T.C. + Hysteresis measurements on brand new LT1027DCLS8-5 voltage reference

[Andreas]

Hello,

and the other verification charts from the remaining 2 ADCs (ADC22 dead bug + ADC23 slotted PCB) see also:
https://www.eevblog.com/forum/metrology/t-c-hysteresis-measurements-on-brand-new-lt1027dcls8-5-voltage-reference/msg1037076/#msg1037076

Both devices hooked up in parallel on LTZ#4.

on 19.09.2016 and 22.09.2016 T.C. measurement of the resulting T.C. after 3rd order correction
(and to show the resulting drift of the temperature cycles).

ADC22 (dead bug) shows large ageing drift on the first cycle
(Temperature correction on the 2nd cycle is still not optimal perhaps due to drift during calibration).

ADC23 (slotted) has large hysteresis on both cycles.

And also the results of the offset and gain drift.
The exact values are in the attached comparison table.

with best regards

Andreas

[Andreas]

First ageing results:

all devices have been calibrated to my ADC13 = 3597.846 mV after 2:1 divider on LTZ#4.
Calibration was between day 1-5.
Day 9-12  temperature cycling + T.C. verification of ADC21+ADC24
Day 14-18 temperature cycling + T.C. verification of ADC22+ADC23

From day 19 on all ADCs at room temperature running 24/7
Drift through 14 days (day 19-33) is between 2.7 and 5.4 ppm
Or 0.2 ... 0.4 ppm/day
or (extrapolated) 8 ...  16.3 ppm/khr

Quite a lot and up to now the drift looks more linear than square root.

With best regards

Andreas

[Andreas]

Hello,

now being at 1000 hrs from first calibration.

5 to 17 ppm/kHr.
too much for volt nuttery.

With best regards

Andreas

[branadic]

now being at 1000 hrs from first calibration.

5 to 17 ppm/kHr.
too much for volt nuttery.

However, this is well within the given typical value of 12ppm/kHr. So the typical 18ppm/3kHr will also match, I guess. Well, the LS8 package differs more than expected from the former hermetic TO package, what ever the reason might be, glue or package itself.

[Andreas]

Hello,

as already announced a hysteresis test with a much smaller temperature slope.
Again on ADC22 + ADC23 like on 22.09.2016 (see above).
Reduced to 0.04K/minute or 2.4 degC / hour.
So this is a more realistic value for my "lab" temperature change.

The hope was that this would reduce the hysteresis.
But all what I can see is a large drift during temperature cycling.
Around 3.1 ppm @25 deg C for ADC22 (against previous drift direction)
and 3.9 ppm @ 25 deg C for ADC23

With best regards

Andreas

[Andreas]

Hello,

this weekend I did a slow ramp of 0.04 K / deg on the remaining 2 ADCs
(ADC21 and ADC24).

ADC21 (dead bug) shows a excellent behaviour: less than 1 ppm change over 30 deg C temperature change.
ADC24 (slotted PCB) shows the well known hysteresis curve of around 4 ppm @ 25 deg C and 6 ppm change over 30 deg C temperature range.

So ADC21 (the only one of the 4 ADCs with selected LT1027DCLS8-5) seems to have nearly stabilized now.
Stability evaluation of the 36 hrs measurement shows a stability of 0.4uV (after 2:1 divider) corresponding to 0.8uV at 0..10V input.  (= 0.08 ppm! StdDev)
And this over the whole 30 deg C temperature range.

With best regards

Andreas

[Andreas]

2kHrs drift data:

first diagram LTZ#4 (with 2:1 divider) in mV in total view since start
2nd diagram calculated VREF-drift of LT1027 references zoomed in for 2nd kHr.
Still up to 8 ppm drift for the slotted PCB´s.

the jump on ADC22 by -3 ppm is after temperature cycling (see above).

Now I am doing 15 mA Loads on ADC#22+#23 over night.
I hope this will give accelerated ageing.

with best regards

Andreas

[Andreas]

3kHrs drift data:

as promised ADC22+ADC23 with 15 mA load over (nearly) every night for accelerated ageing.
ADC22 shows a direction change + large increase of drift to 8ppm/kHr.
ADC23 is nearly unimpressed by the treatment.

ADC21 shows drift below 1 ppm during 2-3kHrs.

with best regards

Andreas

[RandallMcRee]

What if you were to average the four references? Have you considered that?

Eyeballing the graph seems to suggest an overall very good result. Half the noise, too.

Randy

[Conrad Hoffman]

A couple thoughts on humidity, one unlikely, the other not so much-

1) I've heard that bee's wax has remarkably good moisture properties. Perhaps bake the board and dip coat it in molten bee's wax, maybe a couple coats. It worked for resistors and such in the old days.

2) My HP vector impedance analyzer runs the tuning cap in an oil bath to stabilize it against atmospheric changes, or so I believe. Bake the water out of some mineral oil or silicone oil and run the board in an oil bath. Certainly thermal benefits and maybe humidity protection.

[branadic]

ADC22 shows a direction change + large increase of drift to 8ppm/kHr.
ADC23 is nearly unimpressed by the treatment.

Well, ADC22 is dead bug mounted and ADC23 is the slotted pcb. Up to now I was believing dead bug is better, but this latest test now shows a different picture. Any ideas of interpretation of this results available?
Are you planing a similar test with ADC21 (dead bug) and ADC24 (slotted pcb) as a cross comparison?

[Kleinstein]

The dead bug type of mounting could result in higher thermal resistance and thus higher temperature from the 15 mA burn in procedure. Especially the jumps in the curve don't look that good.

The way one solders the chips also has an influence on the starting point. Soldering SMD chips is well enough to reach the 120 C range where epoxy starts to soften / changes structure.

[Andreas]

Hello,

What if you were to average the four references? Have you considered that?

no definitively not. I´ll keep only the best for precision measurements.

1) I've heard that bee's wax has remarkably good moisture properties. Perhaps bake the board and dip coat it in molten bee's wax, maybe a couple coats. It worked for resistors and such in the old days.

The LS8 package is insensitive to humidity.
And at least ADC21 + ADC22 (dead bug mounted) have no influence from PCB epoxy.

ADC22 shows a direction change + large increase of drift to 8ppm/kHr.
ADC23 is nearly unimpressed by the treatment.

Well, ADC22 is dead bug mounted and ADC23 is the slotted pcb. Up to now I was believing dead bug is better, but this latest test now shows a different picture. Any ideas of interpretation of this results available?
Are you planing a similar test with ADC21 (dead bug) and ADC24 (slotted pcb) as a cross comparison?

at the moment the humidity (in winter) is near constantly low at 50% rF here.
So what I see as drift has nothing to do with mounting method.
ADC22 has probably a wrongly mounted chip within package.
(simply behaving strange).
So for deciding about the mounting method we would have to wait until the annual drift is below 1-2ppm/year and then see the rF changes between summer / winter.

What I plan is to build some ADCs with (new) AD586 to compare.
But I can´t remember that any of my AD586 ever had drifts in the 8ppm/kHr range.

The dead bug type of mounting could result in higher thermal resistance and thus higher temperature from the 15 mA burn in procedure. Especially the jumps in the curve don't look that good.

I did not consider the higher thermal resistance up to now. But this should make less than a factor of 2 in effects.
As already said: Those jumps are most probably something strange with the (individual) chip itself or the mounting within LS8-package.
I am convinced that the LS8-package is not a good replacement for the CERDIP nor the metal can package.

with best regards

Andreas

[Andreas]

Hello,

Further drift data after having removed the ageing loading from ADC22 + ADC23

ADC22 (green) again changes direction
and also ADC23 shows (a small) influence.

So I will go back to AD586LQ for my purposes.
Even if they need a higher power supply voltage for proper operation.

For my part I will order some fresh AD586LQ from DigiKey and do a comparison.
I do not remember that they are such drifty as the LT1027.
But who knows if the newer parts also suffer from such drifts
(perhaps due to a different die attach due to RoHS)

With best regards

Andreas

[Andreas]

Hello,

just for comparison I ordered 10 "new" AD586 (datecode 1606) and measured to select for new ADCs.
This time I got a rather mediocre batch of AD586 regarding noise.
Usually I have the AD586 with 2-3 uVpp 1/f noise.
This time I have many above 3 uVpp (up to 9.2 uVpp) and many with popcorn noise (up to 5uVpp).
So there is only one really excellent reference (#42) within the 10.

My selection criterias are:
TC below 1 ppm/K best with a "sweet spot = zero TC" near room temperature
Hysteresis below 1 ppm over 30 deg C span.
noise below 3uVpp
no (<1uV) popcorn noise.

I selected the 2 best of the 10 AD586 to built 2 new ADCs which will be calibrated and then monitored for ageing.
AD586#42 is obviously the best.
AD586#38 was mainly selected because it has a "sweet spot" near room temperature.

Stressless mounting with only 1 pin soldered directly to the PCB was used to compare with the "dead bug mounting" of the LT1027DCLS8-5. The other connections are made by thin "VERO wire".

with best regards

Andreas

[RandallMcRee]

Just FYI these are "on-sale" at Newark/Element 14 (not sure if this is US only, or UK/Europe as well):

http://www.newark.com/w/c/semiconductors-ics/power-management-ics-pmic/voltage-references?reference-voltage=5v&voltage-reference-case-style=lcc&st=semiconductors+Cbargains

Seems like Andreas did not find these outstanding but in case someone wants to experiment this seems like a good deal.

Andreas: did you decide that the AD586 is better? What's your final opinion on the LT1027?

Randy

[Andreas]

Hello,

finally I still want to do some ageing measurements on ADC25+ADC26 with AD586LQ
You can shurely not generalize from few samples.
But what I measured is that the measured LT1027DCLS8-5 have more hysteresis and more ageing drift as that what I am used to from my AD586 ADCs.

But quality may change from batch to batch.
The latest batch of AD586 has a high number of  references with popcorn noise which I do not want to have in my precision cirquits.
If the manufacturer uses a different die attach (e.g. because of RoHS) this may have a influence on the hysteresis.

And it is always a question of requirements:
- I have large temperature and humidity changes in my lab (18-32 deg C and 40-70%)
- my "benchmark" ADC has a standard deviation of around 0.25 ppm measured over 42 days (1000 kHrs) and I still want to do it better.
- I always use battery supply. So I do temperature compensation with a NTC but no heating of the 5V references in my ADCs.

So under normal room temperature a selected AD586 against a selected LT1027 may be a factor 2 better. (But the price is also higher).
Essential is that you use a sort of "dead bug" mounting. Which is easier with a CERDIP8 package.

with best regards

Andreas

[Andreas]

Hello,

now after (finally) having adjusted ADC25+26 with AD586 I can do some comparative ageing measurements.
I selected ADC21 as "reference" because it was the most stable compared to LTZ#4 the last 1000 hours in the ageing measurement.

But there are several surprises:
ADC21 had stabilized to around reading 3597.810 mV for LTZ#4 with a 2:1 divider after 3kHrs. see;
https://www.eevblog.com/forum/metrology/t-c-hysteresis-measurements-on-brand-new-lt1027dcls8-5-voltage-reference/msg1112234/#msg1112234

And now after a longer off time reading has changed to 3597.842 mV (9 ppm drift).

When comparing a 24 hours measurement between the 3 ADCs then it gets obvious that also the temperature compensation of ADC21 has a problem. Nearly 3 ppm drift over a 4 deg C temperature excursion.

After adjustment of ADC21 I had around 1 ppm over a 30 deg C range. See:
https://www.eevblog.com/forum/metrology/t-c-hysteresis-measurements-on-brand-new-lt1027dcls8-5-voltage-reference/msg1061241/#msg1061241
I never observed such a large T.C. drift on any other of my ADCs.

Now ADC25+ADC26 with AD586LQ (with few operating hours for adjustment) are running against ADC21 (with more than 3kHrs operating time) for a new ageing test.

with best regards

Andreas

[Andreas]

Hello,

I now have found the reason for the T.C.-Drift on ADC21:
The T.C. correction was switched off on ADC21.
(probably I have done a ratiometric measurement and forgotten to re-activate the correction).

With active correction ADC21 looks much more stable (light blue) similar to the other ADCs.
But there is still that drift during off-time
now with active T.C. correction it is from  3597.810 mV mV to around  3597.836 mV (or 7.2 ppm)

with best regards

Andreas

[branadic]

Hi Andreas,

is it drift during turn off time or is it drift because of turning the reference off and on again? I wonder which mechanism makes the reference drift without any power supplied. A simple test could be to vary the power at both limits, from normal supply to minimum, maximum and back to normal supply and measure the output voltage.

-branadic-

[Andreas]

Hello branadic,

the ADC21 has been switched off since end of January.
But of course there were some temperature changes up to around 32 deg C in my "lab" which might also introduce some ageing effects.
(the temperature itself is compensated by a 3rd order correction).

Before it was contiously active for around 140 days. (ADC21 readings of LTZ#4 drifted down so the LT1027 drifted up during these days).
And now it seems that it has drifted back to around the value which was in the beginning of these 140 days.

with best regards

Andreas

[Andreas]

Hello,

next 24 hr stability measurement (after some stabilisation time) on my LTZ#4 reference.

ADC25 red (AD586LQ) has execellent stability and lowest noise
ADC26 green (AD586LQ) shows some popcorn noise
ADC21 blue (LT1027DCLS8-5) also shows some instabilities.

and so no surprise also the Allan Deviation reflects the results.
Note that all measurements are done after 2:1 divider. So to compare with other instruments all results have to be multiplied by 2.
ADC25 with averaging over some minutes shows a stability of 0.15 uV (2:1 or 5V range) or 0.3uV (1:1 or 10V range).

With best regards

Andreas

[Andreas]

Hello,

now ADC25+ADC26 with AD586LQ are running for their first 1 kHr.
So attached the ageing drift compared to ADC21 with LT1027DCLS8-5.
(absolute (mV readings) and relative VREF drift).

And also attached the equivalent sheet of ADC21-ADC24 (LT1027DCLS8-5).
Drift of the AD586LQ is at least a factor 3 lower than LT1027DCLS8-5.

with best regards

Andreas

[branadic]

So the picture has slightly changed? ADC26 performs better then ADC25, even though it has popcorn noise? Or has the popcorn noise vanished? In your last post ADC25 was more stable (see Allan Deviation).

Have you performed Allan Deviation on ADCs with slotted reference as well? How do they perform?

-branadic-

[Andreas]

Hello branadic,
 
thats exactly the reason why I have started another stability measurement (will be around 30 hrs).
So today in the evening I will know wether ADC25 or ADC26 is more stable now.

On the other side the standard deviation of my ADCs (when well pre-aged)
is around 0.25 ppm for daily measurements of 1 minute integration time.
So 0.5 ppm jumps (+/- 0.25 ppm) is nothing unusual.

If you mean ADC23 + ADC24 with LT1027DCLS8-5 and slotted PCB:
I have only some older stability measurements which I have done immediately after T.C. adjustment.
(so before the 3 kHr ageing phase with still high initial drift).
Here ADC16 (also a AD586) is given as reference for ADC24 -> minimum a factor 2 better in stability.

So a more recent comparison with ADC24 would be interesting.

with best regards

Andreas