Best out-of-the-box 10V reference?

[nuno]

Yes, but you can check that by the difference in measurements from unit to unit, since you know the tolerance, you can have an idea if the units you have are falling "too much on the same side" or not. But that's a good point, when factoring in the price. LM4231AMF doesn't look very expensive (maybe 4 for $10) with 0.05%, although 10ppm/ºC and some other worse long term specs.

[Kleinstein]

Especially when buying 4 of the same units from one source. Chances are you get some that ran of the same machine one after the other.  So chances are good the 4 units will be within 0.01% to each other - but you still don't know if they are all at the upper or lower end of the span or maybe dead on. Even if one unit is off from the others - you still don't know for sure this is not a bad unit outside the specs. Knowing to have an unusual difference also increases the probability to have this due to an defective part.

So overall the 4 identical unit's will not give a much better accuracy spec than a single unit. To make this work it would be more like using 4 different types or manufactures.

[nuno]

To make this work it would be more like using 4 different types or manufactures.

That would make a good experiment.

I guess that if the probability of getting chips more "far apart" in terms of voltage output were the same for any tolerance, then we could buy the ones with worse tolerance (cheaper) and still get an "average" as good as buying the ones with better tolerance.

[The Soulman]

Voltage references or resistors, the same principle applies:

[Andreas]

Voltage references or resistors, the same principle applies:

Hello,

I have made the first T.C. measurements of some AD587
1 UQ
2 JQ with datecode 1613
3 JQ with datecode 1342

@Lars: I fear you had big luck with your T.C. being below 5 ppm/K.

@Soulman: now try to make your statistics.
I still do not see any gaussian distribution of the output voltage.
Even when the price of the references easily exceeds that of 1000 metal film resistors.

my personal opinion:
even when building a reference with temperature compensation the  T.C. should not be too large.
Usually the temperature sensor is not on the chip. So there is always a temperature difference which can be 1-2 deg C.
So for a excellent device the T.C. should be around / below 1 ppm/K without compensation.

with best regards

Andreas

[Gyro]

I think you'll find that Joe Geller and Lars were using AD587LQ... Those are what Joe fitted to the SVR/SVR-T.

I'm not sure what you would find in comparison, but the LQ, sadly no longer available, was closer initial tolerance and better TC than the JQ. The LQ was specified max 5ppm/'C versus max 20ppm/'C for the JQ.

[Andreas]

The AD587JQ from Digikey (or Mouser) is right now my preferred DIY10V ref as it isn´t humidity sensitive. The last years I have bought AD587JQ´s a couple of times and no one have been over 5ppm/C (probably been lucky but am not sure).

Hello,

I referred to this statement of Lars.

with best regards

Andreas

[Gyro]

Ah, sorry. I missed that quote, lucky it is then!

Chris

[not1xor1]

@Soulman: now try to make your statistics.
I still do not see any gaussian distribution of the output voltage.

but if you average the values you get quite close to 10000mV 
unfortunately the Tc is positive for all the samples

[Andreas]

Hello,

for the diagram I picked the "box" TC (per definition positive).
the temperature gradient is negative for the lower TC values.

with best regards

Andreas

[tszaboo]

Voltage references or resistors, the same principle applies:

I had the privilege of measuring a lot (100+) of LTZ1000 references. It wasnt a gaussian distribution.

[The Soulman]

Voltage references or resistors, the same principle applies:

I had the privilege of measuring a lot (100+) of LTZ1000 references. It wasnt a gaussian distribution.

Than I stand corrected, but than the question: why not?

[Kleinstein]

For the LTZ1000 refs I would expects a more normal distribution, but there still is a chance they come from the same batch and can thus be all similar with maybe a slow movement of the average. So distribution could be a little more even, like square.

For precision 10 V refs, there usually is some checking of quality / grading after the adjustment. So the better units can end up at the higher grade - again having more units close to the limits. The lower grade samples may be missing a significant number of the good ones up to the point of having two peaks in the distribution. Also the specs are often guarantied by testing and rejection the bad ones, nut just by having a low scattering assuming only 0.1% will be off. It may happen for low cost chips / less critical parameters (e.g. TC), that they only test a few samples and rely on not too much scattering for the rest.

[Andreas]

For the LTZ1000 refs I would expects a more normal distribution,

Why that?

the LTZ1000 is a heated device with about 50 ppm/K in the unheated state.
so every little error in output voltage is multiplied by setting a different temperature setpoint.

For the 10V references there are different mechanisms which play a role.
First the zener and then the output voltage divider. (so at least 2 overlapping distributions).

Low T.C. components are often trimmed for tempco by a temperature compensation network which is adjusted during production at 3 temperature points.
Also there I would never expect a gaussian distribution.

with best regards

Andreas

[HighPrecision]

Hi,
got three NOS Burr-Brown REF10KM refs, datasheet is available on web, I haven't found any info here, anybody knows these chips ?

Thanks

[Vgkid]

Here you go.
[urlhttp://www.datasheetcatalog.com/datasheets_pdf/R/E/F/1/REF10.shtml][/url]
That looks like that uses the same datasheet as the ref101, I have.

[tszaboo]

For the LTZ1000 refs I would expects a more normal distribution,

Why that?

the LTZ1000 is a heated device with about 50 ppm/K in the unheated state.
so every little error in output voltage is multiplied by setting a different temperature setpoint.

For the 10V references there are different mechanisms which play a role.
First the zener and then the output voltage divider. (so at least 2 overlapping distributions).

Low T.C. components are often trimmed for tempco by a temperature compensation network which is adjusted during production at 3 temperature points.
Also there I would never expect a gaussian distribution.

with best regards

Andreas

Exactly. The reference itself has a large ~3% variation to begin with. I've used 0.1% resistors to set up the temperature setpoint (very nice thin film resistors from Susumu) , so that has much less tolerance. The Zeners, which had initially larger voltage also run somewhat warmer. But that is not what I saw.
I saw, that I had something, which looked like a normal distribution, where the middle part was missing. Because they bin the parts, and the better ones went to someone, who was paying more for the same part.
If you buy an accurate 10V reference, it will be either:
- Binned, so they will select the best parts and ship it to someone else
- Laser  or electrical fuse, or other electrically trimmed of the internal resistors. Clue: There are NC or DNC pins on the pinout. Trimming uses god know what algorithm to trim it.
- You get surprisingly lucky. Once I asked for 5 sample from a manufacturer, surprisingly, all 5 sample was extremely well behaving much better than datasheet. And then the production parts were much worse. Too bad, I started being skeptic to samples and even datasheets.
So expecting the same part to behave "randomly for tempco" or get normal distribution, or expect it to age in a certain way... Its just wishful thinking.

[HighPrecision]

That looks like that uses the same datasheet as the ref101, I have.

Thanks Vgkid, yes, REF101 is similar but more flexible to old REF10, with some difference in stability @1000hrs. You have the "KM" version ?

[lars]

What I have preferred the last years are actually the AD587 JQ as the LQ isn´t available. That the KQ and LQ isn´t available and that the UQ seems to be trimmed in another way is one reason I think it reasonable to believe a lot of JQ’s will be within 5ppm/C. And of course, I was lucky to get all my AD587JQ within +-5ppm/C. I have nine AD587JQ’s of date codes 1028, 1247, 1517 and 1609 bought from Digikey and RS components.

The datasheet for the AD587JQ says 20ppm/C temperature coefficient and that with the box method. So, at 25C you can have even worse TC. As I, Andreas, Joe Geller and a report from the LISA-project shows the AD587UQ has more like 7-11ppm/C at 25C but the box specification is 5ppm/C.

Joe used the LQ and I think all had the date code 0045. Most of the LQ Joe had were around +2ppm/C at 25C so the simple add-on with a NTC+ resistor worked well.

 I have temperature compensated many 10V references at home (never in a design at work) with NTC’s but mostly used other designs than the SVR-T design. The SVR-T design was just because the trimpot value was already fixed. Enclose a sketch for what I tried to describe earlier as an example for a more general design. It will give an adjustment range to compensate -8ppm/C to +12ppm/C. In this design the sensitivity to drift of all trim components are quite low compared to the SVR-T design. If I know the AD587 have a slight positive TC I can calculate a larger value instead of 1.5M to get less range and some second order compensation. Most AD587 seems to have a slight negative second order component of about -0.03ppm / (°C * °C) around 25C so a larger resistor gives some compensation for this. 1.5M together with 1.36M NTC gives a quite linear compensation, so no second order compensation, but useful for a broad range. I myself has set up an excel sheet for calculation of the NTC compensation.

That I say I prefer the AD587JQ is only true from a hobbyist point and to get a good “transfer standard” to compare and maintain 10V. In my professional work as a design engineer I haven’t designed in an AD587JQ and probably will not. Long long ago I designed in AD688AQ and BQ as they were good choices for +-10V refs with reasonable accuracy in data acquisition systems. As the AD588BQ are similar I guess it is a good choice for 15EUR/USD from eBay if they are not fake. My experiences with AD688xQ compared to AD587xQ are that the 587 is slightly more stable over time and draw much less power (10x) and don´t need dual supplies, so if you don’t need +-10v I prefer the AD587 in all aspects. Having just 2mA current draw is a plus when using say two 9V batteries to power your reference. Today most volt refs I design in are 2.5-5V in SMD of course.

Why 10V even for a hobbyist? I should say because of the traceability chain and uncertainties. If you, like me, after some years comes below 10ppm uncertainty the extra step from say 5 or 7.xV gives extra uncertainty. Not all have a super linear 8 ½ digit DMM or a KVD. My best DMMs are only 6 ½ digits and I never paid more than 100USD for a DMM yet. If you go for a couple of 10V references adjusted close to 10V it is also easy to make a simple x100 amplifier and with just a 3 ½ digit DMM to get 0.1ppm resolution when you regularly compare your references. Of course, after a while you will find it is difficult to measure the absolute long-term drift without repeated calibrations with known uncertainties. But even before that you have learned a lot about drifts between your samples. And temperature coefficients can be checked and adjusted with simple means. My first NTC temperature compensated REF102CM’s I adjusted with just a lamp above the DUT and a 3 ½ digit DMM + amplifier. Three out of four are below 0.1ppm/C at 23C. The fourth I must have made some mistake as it is +0.4ppm/C but as I have a second 10K NTC connected to banana jacks on the front of the box it is easy to compensate afterwards. Some other of my references aren’t compensated but I only compensate afterwards. I prefer the 10V adjusted to within say +-100ppm but some of my better reference have up to 400ppm offset (and 2ppm/C).

As Andreas pointed out you cannot temperature compensate away all TC on the IC. My estimate is maybe 10-20 times for a reference with about 5ppm/C from my experiences with an NTC very close to the IC (and of course mounted in a box, so outside gradients will not affect so much).

As the thread started with suggestion for 10v for MML I would say: DIY with AD587JQ and/or VREF10-003 from voltagestandard.com if you can afford 53USD for it. With the VREF10-003 you get 30ppm for 6 months and as far as I know it is a serious offer compared to many other Vref boards or boxes on eBay. I have only one VREF10-003 but at arrival it was very close and now after more than 1 year it is still just +12ppm. Of course it is humidity sensitive as all plastic refs. It has a cut-out on three sides of the IC but I am not so sure it helps. I think the humidity sensitivity comes from the epoxy in the DIP8 package. Enclose my measurements of it. The result is compensated for temperature (1.3ppm/C).

Having at least three 10V refs is nice as you can compare them. I try to compare my 10V refs each second or third month as that give multiple points over a year, which is nice to see seasonal variations.

Lars

A note: I usually say just “uncertainty” but more mean “expanded uncertainty” as it often is written in calibration protocols with k=2 or a confidence of 95%.

[lars]

Just some other notes from my experiments with 10V refs:

I have tested several LT1031 in metal case (all bought new from ELFA-Distrilec) and was not happy with them. They drifted a lot and especially nasty is that they change a lot if turned off for some weeks after being on continuously for a long time see chart in another thread. https://www.eevblog.com/forum/metrology/new-voltage-reference-from-maxim/msg971413/#msg971413 . The strange thing is that the two LT1021 I have in metal cases (bought on eBay) are much better even if not as good as the AD587xQ for drift and long power interrupts. In my measurements with 3 weeks power off and 3h on  a couple of times the AD587xQ an LT1021CMH is very little affected with less than 1 or 2 ppm , REF102CM about 3-5ppm and LT1031DMH up to 10-40 ppm. Many plastic packages also show 10-30ppm up to maybe 50ppm worst case as I have seen.

Almost all up to 1mA buried zener refs I checked has at some time jumps (pop corn noise) On some it can only be seen at some temperatures on other can be long between that is minutes, days or even months I have seen. (LM399, AD587, REF102, LT1021,1031, 1236 etc). For me it seems that about 0.8ppm jumps is very common but also 0.4 or 1.5ppm. One 6 digit DMM (with LM399) has for 20 years shown 1.5ppm jumps maybe several times per day. As others have shown even LTZ1000 might have “jumps”. So for an uncertainty analysis for 10v refs I always recommend to say 0.8ppm for jumps and also I have seen a step after months or even years of several ppm so getting below 5ppm uncertainty seems not serious for 10v IC refs if you don´t have several of them.

Line regulation is not as bad as the datasheets says in my experience. Last I tested six AD587xQ, two REF102CM and two LT1021CMH with a change of input voltage 15-20-15v the AD587 were best with about 1-3ppm change, the REF102 about 3-5ppm and LT1021 was worst with about 8-10ppm (Note LT1021 had about 5-7ppm/C before compensation the others below 3ppm/C). So datasheets doesn’t say everything. The spec for REF102C is 1ppm/V, LT1021 4ppm/V and  AD587 10ppm/V.

Reverse voltage protection is necessary from bitter experience. I mostly use a schottky diode 40v 1A like the 1N5819 which gives about 0.2v voltage drop at 2mA. Sometimes I have added an LP2951 set for 14v to get better line regulation. The LP2951 is cheap (<1USD) and draws little current and have worked well for me.

Having short-circuit protection with an extra buffer like Alex has with an LT1097 for the LT1021 is probably good but I have never used such protection. One thing to remember with buffers is to test with capacitive loads. Not all can handle that. I have once short-circuited a MAX6350 for a couple of minutes and that gave a permanent shift of -30ppm but that is the only time and I have measured references without protection lot of times. Some time I shall test an AD587JQ short-circuited.. some time…

I have never used protection for over voltage on input or output for my own designs and don´t remember I have had problems either. At work I always design in this but am not sure if it really works. Actually surge or burst tests is seldom needed on outputs like these (if you don´t have extremely long cables).

Having EMI protection is very good as you never know if you have any RF-fields nearby. But it is the same problem as with over-voltage protection. It is not easy to design and verify. Having appropriate bypass capacitors and ferrite beads have worked for me. Have done some EMC tests at home just with an HP8657B signal generator sweeping with a small loop close to the output binding posts from 1Mhz to 2GHz reveals that most of my 10v refs still have susceptibility to relatively small RF fields. My SVR boards and also the 2ppm eBay are worse than the once I have better EMI protection on. I have not tested my VREF10-003. If I remember correct the Fluke 732B has a very low spec of 0.18V/m (needs controlled lab environment!). Normal equipment for home use is tested at 3V/m and the equipment I work with is tested at 10V/m or more. You can really have large errors if the RF signal gets rectified. The really nasty problem with EMI is that you can calibrate your 10V ref in one environment and at home you have different RF fields and may be off many ppm without knowing.

Having the reference in a case is absolutely recommended, mostly due to air drafts. I can´t say that having a metal or plastic case seems to make a difference for me. Neither from an EMC or stability point. One reason the metal case is of little use for EMC is that the noise entering through the binding posts are the same for metal and plastic. Good design of the PCB probably is the best.

I have used different banana jacks and bindings posts including silver and gold plated brass or copper. Except for nickel plated once, I have not seen a difference for my 10v refs. Nickel-plated was a disaster giving easily up to several ppm errors.

Almost all of my 10V refs and resistor references have a temperature sensor. As my reference resistors have no power a PT100 or NTC is my choice for them and to have the same I most often use a 10kohm NTC even on the 10v refs even if I use LM35 on some and it is easier to convert to °C compared to NTC´s.

Lars

[Andreas]

Hello,

I have finished the T.C. measurements on the AD587 which I had in the drawer.
From the remaining references with DateCode 1342 2 were below 3 ppm/K.

If I calculate that from the T.C. the drift over a 10 deg C range can be compensated by a factor of 10
giving 10 uV error for each ppm/K * 10 K / 10
and take the full uncertainity for the popcorn noise
I get the "reference score".

So the yield is 2 excellent devices and 3 fair out of 9 samples.

With best regards

Andreas

[Andreas]

Hello,

just want to suggest a alternative T.C. trimming scheme.
I think this fits more to my relative large temperature range in my "lab".

R2 and RV1 give a linearized temperature dependant voltage of about 1% (100mV) per deg C in a 25 +/- 15 deg C range.
The voltage at TEMP goes down with rising temperature.
If your reference needs it the other way round you have to exchange RV1 and R2.

R1 R5 R4 is the voltage output trimming .
R3 has to be adjusted to the T.C.

But Im also thinking about using a 12 Bit DAC for the fine trimming and the TC correction. This has the advantage that the NTC can be always placed directly at the ground pin which should be connected to the lead frame of the chip.

With best regards

Andreas

[Gyro]

Just a thought, aren't you going to suffer more self-heating of RV1 (~1mW) due to the current through R2? I suppose this could  be good (maybe  ) or bad depending how it mirrors the dissipation of the AD587.

If you delete R2 and connect RV1 to the 10V ref output then you have the SVR-T circuit. Presumably you could still conect RV1 to ground instead to provide opposite TC compensation as you indicated.

Edit: I'm just wondering if R2 actually helps or hinders you.

Chris

[Andreas]

Hello,

the SVR-T cirquit is only linear in a small temperature range.
Since my AD587 references are nearly linear in T.C. over the full 10-40 deg C range I want to have a more linear compensation.
But of course there is nothing for free.

With best regards

Andreas

[lars]

Hello Andreas,

I can´t see why the SVR-T circuit shouldn´t be able to have a linear compensation if needed. With the series resistor set to 80% of the NTC (at 25°C) the compensation will be quite linear over 10-40°C. I had to find my old calculation and add 10 and 40C to the calculation as the old just covered 15-35C. Enclose a chart done with the 2x680k NTC in series with 1078kohm(includes internal resistance of Vref and trimpot node) and the trimpot set to compensate 7ppm/C. The residual is +-1ppm for a compensation range of +-105ppm (10-40ppm). The calculated residual has a max TC of 0.2ppm/C. Probably much better than other errors. I have used the datasheet values for the NTCS0805 680k (from Mouser or Digikey).

Your circuit for me just seems to divide the current between R2 and R3+about 2kohm (from the internal resistance and the trim network). One of my assumption is that the 10v output is very steady relative to ground compared to the change in the NTC-resistor node.
If my assumption is correct the 80% series resistance in your case is R2 in parallell to R3+about 2k and after that your circuit is very similar to the SVR-T. ( I haven´t run a simulation just calculated in my head so I might be completely wrong)

The circuit I showed before has the nice feature that you have a quite linear change of TC of about 4ppm/C per Volt change on the trimpot wiper (TP2). By temporary adding a resistance between the trim pin and 10V you can do more exact calculations by measuring the trimpin and 10V change. My last AD587JQ box had about 4.27ppm/C per Volt calculated.

For those that want to use my circuit, my recommendation is to first set the offset trim pot set to 2.5V (mid). When adjust the difference between the trimpin (TP1) and trimpot wiper (TP2) to zero, the TC compensation will be zero that is very useful for a first temperature test. When adjust the TC trimpot. Check again and maybe do another adjustment. Add R1 or R2 to get within say +-50ppm. Do the offset fine trim. Do a temperature test again to confirm.

This will also work for AD587 in plastic packages but the hysteresis (due to drying) during the temperature tests may be a problem.

Lars