Stability of 5V references compared to AD586LQ

[Andreas]

Hello,

I have found the root cause for the "jumps" at low temperatures.
Found it accidently when trying to add some thermal isolation.
The package is _very_ sensitive to any touching. (package stress).
Jumps of several 10-100uV can be observed.

The fault that I have made is to put the LS8-package directly on the PCB to ease soldering the wire from the other side to the GND Pin.
So the package was touching the PCB (and in between there was the plastic sticker with the number from previous T.C. measurement 10 years ago).

I solved this by rewiring and placing some spacer between PCB and the reference.
After soldering the spacer was removed.
Now I got a "soldering shift" in output voltage and a new fast initial ageing.
So it will take some time until the reference will be more stable.
But I fear there will be still a 3-4 ppm hysteresis.

Another observation was that there might be some temperature sensor drift.
I am using the chip internal temperature dependant trim voltage for the temperature sensing.
And it looks that either the absolute voltage drifts with reference voltage or there is a problem
That I have different reference voltages from LT1236+bridge evaluation
and the 12 Bit ADC (where VRef = VDD from a 5V regulator).

with best regards

Andreas

[Andreas]

Hello,

and the same for ADC#34.

Initial drift is a bit lower because I had more time between soldering and first measurement.

2nd picture: the blue line is from previous T.C. measurement showing the drift by re-mounting.
Against red: the current T.C. measurement. Hysteresis here ~4 ppm. And 1 ppm drift between cycles.

3rd picture: the drift between the temperature measurements red: new green before re-mounting.
But of course there could be also that the sensor8 distance which I use as master temperature
to the LT1236 reference is slightly different due to re-moung.

With best regards

Andreas

[Conrad Hoffman]

Mechanical and thermal stresses play havoc with all references. My plan, as yet unrealized, is to space the device upside down and dead-bug the wiring with fine constantan or manganin. The device is small so the wires should support it just fine. Using ABC (anything but copper) will reduce the thermal flow out of the part and probably increase the stability and lower the power requirements. I'm guessing the added resistance of the wiring will be a constant and not affect anything, but that remains to be seen, depending on the current flow for heating.

[Andreas]

Hello,

Manganin would be good.
Constantan has too much thermal voltage against copper.
And the chips usually have 2-3 deg C more temperature than the environment.
(even when the ground pin is soldered to the PCB).

with best regards

Andreas

[Kleinstein]

Another problem with konstantan wire is that it is hard to solder. Maganin wire is easy to solder.

[Andreas]

Hello,

update from ADC29 + ADC30 with AD586MNZ (DIP8 package).
In the diagrams for humidity (rH on the x-Axis) you can see a drift over humidity and one over time.

on day 113 I had determined ageing and humidity coefficients from the past. see here:

https://www.eevblog.com/forum/metrology/stability-of-5v-references-compared-to-ad586lq/msg5257899/#msg5257899

In daily comparisons to my LTZ#4 (my most stable reference) I tried to predict the readings together with these two coefficients.
so the prediction charts show that there is a more or less good compensation possible.
It looks like the ageing coefficient is decreasing over time (so I get too low predicted values).

with best regards

Andreas

[CalibrationGuy]

Hello Andreas, very interesting data. Your most stable reference, the LTZ#4, forgive me if you've already said it elsewhere, but have you characterized its drift against a known reference like a 732 unit? It would be instructional to indicate the LTZ#4 drift in this graph somewhere.

TomG.

[Andreas]

Hello,

most stable means less than 1 ppm / year.
Calibration points of LTZ are outside the measurement.

If you leave out the outlier on 06.06.2020 (which was influenced by parallel conected instruments during measurement)
the readings from 2016 to 2021 against a 3458A with high stability reference and usually freshly calibrated indicate a drift of -0.9 ppm over 5 years.
(measurement uncertainity 3.8 ppm including transfer linearity).

The outlier is also verified by my most stable ADC16.
(which is drifting nearly equal to LTZ#4 as can be seen e.g. in first post of this thread).
So I have a weekly sanity check between LTZ#4 and ADC16.

with best regards

Andreas

[CalibrationGuy]

Thank you for your reply Andreas, your data is excellent as is the low drift of your reference modules.

Best Regards,

TomG.