Lowest drift, lowest noise voltage reference (ADR1000AHZ)

[Andreas]

Frank: I fear they dont even know at ADI; when looking on the RoHS declarations I can find data for ADR1000A but none for LTZ1000

[Ismsanmar]

This would render the ADR1000 useless for any use in a DMM, or you would need to add a standby power supply.

Perhaps that's why the 8 1/2-digit multimeter ADCMT introduced earlier this year have a battery-backed reference voltage source.

[iMo]

Isn't it because people run the ADR1k on lower temperatures? Perhaps ADI has characterized it at a higher XX degC temperature where the plasticity of the die glue performs in a slightly different way..

[Kleinstein]

Essentially all polymers show the stress relaxation over time. Compared to metals they can not withstand high mechanical stress over a long time. For polymers room temperature is already a relatively high temperature, not too far from the melting point or glass temperature. There are still difference in the time scale and glass temperature. Also the temperature can make quite some difference: at a higher temperature the relaxation can be much faster. So there can be an advantage to running the reference rather hot.
It is not only the epoxy that can make a difference but also the filler (e.g. the bubles with the LTZ1000A) they effect the thickness and local stress.

Another possible difference is how much the chip reacts to stress. This can be different between chips.

[aronake]

LCSC have ADR1000 in stock.

I would think they are cautious on making sure they sell genuine parts and not rejects from binning, but who knows..... Buy at own risk...

https://www.lcsc.com/search?q=adr1000&s_z=n_adr1000

[exe]

Out of curiosity, is it possible to float the die? On bond wires, dead-bug style? Though production cost would go up, I expect.

[KT88]

No. Self-heating would overwhelm any benefits from that, unfortunately...

[Andreas]

Hello,

I now passed 30 kHrs on ADR#01/#02

No visible drifts of references except some measurement noise.
The 10V voltage divider seems to show some seasonal changes.
Temperature or humidity dependant.

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 32 kHrs on ADR#01/#02

No visible drifts of references except some measurement noise.
The 10V/6V6 divider drift is also very stable for a non hermetically resistor.

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

[miro123]

Unfortunately, the ADR1000 has long been unavailable through local distributors like Mouser and Digikey.
Even more bad news for hobbyists like us: my order for the ADR1399 EVM was canceled by Mouser. Due to recent geopolitical changes, the EVM can no longer be sold to private customers in the Netherlands. A valid company tax number and company address are now required.

[Adilpickles]

Dang, that sucks. In the US, mouser & digikey are also completely out of ADR1000s; fortunately, I was able to use my university email and order directly from ADI, even though I am just a student. The LTZ1000a has been hell to get a hold of though.

[exe]

There are a few ltz1000 on lcsc, though very few.

[branadic]

I'm about to hit 1600 days of non-stop observation on my ADR1000 reference, here are recent plots.

-branadic-

[Andreas]

Hmm,

so it looks that F7000-2 is more stable than F7000-1

with best regards

Andreas

[branadic]

Andreas, that is also my conclusion, yet F7000-1 is still within spec.

-branadic-

[Dr. Frank]

Here's my drift diagram for my only build, the ADR #3.



I'm not interested anymore in the overall stability / drift of the ADR, as I have completely finished my studies about stability of buried zener references, have proven the nature of the LTZ 1000 as a Deterministic / Absolute Reference, and now I only would like to improve the characteristics and specification of longscale DMMs, like my 3458A.

The ADR would offer the opportunity to realize a usable 9-digit DMM, i.e. rms noise in the 0.001ppm range, INL 0.02ppm, and drift lower than -0.6ppm/yr., By implementation of the Deterministic Reference by adding a secured real time clock, idle power supply and battery backup, these design changes would yield an Absolute Voltmeter. Its 10V range uncertainty specification would simply be:
 0.01.. 0.001ppm over Lifetime
  +/-0.02pmm of Range   

I have modified my ADR #3 by as well adding a proper 2nd electrical shield around the inner GUARD magnetic shield (with Andreas EMC capacitors), and doubled the zener current for achieving lower noise. T.C. is not yet aligned to zero.

This initiated a new, but faster inrush drift.     

As said, the ADR #3 with its given bad die attach is not usable at all for DMMs, only with the described power back up. AD should make a Design Change on that.

I've heard nothing from Illya, or MiDi, concerning their tests to bring down the noise level of the 3458A (together with the necessary modification with EMC capacitors for the A9 board), by changing the +/- 12V and 5V reference OpAmps and using a 4fold ADR reference.

Has anybody heard about that activity?

Frank 

[Echo88]

Was the PTB or a similar organization in any way used to determine the actual performance of your modded ref against a JJA Frank?
Did Mr Palafox reply to your mail meanwhile?

[Dr. Frank]

I have physically proven the principle by my 20 year long studies and experiments about the stability of the LTZ1000 reference. With my limited possibilities, I can demonstrate currently an uncertainty of 0.02ppm/yr.
A former Physics fellow student from RWTH Aachen, who works as a Technology expert on discrete semiconductors, and who himself designs such semiconductor structures, has explained me completely the build of a buried zener diode.
This was the ultimate theoretical proof of concept, i.e. that the LTZ1000s drift rate is non stochastic, but deterministic.
I got in contact with PTBs dept. for electrical units and will get the possibility to monitor one of my LTZ1000 references for three months vs. their JJ array.
This will set the initial voltage value to an uncertainty of 0.5nV/V.
I assume, that it will be possible to further reduce the drift rate uncertainty by another factor of 10..100, so to determine the upper uncertainty limit, i.e. to 0.001ppm absolute or per year.
Only problem would be that parasitic effects, like residual T.C., humidity and air pressure sensitivity come into play.
The LTZ1000 chip itself might still show random walk effects, but which are of reversible nature, whatsoever I have observed over this said time period.


I already have modified my FLUKE 7000 reference, by removing the 1V output and instead making the 7.1591V directly accessible at the former 1V jack.


This modification will as well be possible for any 732X voltage reference, so that LTFLU based references can easily be turned into an Absolute Reference.


Frank

[Kleinstein]

Even if most of the dirft part is deterministic, it is difficult to accurately correct it. One is limited in how accurate the drift can be measured. Especially in the early time there can be additional faster contributions. So one could no use the initial (e.g. a few months) data to estimate the drift parameters. Not all of the LTZ1000 will drift the same and thus individual testing would be needed.

[Dr. Frank]

Even if most of the dirft part is deterministic, it is difficult to accurately correct it. One is limited in how accurate the drift can be measured. Especially in the early time there can be additional faster contributions. So one could no use the initial (e.g. a few months) data to estimate the drift parameters. Not all of the LTZ1000 will drift the same and thus individual testing would be needed.

Sorry Kleinstein, but you obviously have STILL not understood, that your objections are completely unfounded.
Once again, with my LIMITED possibilities, I ALREADY have demonstrated an uncertainty limit for the drift rate for each individual reference (9 EA) to 0.02ppm/yr, over 9 years of observation. Latter time period only made it possible for me to achieve such low uncertainty.

The PTB anyhow, will be able to confirm and improve this uncertainty limit, as their JJ array allows 1000 times lower transfer uncertainty, and probably 100 times lower transfer noise, by use of e.g. an HP34420A, instead of my HP3458A.

Frank

[dietert1]

When i reported some years ago about my LTFLU experiments and found that the previous belief about sqrt(time interval) variation should be replaced by a rather predictable drift, everybody was shouting at me. At the time i proposed the image of a radioactive decay involving different substances with different time scales. Deviations between my LTFLU references after subtraction of the dominant exponentional was predictable to about 10E-8 (more or less the measurement noise).
This seems to hold for well made buried zener references run continuously with battery backup, e.g. those Fluke 732x references. At the time i did not have battery backup. My references were made to run at near ambient temperature and power breaks caused sub-ppm deviations.
Meanwhile Andreas and others showed many many examples with this behaviour, still with different measurement noise limits. Even a setup with a 1 ppm noise limit can reveal the drift decay over several years.

Regards, Dieter

[Dr. Frank]

Hello Dieter,
evidently I do not belong to these "Everybodies", so I'm very interested in your findings, as the LTFLU should show a very similar drift behaviour like the LTZ1000, because both are based on the same or similar semiconductor technology of that time in the 80ties.
Have you described your findings, i.e. the drift rate characteristics, somewhere here on eevblog?
Have you really seen an exponential drift rate, or more or less a linear one?

The shape of 732Xs drift is also not exponential, but the stochastic drift of the gain resistors is overlaid, what I have seen from some examples.

I also can't figure out, what you mean with an 'exponential' drift rate, as this would imply, that the drift would 'explode' over time. Or do you mean 'logarithmic', instead?

What do you mean with a 1ppm noise limit?
By Andreas improved LTZ1000 design, we achieve a 0.02ppm noise limit for the LTZ1000, which I also would expect for the LTFLU circuit. Only by this low enough noise limit, and reduction of other fluctuations, it was possible to make this discovery, actually.

Thank you, Frank

[dietert1]

I mentioned radioactive decay as an example and it should be obvious what kind of exponential is meant. And yes i reported some results of my tests here in this forum. One might use the search.
I just found:
https://www.eevblog.com/forum/metrology/ref70-new-from-ti/msg3555645/#msg3555645
There is some TI paper on the subject, too.
Your plot above just demonstrates the rapid decay of disturbances introduced by your handling, while the whole observation exhibits a drift with a much larger time constant. In my LTFLU observations i saw a time constant of several thousand hours. These slow processes are well predictable for a comparable timescale in the future.

Regards, Dieter

[Andreas]

Hello,

I now passed 33 kHrs on ADR#01/#02

No visible drifts of references except some measurement noise since over 1 year now.
The 10V/6V6 divider drift is also very stable for a non hermetically resistor.
Quintessence: it needed ~1000 days to stabilize the 2 samples.

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

[MiDi]

The 10V/6V6 divider drift is also very stable for a non hermetically resistor.

Thanks Andreas for sharing this valuable information.
What is the RN used for the boost to 10V?
Are there schematics and pictures of your ADR1000 PCBs published?