DC Voltage References Battle - Overview of market

[plesa]

This table can be useful ( I do not remember who is author of this, but it was posted on eevblog some time ago).
https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/?action=dlattach;attach=40252

For me is favourite reference LM399AH, LTZ1000 (A) and LT1027.

[Kleinstein]

The extra OPs inside the AD588 are not usefull for temperature controll, as there is no temperature sensor on the chip.

The more valuable part are the on chip resistors to get specific values for the ref voltage. Even than one might decide to use external OPs to reduce power consumption.

There are a few ref chips like Ref02 that do have a temperature sensor on chip and could in principle be used to get them termalized. Up to the point of using a LM723 to controll its own temperature.

[Alex Nikitin]

the only test in this range (XFET) was (only 1 new sample) ADR4550 (violett = ADC18)
against two (aged) LT1027 in plastic DIP (ADC4 + ADC8)
and one (aged) AD586LQ (ADC13) in CERDIP
All measuring a LTZ1000A with a precision 2:1 divider (LTC1043).

I think it gets clear why I did not try a 2nd sample.
nearly 100 ppm drift over 300 days and not calming down to a sub 10ppm level.

Thank you, Andreas, however the ADR01 is a refined band-gap, pin compatible with the REF01. The "BRZ" version in SOIC8 suppose to have better than 3ppm/C (1ppm/C typ) tempco and less than 50ppm/1kh long-term drift. I'm looking at that particular reference as it has a temperature monitor output and can be compensated to almost zero tempco in a narrow range, I think.

The ADR3450ARJZ on the other hand is possibly the cheapest option for less than 10ppm/C tempco.

Cheers

Alex

[eas]

I assume it is one of these offers from China, I have looked to also:
http://www.ebay.de/itm/252001708799?_trksid=p2055119.m1438.l2649&ssPageName=STRK%3AMEBIDX%3AIT
http://www.ebay.de/itm/251943908859?_trksid=p2055119.m1438.l2649&ssPageName=STRK%3AMEBIDX%3AIT
I have daubt, this are really BQ version, as just the chip cost > 22 USD in higher volumes. Just assume this are "refreshed" JQ or AQ types.

Yes, it is one of those. Same board design, but my PCB is green. The 15v batteries it is intended to take are hard/expensive to get in the US.
It could be a relabeled version of one of the JQ or AQ chips, but I doubt it since the labeling is embossed into the ceramic. I suppose someone could have gone to the trouble, but that seems unlikely. The bare chips I purchased on the other hand...

My "lab" has been packed up while my inlaws visited, but they are leaving today. I'll unpack it and collect some voltage + ambient temp readings over a few days.

[Gyro]

The extra OPs inside the AD588 are not usefull for temperature controll, as there is no temperature sensor on the chip.

Sorry, I should have been clearer, what I was thinking was more in terms of having a temperature sensor (diode or thermistor) strapped to the ceramic body and using one of the opamps as a combined comparator an heater. The whole thing being well insulated of course.

Another Edit: I suppose the opamp could be used to drive an external heater transistor strapped to the other side of the package too.  Maybe the whole idea is a non starter as it would also need scaling resistors to compare the temperature sensor to the main reference (with their own TC and drift).

[Andreas]

Thank you, Andreas, however the ADR01 is a refined band-gap, pin compatible with the REF01.
 I'm looking at that particular reference as it has a temperature monitor output and can be compensated to almost zero tempco in a narrow range, I think.

Hello,

sorry for mixing up. I only know ADR as XFET (ADR445, ADR435, ADR425, ADR4550 ...)
But Band-Gap references also suffer from larger ageing than buried zeners.
(at least for the reference element itself).

If you want to compensate the T.C. to get further improvement then 2 other effects will annoy you:

1. humidity: may be up to 0.5ppm/% relative humidity for plastic packages
    (you can see this in the curve for the LT1027CCN8-5 of ADC4 + ADC8)

2. Thermal hysteresis. This depends on IC package but may be
    in the 10-30ppm range especially for non stiff (small) packages
    for a small temperature range (e.g. 10-40 deg C).
    Thermal hysteresis is also dependent on temperature change speed (gradient).

I have already thought of using the trimming pin of a LT1236AILS8-5 as temperature output to compensate its T.C.
But this needs buffering and probably a 2nd high resolution ADC to get a temperature resolution high enough to
get good results without additional temperature steps in the output.

Another possibility for temperature compensation is described on the page of geller labs.
http://www.gellerlabs.com/Voltage_References/SVRTH/SVR%20THR%20Sch.jpg
This can be used with every reference with a trim pin for a small temperature range.

with best regards

Andreas

[Andreas]

Reviews

AD584KH quad-voltage reference - TiN's review - in works

Hello TiN,

it would be good if you check the voltages of the AD584 after 1 year again in your review.
From a german forum I have the figure of 1.76 mV Drift for the 10V within 1 year against a PM2534 Multimeter.
So for me the AD584 references are perhaps a cheap way to import the "volt" in you home lab.
But you should not rely on that the voltages are stable over a longer time.

with best regards

Andreas

[Alex Nikitin]

Another possibility for temperature compensation is described on the page of geller labs.
http://www.gellerlabs.com/Voltage_References/SVRTH/SVR%20THR%20Sch.jpg
This can be used with every reference with a trim pin for a small temperature range.

with best regards

Andreas

Hi Andreas,

It is quite easy to make independent adjustments for both voltage trim and tempco, to compensate the slope at a particular temperature. You don't need a DA-AD conversion for that.

Also, I have an old idea (possibly 20 years old) about a voltage reference, however up until now I've never tried it in practice (silly, I know  ). It is a very simple circuit and easily allows me to adjust it to less than 10ppm/C tempco at room temperature. With all this talk about stable references I've wired it up today and will see how stable it is till tomorrow (my lab gets about 5C cooler overnight). It is just a good fun to make a reference out of standard parts, not designed to be a reference! Right now it is on the bench, in open air, and the voltage stays inside a +/-2ppm range for an hour so far  .

Cheers

Alex

[TiN]

Woa, lots of activity here, need to catch up

Andreas
I'm not sure that would be possible, as I'll send it to my friend in Ukraine, and shipping it back and forth would cost ten times more than actual reference .
And it's not supposed to be voltage standard by any mean, only like a transfer standard.

[pelule]

Quote
The extra OPs inside the AD588 are not usefull for temperature controll, as there is no temperature sensor on the chip.

For a temp-control you need a "heater" and a "sensor".
The "sensor" should have an as close as possible thermal connection to the chip.
The "heater" isn't that critical, but a close thermal connection would support a good regulation and a minimum thermal mass.
At the u723 the power transistor was the "heater", the -2mV/°K base-emitter voltage drift of the current control tranisitor was "semsor".
At the AD588 it might be possible to use the offset drift (~1µV/°K) of one of the OpAmps (A3, A4) as a "sensor" by amplifying it with for example a gain of 100.000 (~100mV/°K) and use an opamp as the "heater" by driving a load (up to 10 mA @ 15V = ~150mW).

[Alex Nikitin]

I have an old idea (possibly 20 years old) about a voltage reference, however up until now I've never tried it in practice (silly, I know  ). It is a very simple circuit and easily allows me to adjust it to less than 10ppm/C tempco at room temperature. With all this talk about stable references I've wired it up today and will see how stable it is till tomorrow (my lab gets about 5C cooler overnight). It is just a good fun to make a reference out of standard parts, not designed to be a reference! Right now it is on the bench, in open air, and the voltage stays inside a +/-2ppm range for an hour so far  .

I've opened a separate thread on that reference.

Cheers

Alex

[plesa]

Hello TiN,

it would be good if you check the voltages of the AD584 after 1 year again in your review.
From a german forum I have the figure of 1.76 mV Drift for the 10V within 1 year against a PM2534 Multimeter.
So for me the AD584 references are perhaps a cheap way to import the "volt" in you home lab.
But you should not rely on that the voltages are stable over a longer time.

with best regards

Andreas

I can check on two AD584JH in my drawer ( worse spec than KH ) If you want.

[Andreas]

I can check on two AD584JH in my drawer ( worse spec than KH ) If you want.

Hello plesa,

on the one side it would be interesting (hermetical package) if all AD584 are that bad (> 1.7 mV in 1 year) or if it was an outlier.

on the other side: the devices have much too much T.C. and I measured >10 ppm hysteresis over a 10-40 deg C range.
Since I am looking for 5V references with
- humidity stable behaviour (hermetical package)
- T.C. around 25 deg C with less than 1 ppm/K
- hysteresis of less than 1 ppm over a 10-40 deg C band.
I decided to put my single sample of AD584JH back in the drawer.

And since the AD584KH is obsolete by the manufacturer since years the probability of
rebranding/remarking of all those KH devices is very high.

I also made the experience that the higer graded devices are not always better at room temperature.
Usually the adjustment of T.C. is made at the temperature limits.
The temperature gradient near 25 degrees may suffer from this procedure.

With best regards

Andreas

[lars]

In beginning of June I bought one AD584KH board with measurement data like the one TiN tested. It arrived after two weeks and the label said 10.00262V at 21C measured with an Agilent 34401 in 2014 (I have not cared about the 2.5, 5 or 7.5V yet). I had it powered a couple of days. As it was summer and I have no AC in my lab the temperature went between 23 and 27C. The temperature coefficient was -3ppm/C so extrapolated my reading at 21C was 10.00266V (with 5ppm uncertainty) that is +4ppm  from original reading. Now I have run it for a couple of days again. Now with about 19-23C. At 21C the reading is 10.00268 that is +6ppm from original reading, not bad for this low price. The AD584KH fluctuates a couple of ppm over the days (and years for another old AD584LH I have!).

I was really glad to see that the AD584KH board seems to have low sensitivity to air around. My first experience with an AD581KH  (very similar to the AD584KH except it is only +10V) more than 20 years ago was that I had to cover it from air movements. That time I had it with long leads on a breadboard and very soon mounted it in a box. Probably the mounting of the AD584KH with almost no leads to the board is very clever in this case.

Are the AD584KH really obsolete? I see that you can buy it at the Analog Devices web site and Mouser also have stock.

Lars

[Andreas]

Are the AD584KH really obsolete?

Hello Lars,

sorry I mixed that up with the "L" grade which was removed between Revision B and C of the data sheet.

There are still many boards with LH types on the web:
http://www.ebay.com/itm/AD584LH-4Ch-2-5v-7-5v-5v-10v-High-Precision-Voltage-Reference-Module-15-/261320055120?hash=item3cd7e39950:g:-00AAOxy2YtRvgw9

With best regards

Andreas

[lars]

Now I have bought another reference like the TiN one. I also bought two other with the square battery on from two different sellers.
Received the “TiN” style just after one week on Oct 20. This time the label said Agilent 3458A 2015 26C and the 10V was specified as 10.00801. I measured 10.000795. Have measured it again  today at 10.00797 so just about 5ppm below. The temperature coefficient was about -3pppm/C again but the datecode on the AD584KH said 9213 (last time 0635). So both of my 584KH board has been within 10ppm not bad. My guess for an uncertainty specification would still not be better than 50ppm (K=2). My assumptions are 30ppm from the Agilents, 30ppm due to temperature coefficients up to 15ppm/C and the customer is able to be within 2°C from stated temperature, and finally maybe 20ppm from drifts since measurement. This summed, as in GUM, gives about 50ppm. My thought is also that sellers should be conservative as these products goes to hopefully young hobbyist that can´t afford to buy a lot of references to build up statistical confidence. Also giving correct measurement uncertainties could help young (and old) to learn the importance of measurement uncertainties I hope.

I have also received one of the other two of the different style but no measured values were supplied. It had an AD584LH with datecode 1015. I measured 10.00231 at 23C both at arrival and now. The temperature coefficient is about +3ppm/C.
 
I enclose a chart from Excel that might be interesting for some. It shows a few selected references of my (about) bimonthly checks. As reference I use an AD581KH with datecode from 1984 that is only powered a couple of hours every second month.  It is actually 10.00367 and has a temperature  coefficient of about +1.5ppm/C and no temperature compensation but as most of my DIY references contains a 10k NTC (same as Digikey BC2301-ND). As a side note the drift of the NTC is below my measurement uncertainty for temperature and that over more than 20 years!
The AD584 is an AD584LH with date code 7945 and made in Mexico stamped on. It has a temperature coefficient of about +1.5ppm/C. It is only powered up every second month.
The three AD584-2 to -4 are AD584JH with datecodes from 1989 and have temperature coefficients from 0 to -6ppm/C. They were only powered up sporadically 1998-2001 but since 2001 they have been mounted together with some other references in a box and continuously powered and the internal temperature was about 36°C inside in average (ouside 23C). 2011 I completely rebuilt that box and took away the AD584´s.
The AD688AQ is a ceramic package mounted with trimmers I salvaged and cut out from an DAC board more than twenty years ago. It is only powered every second month.
Last the four REF102CM have datecodes  from 1998 I think. Around 2000 I mounted two each in a box with NTC temperature compensation. The REF102CM´s have been continuously powered since around 2000-2001. As can be seen they drifted the first years but probably not so much now.

Lars

[Andreas]

This time the label said Agilent 3458A 2015 26C and the 10V was specified as 10.00801.

Hmm,

why are there only 6.5 digits when measured by a 8.5 digit device?

With best regards

Andreas

[lars]

I should also select a resolution of 10uV as the AD584 bandgap device is specified for 30uV p-p 0.1-10Hz if I remember correct

Lars

[Andreas]

Hello Lars,

with 100 NPLC (2 sec) integration time which I would choose for "precision" measurements
you will have only a part of the 0.1-10 Hz noise.

With best regards

Andreas

[lars]

You are correct but also remember you are in the flicker region. So just over minutes you can have at least a 1ppm shift

Lars

[splin]

As a side note the drift of the NTC is below my measurement uncertainty for temperature and that over more than 20 years!
Lars

Lars,

Can you tell us anything about how you measure the NTC drift and how you calibrate it? What is your measurement uncertainty for temperature?

Thanks,

     Tony H

[lars]

I have been fortunate to have access to calibrated temperature probes and calibrated my reference probes to them so I have about 0.2C uncertainty as best. As I have done measurements on my voltage and resistance references about every second month for around 20 years I have quite good statistics. If I do a regression analysis on the NTC´s compared to the outside probes I get about +-0.02C/year. The NTC´s are 10k Philips/BC Components same as the the Digikey BC2301-ND.

Lars