Lowest drift, lowest noise voltage reference (ADR1000AHZ)

[branadic]

It's not just me, the temperature effect is real.
I used WebPlotDigitizer to get the samples from your plot and Plotter to remove drift by applying a hyperbolic fit, which worked better than the parabolic fit, though I know both aren't the right choice, but will do for the moment.
If I then replot the data with the drift removed from them we can clearly see a temperature influence. As it appears in similar amount on all three datasets, is it systematical and coming from your measurement device? If you would remove that temperature dependency from the data, how does the drift curve appear?

-branadic-

[Andreas]

Could you plz. plot drift (with the drift rate itself removed for sure) vs. ambient temperature? Is it just me or is there a small correlation?

I have drawn values over temperature from day 705 where the drift is rather stable.
I see no direct correlation drift over temperature on the ADR1000 measurements.
(the T.C. is adjusted to -6 ppb/K and -4.5 ppb/K on the 6.6V outputs and +20 ppb/K on the 10V output).
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5138967/#msg5138967

The ageing measurements on ADR1000A are always corrected by a LTZ#4  measurement.
(my most stable reference over time with -21 ppb/K)
So also the temperature coefficient of the 4 ADCs + both DMMs are corrected.

But I have to admit that this reference measurement on LTZ4 is the weak point.
If it fails, you have a common cause error which is also visible in LTZ#9 measurement.

When you look at the "dent" on day 726 it is not temperature related.
In the averaged measurements of the raw values on 15.03.2024 I can see that ADC#15 and ADC#17
measurement failed for some reason on LTZ#4 compared to 08.03.2024.
(not temperature related, probably bad contact or EMI related during measurement)
So normally I should exclude this measurement on day 726 from data.

with best regards

Andreas

[branadic]

I think the temperature dependence is obvious in your previous post, at least to me.
I can see that temperature dependence subtracting different fit functions from your data, such as natural log y=A*ln(x/B) with:
A=-1.0447910452e+00 and B=5.2486706950e+00 for ADR#01 10V corr
A=-1.6667803966e+00 and B=1.0955854281e+01 for ADR#01 Vz corr
A=-1.7280758581e+00 and B=1.2283777186e+01 for ADR#02 corr

or cubic function y=ax³+bx²+cx+d. You can check that yourself using e.g. https://curve.fit/
What is left clearly correlates with your ambient temperature.

-branadic-

[iMo]

@Andreas: when you look at the graph from a distance you may see - when the temperatures are typically "higher" the curve's downwards trend (the trend which may fit a function branadic has extracted) is slowing down a little bit, when the temperatures are typically "lower" the curve's downwards trend is speeding up a little bit..

[aronake]

Stock levels at analog.com of ADR1000 is gradually going upward. In July i saw 52, two weeks ago there was around 127, and now 444. So seems production is ramping up.

After some fiddling around I managed to order 5 last week. First I managed to get quotes but delivery was rejected. I though that could have to do with me living in Hong Kong. But contacted Analog customer support to ask the reason, and they told i had to order from a company e-mail. I have a company here in Hong Kong, so switched log-in to company e-mail and used company e-mail for order and then worked and i got a confirmed order. Two days later i got a tracking number, but items not sent yet. Seems sending date is scheduled for 29th now after two updates from analog. So looks like things are moving, but can not be sure until the ADR1000 are here....

[aronake]

344 in stock now.

Who has ordered recently and how has it been going? Got delivery?

My status with the fedex trackning number is still Lable Created after a week, so seems Fedex has not received the package yet. Thats quite strange

[aronake]

344 in stock now.

Who has ordered recently and how has it been going? Got delivery?

My status with the fedex trackning number is still Lable Created after a week, so seems Fedex has not received the package yet. Thats quite strange

After a week with the fedex shipment in status "Label Created" I reached out to Analog customer support to be told first that the order was cancelled "due to trade compliance". When asking if they could give more details as I much rather buy from them than from unauthorized dealers as no authorized dealers have ADR1000 in stock I was told "ADI eshop cannot cooperate with your company".

In the meantime I managed to get a ADR1000 from an unauthorized dealer. But that one is very far out of specifications when it comes to stability. Either it was broken or some reject for quality issues or fake. If fake, someone has put lots of effort in making a fake. It has realistic characteristics on the current draw on the heater and from time to time, it can keep within 5 uV peak to peak for some hours. The board I am using it on is very close to reference design, instantly worked with LTZ1000 I tested it on. Some slight modifications made now to be closer to ADR1000 component values.

I can understand that Analog do not want to sell to everybody, but at the same time they are promoting the terrible business companies of selling fake and broken components.

[Andreas]

Hello,

I now passed 24 kHrs:

no clear direction of drift.

Detailed explanation of the diagrams see post of 14 kHrs
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5138967/#msg5138967

with best regards

Andreas

[Andreas]

Hello,

I did some interpolations to better approximate drift.

First: Error curve of coefficient for SQRT(Time)
After subtracting the (overall averaged) interpolation from the measured value the error curve shows different slopes before 7.5 kHrs and after 7.5 kHrs.

2nd: after leaving off the first 7.5 kHrs and calculating the error curve for the remaining part over time.
3rd: the same over temperature (with offsets of 0.5 ppm to better show the individual temperature slopes).
ADR#1 10V shows a significant slope whereas the other slopes are much lesser or hidden within measurement noise.

4th: simultaneous interpolation over SQRT(Time) and Temperature with time as X-AXIS.
5th: the same over temperature. (showing the measurement noise).

Results for Tempco of combined interpolation.
ADR1000#1 6V6: 18 ppb/K interpolated
ADR1000#1 10V: 39 ppb/K interpolated
ADR1000#2 6V6:  1 ppb/K interpolated

Each measurement is done relative to LTZ#4 which has -21 ppb/K (so this negative drift increases the result by +21 ppb)
The dertermined T.C. values for the used references and the expected T.C. for the ageing measurement is as follows.
ADR1000#1 6V6:    -6 ppb/K -(-21ppb/K) = +15 ppb/K expected
ADR1000#1 10V: +20 ppb/K -(-21ppb/K) = +41 ppb/K expected
ADR1000#2 6V6:    -5 ppb/K -(-21ppb/K) = +16 ppb/K expected

But of course there is a uncertainity of several ppb/K in the measured values..

with best regards

Andreas

[Andreas]

Hello,

I did some interpolations to better approximate drift.

First: Error curve of 3rd order (cubic) interpolation over time (kHrs)
After subtracting the (overall averaged) interpolation from the measured value the error curve shows outliers up to 1 kHrs so I left off the first kHr for the following interpolations.

2nd: after leaving off the first 1 kHr and calculating the error curve for the remaining part over time.
3rd: the same over temperature (with offsets of 0.5 ppm to better show the individual temperature slopes).
again ADR#1 10V shows a significant slope whereas the other slopes are much lesser or hidden within measurement noise
but seem to get a higher order tempco at the edges due to the 3rd order ageing coefficients.

4th: simultaneous interpolation over time (3rd order) and Temperature (linear) with time as X-AXIS.
5th: the same over temperature. (showing the measurement noise and the seemingly nonlinear T.C. behaviour at the lower edge).

Results for Tempco of combined interpolation.
ADR1000#1 6V6: 29 ppb/K interpolated
ADR1000#1 10V: 42 ppb/K interpolated
ADR1000#2 6V6: 12 ppb/K interpolated

The expected values are as above.

with best regards

Andreas

[Kleinstein]

The is quite some correlation between the ADR measurements. A correlation between the 10 V and 6.6 V from the same ref. is no surprize, but there is also quite some between the 2 references. The correlated part could be from the ADC, divider and it's reference or possible some common influence like EMI or pressure or humidity. With the rather sharp jumps humidity is not that likely.

[Andreas]

As already mentioned:

But I have to admit that this reference measurement on LTZ4 is the weak point.
If it fails, you have a common cause error which is also visible in LTZ#9 measurement.

When you look at the "dent" on day 726 it is not temperature related.
In the averaged measurements of the raw values on 15.03.2024 I can see that ADC#15 and ADC#17
measurement failed for some reason on LTZ#4 compared to 08.03.2024.
(not temperature related, probably bad contact or EMI related during measurement)
So normally I should exclude this measurement on day 726 from data.

with best regards

Andreas

[branadic]

Here is an update of my ADR1000 reference against two Fluke 7000. The reference appears to be stable, but one of the F7000 drifts more than the other, still within spec though.

-branadic-

[Andreas]

Hello,

I recently passed 25 kHrs:

There seems to be some common cause drift on all measurements
-> obviously my LTZ#4 (reference for relative drift in this case) drifted by 0.2 - 0.3 ppm during the last 1 kHr.

Detailed explanation of the diagrams see post of 14 kHrs
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5138967/#msg5138967

with best regards

Andreas

[Andreas]

Hello,

I now passed 26 kHrs on ADR#01/#02

no new comment against last post.

Detailed explanation of the diagrams see post of 14 kHrs
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5138967/#msg5138967

with best regards

Andreas

[lichen867]

login & reply to experss my respect to you.

[Halcyon]

login & reply to experss my respect to you.

Don't pretend to be from somewhere you're not. Clearly a spammer/illegitimate user.

[branadic]

Here are some recent plots of my ADR1000 against 2x F7000.
We can spot a few things: the moment when a room-controller to the heating was installed (around day 1200) and the moment when I've connected an additional scanner with a second meter to the reference some 50 days ago.

-branadic-

[Andreas]

Hello,

I now passed 28 kHrs on ADR#01/#02

temperatures are rising in my lab and also humidity is rising.
So the measurement values get a bit unstable.

Detailed explanation of the diagrams see post of 14 kHrs
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5138967/#msg5138967

with best regards

Andreas

[Andreas]

Hello,

I now passed 29 kHrs on ADR#01/#02

No visible drifts of references and 10V voltage divider. (all within my lab limits).
Since all is more or less stable I left off the sqrt(time) diagrams.

Detailed explanation of the diagrams see post of 14 kHrs
https://www.eevblog.com/forum/metrology/lowest-drift-lowest-noise-voltage-reference/msg5138967/#msg5138967

with best regards

Andreas

[dietert1]

The sub-ppm stability of the divider is quite impressive. And your tests demonstrate the behavior that i found in my LTFLU experiments some years ago and that seems to be the trick with those Fluke zener references: Continuous operation for years including battery backup. There are some really large time constants involved.

Regards, Dieter

[branadic]

Here is an update from my setup. The ADR1000 reference and F7000-2 are either very stable or exhibit at least the same relative drift.

-branadic-

[dietert1]

We have to keep in mind that even an apparent continuous drift like the one between branadics two F7000 references is compatible with a very long time constant exponential decay. In fact those long time constants are a consequence of very (almost perfect) hermetic enclosures and very stable (almost perfect) temperature sensors.
The last example of this i have are near hermetic enclosures that i made with Weidmüller Klippon aluminum enclosures. Those have rubber seals that result in time constants of some weeks for air pressure containment. They are able to suppress atmospheric pressure changes of 30 mBar as caused by weather down to changes of a fraction of a mBar. Still it means if i close one of those boxes, i have to wait for a day with medium air pressure. Otherwise i will have a delay of some months with initial pressure drift if i want stable air pressure inside.
Now if you think about an industrial metal IC package with glas seals, the time constant will likely be much longer, although the air volume inside is small compared to those Weidmüller enclosures.
I seem to remember that vacuum tubes needed some getter to trap gas atoms in order to maintain good vacuum over the years. Those gas atoms evaporated from heated metal surfaces or maybe entered by imperfect hermetic seals of the tube electrodes.
If one doesn't have such a trap one will see the seasonal modulation present in Andreas' log of the divider ratio. And the better the device package, the smaller the modulation will be and the longer will be the initial time constant.

Regards, Dieter

[Dr. Frank]

Here's my last drift diagram of ADR#3. After adding a small transformer in my 3458A, it will go in there, making it a 9 digt instrument, at 6ppb StD.
You can see the EMP event by a nearby thunder strike at July 1st, and I have assembled it finally in a proper double shielded case, which caused the negative shift.
I already reported on its creeping characteristic, requiring 2 weeks to recover from an elongated time-out.
This would render the ADR1000 useless for any use in a DMM, or you would need to add a standby power supply.
Unfortunately, no other volt-nuts have repeated my tests, so I don't know yet, if this behavior is unique to my sample.

When I recently made all the changes during testing noise performance of my 3458A, and heating it 5°C higher, I recognized, that even after a 1 day long interruption the ADR shows a long termed creeping effect as well.
The +5°C change lets it creep even after 2 days.
I discussed with John R. Pickering about his tests on modified F7000 reference assemblies.
He also tried to anneal the die attach of the ADR by using his modified heating process in his mothers oven.
This did not work, as well the F7000 hysteresis removing method did not work, especially as the ADR does not show any hysteresis effect after a big temperature excursíon.

So I conclude, that creeping behavior is a common characteristics of all ADR1000.

The die attach of the LTZ1000 is an epoxy, and obviously has the characteristics of being solid, non elastic.
The baking process stresses and cracks weak inner bonds in the glue, so it does not change any more it's elastic properties.

The ADR1000s die attach can be described as being a viscious, elastic fluid, like tar.
Such a fluid does not show hysteresis, but it takes a lot of time to creep from one tension equilibrium to the other.
I did not yet do a reasearch, which die attach material shows exactly this behavior.

I will ask Analog and make the proposal of an ECR, to change to the LTZ1000s adhesive.

Frank

[Andreas]

I will ask Analog and make the proposal of an ECR, to change to the LTZ1000s adhesive.

I hope they do not find out that the ADR die attach is cheaper and change to the ADR adhesive on LTZ1000

with best regards

Andreas