can't upgrade the resistors, there's no documentation available, at least not that I can find. :-|

My DMM modules have on-board temperature sensors, readable by software so I could make a curve fit and correct output for temperature. But I might have to calibrate the dependency for each resistance range.

Or build a temperature controlled cabinet for the whole thing.....

[Andreas]

Hello,

when looking closer at your RTD_circuit_3.png I can only see one regulated current through the DUT.
The regulated current of DUT is proportional to the VRef and proportional to the 25K resistor.

The Reference resistor has at the lower pin the same 2.5V.

But the current in reference resistor is not regulated:
At the upper pin I see that U3 is amplifying the offset of U1 with his open loop gain.
(So actually amplifying the difference of offset of U1 + U3).

Or do I get anything wrong?

So which of the setups do you use for your T.C. measurements.
And how much offset drift do the OpAmps have?
A standard OpAmp has about 5uV/K offset drift. So with 100mV over the DUT resistor 1 deg C temperature difference of a standard OpAmp will give about 50ppm/K error.

With best regards

Andreas