AD587LW - 10V precision travel standard

[Andreas]

Ups I did it again:

(putting one of my references into the freezer for ~8 hours).
see also:
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg2449794/#msg2449794

AD587LW#1 had drifted back to the old value before the first freezing.
So I wanted to know wether I could trim it back near inital calibration by freezing again
or wether the reference is hardened by first freezing so no new change would happen.

Temperature logger shows that -20 deg C where reached.

And: we have again a step in output voltage: this time by -6.1 ppm. (previously -4.8 ppm)

with best regards

Andreas
 

[iMo]

Why so steady drift upwards after so many years? It is a metal can, isn't it?
Wouldn't be better to get rid of all those settings resistors and thermistor around it and live with something like 10.034.256?? It is only a number..

[3roomlab]

so is it possible that there is a temperature where it does 0ppm drift?
maybe @ 5C ? 

[iMo]

It is a travel standard so messing with ovens/coolers is not the right way to go, imho. I asked above as the drift could be caused by surrounding resistors and trimmers. I think a naked 587 (with the buffer) and a diode or thermistor for measuring the temperature close to the chip with a calibration graph glued on the top of the box would be less prone to such long term steady drifts..

[Andreas]

Why so steady drift upwards after so many years? It is a metal can, isn't it?
Wouldn't be better to get rid of all those settings resistors and thermistor around it and live with something like 10.034.256?? It is only a number..

Hello,

it is a CERDIP8-package not a metal can.
Without those trimmers and termistors around we would have 1-2 ppm/K temperature drift.
So we would have annual variations of more than 20 ppm with my lab temperature range of about 18 - 33 deg C.

With the compensation I have only problems at very hot days (see those "dips" at above 30 deg C).

from my 4 samples 2 are drifting with about 2 ppm/year (which is not too much) the other 2 are much lower in drift compared to my LTZ-references.

with best regards

Andreas

[Andreas]

I think a naked 587 (with the buffer) and a diode or thermistor for measuring the temperature close to the chip with a calibration graph glued on the top of the box would be less prone to such long term steady drifts..

Hello,

I am dreaming of a small processor with 12 Bit A/D to measure temperature and outputting a correction voltage either per PWM or a 12 Bit DAC (LT1257). So with a 3rd order polynomal one could correct +/- 2ppm/K of selected AD587 over a 18-38 deg range with a resolution of  ~100 nV.

with best regards

Andreas

[iMo]

fyi - I've been using the vintage ADS1100 (6pin complete i2c 16bit adc) for converting temp from an LM35 sensor, compensating my 34401A

[DavidKo]

I think a naked 587 (with the buffer) and a diode or thermistor for measuring the temperature close to the chip with a calibration graph glued on the top of the box would be less prone to such long term steady drifts..

Hello,

I am dreaming of a small processor with 12 Bit A/D to measure temperature and outputting a correction voltage either per PWM or a 12 Bit DAC (LT1257). So with a 3rd order polynomal one could correct +/- 2ppm/K of selected AD587 over a 18-38 deg range with a resolution of  ~100 nV.

with best regards

Andreas

Maybe you can use similar approach as in AN86, only you need to add the code for the temperature measurement and correction.

[Andreas]

Maybe you can use similar approach as in AN86, only you need to add the code for the temperature measurement and correction.

The more interesting question is here: how can I monitor that the poor processor is actually doing that what is intended.
And not hanging e.g. into a reset-loop or forgetting the calibration constants while outputting the minimum/maximum possible voltage?

with best regards

Andreas

[guenthert]

The more interesting question is here: how can I monitor that the poor processor is actually doing that what is intended.
And not hanging e.g. into a reset-loop or forgetting the calibration constants while outputting the minimum/maximum possible voltage?

with best regards

Andreas

       The same as with any component.  If your VR is stable, you're going to assume all is peachy, if it jumps or drifts excessively you investigate.

     As for how to investigate software bugs, I don't think this application is much different from other MCU applications.   At first thought it seems even simple enough to be implemented as FSM, in which case the states could be enumerated and the index of current one output on otherwise unused pins (if that extra noise is deemed acceptable).

     And don't be afraid of reset-loops: the firmware of e.g. Keithley 181 always resets after each measurement cycle.

[miro123]

The more interesting question is here: how can I monitor that the poor processor is actually doing that what is intended.
And not hanging e.g. into a reset-loop or forgetting the calibration constants while outputting the minimum/maximum possible voltage?

with best regards

Andreas

i have few questions
1.Do you want only to monitor the temperature or do you want to do compensation? or just say it with other words does your application requires exact 10.00000V
2. How you gonna sense the temperature? I asking this question because i struggling to get it right. I've build TC oven I can easily achieve sub-milikelvin sensitivity 'mK' and reasonable long term stability. unfortunately the mesurement erors are several orders of magnitudes worse - e.g placing the probe few cm further away gives +-100..200mK, small smd component cannot be measured directly
I come to conclusion that sensing of temperature can only be done on silicon die - e.g. LM399, LTZ1000 or REF50xx 

[Andreas]

Hello,

with the AD587LW project I am compensating T.C. according to the suggestion from Lars Walenius in the analog domain.
This works good for a small temperature range and you can do linear and partly 2nd Order temperature compensation.

Of course a oven will give a much better stability but I also want to have a 4-7 days continuous operation on battery.

Since the temperature range of analog compensation is too small for my "lab" temperatures it would be good to have a 3rd order temperature compensation.
But if you do this you also have to be shure that the compensation is "bullet proof".
So I actually want to monitor that the temperature compensation is working right.
(on a oven you also have a LED indicating that the setpoint temperature is reached):

For temperature sensing a good thermal isolation of the NTC/Voltage reference area against environment is essential.
Most of the temperature is distributed by the component leads. (so at best connect NTC ground to Ground pin of AD587)
And also selecting devices which have a low T.C. (< 2 ppm/K) in the temperature range of interest helps to maintain stability.
SMD components have usually lesser stability than DIP/CerDIP packages.

with best regards

Andreas

[magic]

I asking this question because i struggling to get it right. I've build TC oven I can easily achieve sub-milikelvin sensitivity 'mK' and reasonable long term stability. unfortunately the mesurement erors are several orders of magnitudes worse - e.g placing the probe few cm further away gives +-100..200mK, small smd component cannot be measured directly
I come to conclusion that sensing of temperature can only be done on silicon die - e.g. LM399, LTZ1000 or REF50xx

You could try to measure chip temperature by monitoring its quiescent current, but I don't know how much TC it has and it could be too low to be practical.
Other than that there seems to be no hope for AD587. Somebody torn down a few units and even the "factory test pins" had no internal connection.

[Andreas]

Hello,

somewhat off topic here but has to do with AD587:
Branadic asked me wether the zener reference voltage (visible on Pin 8 ) is typically "better" in T.C. than the output voltage.
I found this question very interesting. Especially because there is the Joe Geller Patend using a LM399 connected to Pin 8 to improve the T.C. of a AD587 to below 1 ppm/K.
https://patents.google.com/patent/US7382179B2/en


So I made a measurement.
My K2000 (temperature compensated) is connected to Pin 8 (which is filtered by a 1uF MKS02 capacitor).
My ADC27 is connected to the 10V output of the reference.
My 34401A is connected to both voltages measuring the ratio between VSense = 10V and Input = 7V

Here the results of 2 samples
AD587JQ#04 has a relative high T.C. of ~7ppm/K within 10-40 deg C
AD587JQ#08 has a relative low  T.C. of <2 ppm/K within 10-40 deg C

AD587JQ#4
Vout  ADC27  225.3 ppm / 32.4 deg C = 6.96 ppm/K  9998.078 mV @ 25 deg C
Vz    K2000  229.8 ppm / 32.4 deg C = 7.09 ppm/K  7132.876 mV @ 25 deg C
Ratio 34401A  4.44 ppm / 32.4 deg C = 0.14 ppm/K
calc  ratio   4.81 ppm / 32.4 deg C = 0.15 ppm/K


AD587JQ#8
Vout  ADC27  54.8 ppm / 32.9 deg C = 1.67 ppm/K 9999.375 mV @ 25 deg C
Vz    K2000  63.2 ppm / 32.9 deg C = 1.92 ppm/K 7136.059 mV @ 25 deg C
Ratio 34401A  8.6 ppm / 32.9 deg C = 0.26 ppm/K
calc  ratio   9.0 ppm / 32.9 deg C = 0.27 ppm/K

here the initial drift was rather high so I had to wait 2 temperature cycles until the values stabilized somewhat.

In both cases the output voltage divider has a relative low T.C. (well below 1 ppm/K).

With best regards

Andreas

[branadic]

Thanks Andreas for performing this measurement and presenting the results. So that literally means, that the third order component in the t.c. is actually coming from the temperature compensation at the zener diode itself and is not added by the output stage, which makes it hard if not impossible to compensate for, sad news.

-branadic-

[Andreas]

Hello,

some update on ageing drift data of my 4 AD587JQ references now after ~ 4 years (>1400 days)

The drift data is relative to my LTZ references.
Most stable reference is LTZ#4 (green line in charts).

#01: this device I had 2 times over night in freezer (which gave ~-5ppm jumps) otherwise the drift would have been ~10 ppm in 4 years.
#02: ~4 ppm in 4 years against LTZ#4. so pretty stable.
#03: was 1 time in freezer (see jump) otherwise near zero ageing drift but some seasonal changes due to internal temperature changes.
#04: also nearly no ageing drift against LTZ#04 but rather sensitive to temperature when internal temperature rises above ~30 deg C

with best regards

Andreas

[mawyatt]

Hello,

somewhat off topic here but has to do with AD587:
Branadic asked me wether the zener reference voltage (visible on Pin 8 ) is typically "better" in T.C. than the output voltage.
I found this question very interesting. Especially because there is the Joe Geller Patend using a LM399 connected to Pin 8 to improve the T.C. of a AD587 to below 1 ppm/K.
https://patents.google.com/patent/US7382179B2/en
Andreas

The Joe Geller idea is very clever, and your measurements suggest that the AD527 doesn't contribute much additional temp drift with the amp and gain resistors, as the dominate TC source seems to be the buried zener.

Wonder why National did not have a single chip that basically combined the LM399 with on-chip heater and the AD587 amp resistor network to get to 10 volt output? Or maybe they did and I just didn't know about it?

Anyway, thanks for the post and now you've created an intriguing question of whether this LM399 and AD587 combo might be an attractive alternative to a LM399 with discrete amp and resistors??

Best,

[Andreas]

Hello,

with the AD587 output ratio I do not have the flexibility that I want.
(converting from 6.6 to 7.2 V to exactly 10V).
Even if the ratio stability is rather good.

So I´d rather use a high resolution DAC with specified ratio drift for that task.
(So similar to Ian Johnstons approach).

with best regards

Andreas

[Kleinstein]

The LM399 is running quite hot and this can effect the stability of resistors.
A stable 10 V reference is not such a common application that it needs a special chip for this. There is still the option to have a discrete zener, resistors and temperature regulation that includes the resistors.

NS had a LM369 zener reference with 10 V scaling, but without a heater. So somewhat similar to the AD587.

[mawyatt]

50 years ago or so National had a Hybrid 10V reference but don't think it had a heater either. We used these a lot, but recall they were somewhat expensive.

I'm sure there was a very good reason for not including the heater, otherwise they would have done it!!

Best,

[kleiner Rainer]

50 years ago or so National had a Hybrid 10V reference but don't think it had a heater either. We used these a lot, but recall they were somewhat expensive.

I'm sure there was a very good reason for not including the heater, otherwise they would have done it!!

Best,

National Semiconductor LH0070 - there are two versions of it: a hybrid module (there comes the "LH" for hybrid from) built around a LM329 reference, some laser trimmed resistors and a LM741 and a newer version built around a LM169, but still on a hybrid.
I got a sample of the older version during my internship, forty years ago - it is still working and of course well aged.

Link:

https://www.richis-lab.de/REF09.htm

I had a link with pictures of the first version, but it seems to have disappeared.


Greetings,

Rainer

[Andreas]

Hello,

obviously I missed to repeat 1/f noise measurement after population of the transient zeners in ~April 2019
So after doubt of IMO I repeated the noise measurement to see if there is any increased noise by the zeners.

I could not see any significant change in 0.1 - 10 Hz noise.
The values in the table are averaged values out of 15-19 repeated measurements with 100 sec measurement time each.

with best regards

Andreas