ADR1399 reference

[branadic]

I think it's worth having a separate thread for the ADR1399 to not mess up the threads about LM399.
About 2000 h ago I've started monitoring one of two ADR1399 with a Prema 5017. After playing with the integration time in the first days I kept the setup as is afterwards and recorded the output voltage. Unfortunately, the Prema has quite some t.c. and noise contributing to the measurement, so it's not the meter of choice to observe this kind of references, but all I had available at that time. Will change that setup soon and add the ambient conditions as well.

-branadic-

[jonpaul]

Brandiac, have connexions at ADI, hoping to get the 1399s and setting up tests here, have latest Keysight 34465A in CAL.

Any tips on your set-up? How does 34465a compare to your meter?

Bon Soirée


Jon

[macaba]

Attached is output voltage deviation vs. heater voltage. 0.37ppm/V.

[Kleinstein]

Brandiac, have connexions at ADI, hoping to get the 1399s and setting up tests here, have latest Keysight 34465A in CAL.

Any tips on your set-up? How does 34465a compare to your meter?

Bon Soirée

Jon

Directly measuring a reference with a DMM will give the combined drift and noise of the DMM and the DUT.  The Prema 5017 by now likely has a well aged LM399 reference and could thus be low dirft.  For the KS34465 it depends on how much that meter was used before. It may help if one has a calibration history.

The noise in both case would be to a large part the meters internal noise form the LM399.

[Andreas]

Attached is output voltage deviation vs. heater voltage. 0.37ppm/V.

Hello,

How is this measured? with or without thermal isolation of the ADR1399? (on both sides of the PCB)

Additional question:
Does the output voltage change also (like on LM399) when changeing the orientation. (laying on left / right side or legs above plastic cap ?)

See also for a LM399 based instrument here:
https://www.eevblog.com/forum/metrology/project-pimp-a-keithley-2000/msg1106829/#msg1106829

with best regards

Andreas

[bastl_r]

hello guys
I found on ebay very cheap resistors with only 2ppm for 1,5€ each.
With 2 sets of these and a Trimmer you can create a voltage divider for the ADR1399 to 10V.
2pcs 31k623 in series and 1pcs 26k1 plus a 200Ohm trimmer.
Here are the links to the resistors:
https://www.ebay.de/itm/144374605346
https://www.ebay.de/itm/144475379838



regards

[branadic]

Why bothering with expensive single resistors, if you can use a resistor network for less the price, that can be configured in a ratio required?

https://xdevs.com/guide/rnet_ratio/

They offer good t.c. matching and the network can be trimmed for the final output voltage by paralleling resistors to a single element within the network, that way the t.c. of the parallel resistor is dampened by the network.
Final t.c. trimming can be accomplished by adding copper in either the upper or lower branch of the network, depending on the sign of the remaining t.c. (see attachement).

-branadic-

[bastl_r]

Hi
Thank you for the tip.
But i don't understand how i can adjust a calculatet ratio of 1.41520412358094.

Can you explain what is to dofor this ratio?
One of my ADR has 7,06624V and the other 7,07143V...

Regards

[branadic]

Simply start with a 2:5 ratio = 0.4 and trim the "5" part of it close to 2:4.817 = 0.4152 with parallel resistors.

-branadic-

[dietert1]

The proposed design with 26100 and 26300 Ohms isn't wrong, except one should not use a trimmer. A trimmer isn't stable enough after some years. Also it is easy to lose calibration without notice.

Better calculate the value needed and get a good ( < 15 ppm/K) resistor of that value. The tolerance of the 26100 resistor will be about 2.6 Ohms. Better choose a series trim resistor on the high side, then one can do ppm calibration by some parallel resistor to that trim resistor, without touching the foil resistors anymore.

Regards, Dieter

[branadic]

New datasheet available https://www.analog.com/en/products/adr1399.html showing ADR1399 in LCC package. Good thing about that package is: "...The LCC version adds four pins, two of which are not internally connected (NIC) and the other two split the active reference into force and sense action on the top and bottom of the shunt..." but that comes at the cost of "...The heater power for the LCC version is about 3× to 4× greater than the TO-46 version...".

-branadic-

[alm]

A higher heating power means stronger thermal gradients, which I'd expect to cause higher temperature coefficients.

[Kleinstein]

The actual heat needed for the LCC package version will depend a lot on the layout / PCB. With some isolation and thin traces it can be quite viable to create quite some thermal impedance on the PCB level. With a suitable PCB and cap the heater power may be comparable.

The higher hyteresis would be a weak point for the LCC version - though this may also depend on the PCB / holder.

[miro123]

I'm looking forward on LCC version. Leaving kovar and adding sense pins are welcomed changes. I wonder what can be made by hobbyist in insulated "dead bug" configuration.
Professional world will most like appreciate LCC voltage reference @5/4.096/2.5V

[branadic]

Meanwhile, 3000 h have passed and measurement is still ongoing. As announced ambient sensors were added, hopefully giving a better clue what's going on. Maybe I should switch from Prema 5017SC to Prema 6048 to get better resolution and stability plots. But that's a task for the near future.

-branadic-

[branadic]

I was asked to perform a 1/f noise (0.1 - 10 Hz) measurement on the ADR1399, so I did exactly that, result attached. A nice match for post 2000.
By the way, this reference was used to verify, whether the Prema 5017 is really limited by its ADC in terms of noise as Andreas assumed or by the low frequency noise of the LM399 used as a reference in there and Andreas was totally right. There is absolutely no improvement when replacing the reference. A short article about it can be found here.

-branadic-

[WillTurner]

The ADR1399 datasheet revision 0 (not the new datasheet listing the LCC package, which I am yet to read) has the following statement listed under "Thermal Resistance" on page 4 of 11 which may be missing in the new revision :-

  "To reduce heater power, avoid bringing power planes close to the device except on the bottom of the PCB, and use a mesh ground instead of a solid plane. In addition, mounting the device at the full height of its leads, approximately 1cm above the PCB surface reduces the heater power."

Does that put an end to the long/short leads debate[5]?

The LCC package is an attractive option. I guess mechanical relief slots, and edge of board location[3, 4] come into play, or dead-bug style[1, 2], which looks promising :-).

References
[1] 

ADC20 is built with a hermetic LS8 package mounted dead bug style in a PCB cutout to avoid PCB stress.
:
:
The LS8 hermetically package shows +/-14 = 28uV or 8 ppm hysteresis.
:
:
With best regards

Andreas

[2]

I have a small pcb with a voltage vref LTC6655B. Unfortunately, the long-term drift specified only for LS8 is 20ppm/sqr(1000) ... I have MSOP package though.

:
:
it has a reason why the (typical) drift is only specified for the hermetically package.
:
The LS8 package is better, but only if you mount it dead bug style on the PCB so that there is no influence from the PCB.

With best regards

Andreas

[3] 

https://www.eevblog.com/forum/metrology/slot-holes-around-ltz1000(a)/msg1531931/#msg1531931

[4] That App Note that discusses mechanical relief which was quoted long in the past ...

[5]

I once investigated LM399 with short and long leads with and without slots.

-branadic-

Edits 1. : Add some references, dead-bug bit.

[Kleinstein]

For the LM399 there was never much debate over long or short leads. Long leads are preferred, just need a bit extra effort for soldering (e.g. a temporary spacer) and the extra hight.
The longer leads also give some extra isolation from mechanical stress.

The long / short leads discussion is with the LTZ reference with higher current (and thus more effect of the lead resistance) and a usually lower temperature setting and thus less heat flow through the leads.

The question is more how much thermal/stress relief is needed / sensible on the PCB. A point to keep in mind here is that it is not only about the stress to the reference, but also stress cause by the heated reference effecting other parts (e.g. resistors) on the PCB.  The heated reference is not so much a sensitive part, but in the thermal sense also a source of disturbance. While the reference itself stabilizes fast, it can take quite some time before the parts around the heated reference get a stable temperature (and even worse a stable humidity level).

[KT88]

"To reduce heater power, avoid bringing power planes close to the device except on the bottom of the PCB, and use a mesh ground instead of a solid plane. In addition, mounting the device at the full height of its leads, approximately 1cm above the PCB surface reduces the heater power."

This statement is about power consumption. The ADR1399 in LCC package has a junction to ambient thermal resistance of 125°C/W with a JEDEC board (I didn't find that layout though).
The layout recommendations are meant to further increase the thermal resitance to save power. The LTZ1000A has 400°C/W in comparison.
Other than the TO-46 package the LCC package doesn't use Kovar thus having a lower influence of the Seebeck effect. This means that thermal gradients would have a lower impact on errors than with a TO-46 case.
If highest performance is the goal I would still aim for lowest thermal gradients on the board and sacrifice a bit more power...

[branadic]

For the LM399 there was never much debate over long or short leads. Long leads are preferred, just need a bit extra effort for soldering (e.g. a temporary spacer) and the extra hight.
The longer leads also give some extra isolation from mechanical stress.

There was a discussion about short vs. long leads before, you must have missed that. Why was there a discussion? Well, you find short leaded LM399 in HPAK meters sitting in sockets, so Keysight must have missed your arguments for long leads. There are other examples were short leads were used, but there are also also examples for long leads, same goes for mounting the reference upside down or straight into the air.
I also made a few thermal imaging videos about short vs. long leads, slots vs. no slot that you can find here:

https://www.youtube.com/user/TheHendi78/videos

Everyone can draw his conclusion based on that results.

The heated reference is not so much a sensitive part, but in the thermal sense also a source of disturbance.

I wouldn't sign that either, the reference is quite sensitive to tilting, as the heater resistor is located only at one edge of the reference, so it's to be expected.

-branadic-

[Kleinstein]

I was asked to perform a 1/f noise (0.1 - 10 Hz) measurement on the ADR1399, so I did exactly that, result attached. A nice match for post 2000.
By the way, this reference was used to verify, whether the Prema 5017 is really limited by its ADC in terms of noise as Andreas assumed or by the low frequency noise of the LM399 used as a reference in there and Andreas was totally right. There is absolutely no improvement when replacing the reference. A short article about it can be found here.

-branadic-

I had a quick look at the 5017 circuit. One point is that this meter, like the HP3457 has a prime range of some 3 V. The 30 V range already goes through the input divider and correspondingly has 10 M input resistance. So the range to test would be the 3 V range. The 30 V range may be limited by the divider / amplifer. The ADC part is a bit confusing and may also have quiite some noise.

[MK]

For the LM399 there was never much debate over long or short leads. Long leads are preferred, just need a bit extra effort for soldering (e.g. a temporary spacer) and the extra hight.
The longer leads also give some extra isolation from mechanical stress.

There was a discussion about short vs. long leads before, you must have missed that. Why was there a discussion? Well, you find short leaded LM399 in HPAK meters sitting in sockets, so Keysight must have missed your arguments for long leads. There are other examples were short leads were used, but there are also also examples for long leads, same goes for mounting the reference upside down or straight into the air.
I also made a few thermal imaging videos about short vs. long leads, slots vs. no slot that you can find here:

https://www.youtube.com/user/TheHendi78/videos

Everyone can draw his conclusion based on that results.

The heated reference is not so much a sensitive part, but in the thermal sense also a source of disturbance.

I wouldn't sign that either, the reference is quite sensitive to tilting, as the heater resistor is located only at one edge of the reference, so it's to be expected.

-branadic-

Long leads means the heat loss is less, but it is easier to get the kovar/copper thermocouples at different temps, short leads I suspect may make it easier to keep the thermocouples under control, but give you other thermal problems.

[Andreas]

Hello,

i do not agree if I look at the pictures branadic did with some LM399 with short/long legs and slotted/unslotted PCB.

https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg278247/#msg278247

The Pins on the PCB are much more uniform on the unslotted PCB with long leads.

With best regards

Andreas

[MK]

Hi Andreas,

I do not think the camera has the resolution to prove it for either of us, 40uV per degree.
We need the pads to have less than 20 millidegree difference, the photos cannot provide that.

[Kleinstein]

Hi Andreas,

I do not think the camera has the resolution to prove it for either of us, 40uV per degree.
We need the pads to have less than 20 millidegree difference, the photos cannot provide that.

The requirements for the reference are not that strict: the LM399 reference is not that accurate to need better than 1 µV stability. The popcorn type jumps are usually in the 4 µV range and 1 ppm drift corresponds to 7 µV.  Another point is that a stable temperature difference only adds to the reference and is thus not a problem. Only the change in the temperature difference is that causes problems.

The camera is still no sufficient to directly resolve the temperature differences. However the camera can still help and show the temperature gradients a bit away from the pins. Due to thermal coupling the differences at the pins is expected to be quite a bit smaller than the differences some 5 mm away. The larger differences around the reference is that can be resolved and compare different configurations.
One has to balance between temperature gradients and voltage drop with long thin copper traces.