Ultra Precision Reference LTZ1000

[jd]

Hi Frank,

Have you looked at LT5400? http://www.linear.com/product/LT5400 Networks of 4 resistors,  2ppm / 2k hours long term ratio stability. And not that expensive really.

A unique part as far as I have seen. On paper they look very interesting for making ratios, provided you can get lucky with the values available. But look, you can make the temperature controller divider using R4=(10+1) in series and R5=(10+1) in parallel. = 11k and 0.9090909k, equivalent to 12.1k + 1k. Perfect!

John

What for?
My design goal was < 1ppm/yr., and I achieved that without any schmuck devices or exotic PCB slots.

Use ordinary precision wirewound resistors and solid thermal management. That's it.

I had assumed that the slots are for thermal management. They should prevent any external thermal gradient from appearing across the zener terminals. On the Datron reference they also have wide copper tracks around the periphery of the square, which would tend to "short out" any thermal gradient. (Provide a "thermally grounded enclosure") Also the long thin PCB segments will tend to further isolate the LTZ1000 thermally, i.e. the temperature of the zener lead junctions will be nearly that of the LTZ1000 rather than the rest of the board.

Stability is now < 1ppm / 4yrs.
Frank

My original interest was for use in the 7.2:10V step up, but I think a PWM approach will be better for that. But in fact it ought to be a good solution for the temperature controller divider too, cheaper than two wirewounds, and much more compact of course. But in fact for my next reference I am using wirewounds for the precision resistors (as you suggest). And no slots either

John

[jd]

Hello all volt nuts - has anyone already experience using the LT5400-6 and LT5400-7 as matching resistor network [0.2ppm/K matching temperature drift] ... - one can easily fabric the 12.5k:1k0 and 12k0:1k0 ratios 12k5 = 5k0 + 5k0 + 1k25 + 1k25  or 12k = 5k0 + 5k0 + 1k0 + 1k0. The LT5400-x networks are low cost.

I don't think either of those get you the ratio, you need to do both parts of the divider in a single device as far as I can see. I.e. the 1k resistor has to be part of the same device as the 12.5k or 12k. The best I could find is (5+5+1)k : 1k, or 11:1, which only works if you use a much smaller R3 ~10k.

Another idea I got in mind - has anyone already considered to replace the Op Amp controlling the heater element with a low power (galvanic isolated) microcontroller that is executing a PID algorithm ? How close could we get in delta mK - would a sensor based on a thermo-couplebe accurate enough ?

Well I have considered it. There is some kind of LTZ heater PWM going on in the Datron reference circuit, but I have not figured it out exactly yet; not sure if it is only used for transportation mode for example.

I don't think a thermocouple would be accurate, you are probably not going to improve on the on-die sensor. If using an external sensor a thermistor is best for non-extreme temperature measurement.

Really, the more I study the canonical datasheet circuit, the more perfect it looks already.

[EEVblog]

NOTE: Several silly argument posts removed.
Come on guys, talk about voltage references instead...

[Blackart]

Thanks for you responses re the 400k Ill play with this in the prototype unit. Ive order all the bits for my ref and now have to wait for them. Then the fun begins.

As for checking the drift I don't have the luxury of a 3458A/1082 8 digit meter. But I do have a datron 4700. I was thinking I could weakly calibrate this to one of the one of the LTZ1000 REFs and then null the others of against it. At least the intrinsic drift go the 4700 is pretty low to start with.

Does any body have a better idea ? Oh Ive bought 5 off LTZ1000 to play with.

Lucas

[Andreas]

I found that you can use the resistor to fine-tune the temperature coefficient in the LTZ1000 circuit, so start with 400 K and put the circuit in a temperature controlled environment (I used a beer fridge that could heat/cool  ) Then you might change it to e.g. 470k to reduce the TC below even 0.05ppm/K.

This will also take care of any concerns about opamp TC.

Hello,

when looking up to my TC-measurements (LTZ1000A) that I did, I cannot see how the TC could be bettered by such a measure. (especially above 20 °C environment temperature).
Most of the TC above 20 degrees is related to some kind of hysteresis. There is no linear relation to the environment temperature which could be compensated in my case. The hysteresis might come either from the temperature gradient, from PCB, the LT1013 in plastic case or some effect on the LTZ1000A.

The best measure would be to heat the whole reference to a constant temperature in my case. But with battery operated device this will not be possible.

The diagrams show output voltage (divided by 2) over ambient temperature of the reference and over the internal temperature sensor near the LTZ1000A on the PCB.
Temperature gradient is 0.1K / minute.

With best regards

Andreas

[Dr. Frank]

Hello,

when looking up to my TC-measurements (LTZ1000A) that I did, I cannot see how the TC could be bettered by such a measure. (especially above 20 °C environment temperature).
Most of the TC above 20 degrees is related to some kind of hysteresis. There is no linear relation to the environment temperature which could be compensated in my case. The hysteresis might come either from the temperature gradient, from PCB, the LT1013 in plastic case or some effect on the LTZ1000A.

The best measure would be to heat the whole reference to a constant temperature in my case. But with battery operated device this will not be possible.

The diagrams show output voltage (divided by 2) over ambient temperature of the reference and over the internal temperature sensor near the LTZ1000A on the PCB.
Temperature gradient is 0.1K / minute.

With best regards

Andreas

Hello Andreas,

best candidate for this kind of hysteresis is the LTZ itself.

This has already been pointed out by Pickering, but without any hint of the magnitude order of this effect.

I also have seen hysteresis, but to a much higher degree, as I changed the LTZ temperature from -18  to +80°C, i.e. -18° for the whole circuit, +80°C for the LTZ alone. That rules out hysteresis effects from the other components.

Effects from the PCB and the LT1013 case also have not yet been demonstrated yet.

You also use wirewound resistors, normally they do not show hysteresis at all, in constrast to metal foil types.


What I do not understand: The absolute change in Uref is - 0.08ppm/K only, although it should be -50ppm/K.

Have you really compensated the T.C.?

Frank

[jd]

Hello,

when looking up to my TC-measurements (LTZ1000A) that I did, I cannot see how the TC could be bettered by such a measure. (especially above 20 °C environment temperature).
Most of the TC above 20 degrees is related to some kind of hysteresis. There is no linear relation to the environment temperature which could be compensated in my case. The hysteresis might come either from the temperature gradient, from PCB, the LT1013 in plastic case or some effect on the LTZ1000A.

The best measure would be to heat the whole reference to a constant temperature in my case. But with battery operated device this will not be possible.

The diagrams show output voltage (divided by 2) over ambient temperature of the reference and over the internal temperature sensor near the LTZ1000A on the PCB.
Temperature gradient is 0.1K / minute.

With best regards

Andreas

Hi Andreas,

I have not looked properly at hysteresis so I may have missed it. I did not notice any but it could be there, I will check more next time.

Here are some plots I made with two different TC resistors, the datasheet 400k and 330k. With my particular layout there was a clear TC with 400k which could be made to vanish with 330k.

The red line shows the temperature of the LTZ "module"; it is self-regulating to an extent so the excursion is a lot less than the module "ambient" (fridge) temperature it is in. The black line is the 3458a internal temperature which tracks that of the room.

[jd]

JD, are you using the LTZ1000 or the LTZ1000A?

Hi,

I am using LTZ1000.

[Andreas]

What I do not understand: The absolute change in Uref is - 0.08ppm/K only, although it should be -50ppm/K.

Have you really compensated the T.C.?

Hello Frank,

perhaps I did not explain it clearly enough:

The tempco of the LTZ1000(A) without heating (or by changeing the temperature setpoint) is around +50ppm/K.
In the measurement above the internal heater of the LTZ1000A was active. So the datasheet value of 0.05ppm/K is the target.
So the TC is not compensated but the chip temperature is held more or less constant.

TC compensation is done only on my ADCs (LTC2400) which I use for measuring the LTZ output values.

I plan on using a 16-bit PWM instead of a divider resistor for the temperature control set-point--   

If the Zener is making about 7.2V, then a 16-bit PWM should give me about a 0.05-deg-C step for each count-- thus allowing very fine grained control.

This 0.05 K step will produce about 2.5ppm/step (50ppm/K) output voltage change.
The interesting question here is how the remaining ripple of the PWM after filtering is minimized.
In order not to increase the noise of the LTZ (1.2uVpp) the remaining ripple should be well below 0.003K (or 7uVpp).

I have not looked properly at hysteresis so I may have missed it.

The red line shows the temperature of the LTZ "module"; it is self-regulating to an extent so the excursion is a lot less than the module "ambient" (fridge) temperature it is in. The black line is the 3458a internal temperature which tracks that of the room.

In your case the temperature change near the LTZ is only around 8 degrees against 25 degrees of my PCB temperature change. Since the hysteresis squares (approximately) with the excursion of the temperature it should be negligible in your case.
Btw. what is the exact type of beer cooler that you are using?
I use a standard thermoelectric cooler (car box).
But I have to use additional ice packs in styrofoam + some fan controller to reach the +10 degree Celsius.

What I am wondering, is which device has less thermal hysteresis-- the LTZ1000, or the LTZ1000A?  Does anyone have data on this?

Interesting question.

According to my measurements on several AD586LQ references the hysteresis shows large differences between individual devices.

Further I have the suspect that the hysteresis is larger on newer devices. And AD: has changed the type of die attach for the hermetically devices according to a change note on their web side. (probably due to ROHS?)

So you would have to test several LTZ1000 + LTZ1000A from around the same date code to answer the question.

With best regards

Andreas

[Galaxyrise]

Of course there are the other bits that need to be addressed as well-- I want my 10V reference to have battery backup for an extended period of time (for shipping the device "hot" to/from a cal-lab that has a JJA)....  I am leaning towards the CALB 40Ah if I finally decide to go with the Peltier-cooler method...

In my experimenting with Peltier cooling so far, cooling performance of the hot side is very important for performance of the cold side.  This would seem to further complicate mailing it "hot", where the reference will be inside a bunch of packing material.  Did you have a plan for mailing a peltier-cooled reference or is not mailing it a reason you don't care the CALB isn't flight qualified?

[CaptnYellowShirt]

Yes-- the Peltier system will need to breath outside air in order to function correctly.  My plan is to build a special carrier with inlet/outlet vents for airflow from the fan that cools the hot-side of the Peltier device.  The carrier will act as the "shipping container".

You could pipe the heat to phase-change sink of some sort. Dry ice (maybe too cold)? Regular ice? One of those semi-solid freezing gel packs? etc

[CaptnYellowShirt]

Hmmmm... That's a good thought-- but are those things allowed on cargo planes-- (I know that some of those you mentioned are not allowed in the cabin of a passenger airliner)?

Air flow might be questionable:


But I'm sure you can find something that'd be allowed. Kerosene?

[Andreas]

Maybe the easiest thing to do is restrict the time of year to ship the reference off for calibration to the more temperate seasons-- (avoiding winter and summer-- perhaps in the fall when it is dryer)...  This would only happen once a year anyway...

I do not know how the humidity behaves in your region.
Here in my region the smallest change in humidity is in spring after winter.
So this is the best time to calibrate all your gear if you want to import the "volt" into your lab with the smallest changes during transport.

With best regards

Andreas

[quarks]

What is the ideal value/range regarding humidity in your opinion?

My very best gear uses LTZ1000 and I now have a Fluke 732A. So far I have not seen an effect with changes from 45% up to 55% on this gear.

Right now I have around 45% and made very interesting meassurements with my TEK 4050. As I already posted several times, the Fluke 8846A / TEK DMM 4050 are great meters. In my opinion the very best you can buy. I only wonder how it is possible with "only" a LM399 reference inside, because  my 4050 even challenges my gear with LTZ1000 reference.

[Andreas]

Hello,

There is no best value. In best case it is constant.
Most instrument datssheets specify < 80% rH non condensing.
The effect is 0.5 ppm/%rH with buried zener voltage references in plastic case. (LT1027CN)
Something similar has been shown in early datasheets from LT1236AILS8 in hermetic case mounted on pcb material.
In this case the changes of the pcb create forces on the chip.

With references (LM399) that have good decoupling from pcb the effect is lesser.
I see around 1-2 ppm over 1 year (cycle) when comparing a well decoupled (from pcb) AD586LQ against a LTZ1000A.
So with a 6.5 digit multimeter you might probably have difficulties to see it.

With best regards

Andreas

[Galaxyrise]

I don't know how radical I need to get in cooling the LTZ1000--- maybe just running it at 25C die temperature (+/- 0.001C) will be sufficient, and further chilling might be wasted because of no further decrease in drift rate.  All of that remains to be determined after construction and testing.  My plan is for a die temperature of 0C, but that may be "overkill", and if 25C *is* sufficient, then the Peltier device will have much less work to do, and the power needs from the battery will be more reasonable.

I'm thinking of trying a three layer approach, where the die temperature and the support circuitry are both held at 10C.  I'm not sure if I'd ever know I succeeded in reducing thermal gradients on the leads, but it seemed like a neat idea.  I'm also guessing 10C should be better for the stability of all the precision parts, so long as they're hermetically sealed.  My lab tends to be 15-20C, 25-65% humidity; so 10C is above the highest dew point and relatively low power to maintain, simplifying the build.  Still many Saturdays and parts orders to go before I know if it's within my abilities to realize this 3-layer design, hehe.

It will take some things out of their tempco sweet spot, but hopefully the temperature regulation will be tight enough to compensate. I will need some good power filtering, too, since having an inner peltier cooler means bringing that bucked voltage into the shielding box. Thanks for posting your PWM filter

As this is going to be the most stable thing I have by several orders of magnitude, I may never know if I chose well.  If I ever go for calibration against a primary standard, my "plan" is to have a road trip  

[CaptnYellowShirt]

 I am going to contact them to see if they would give us "Volt-Nuts" a break on calibrations-- (no certificate of compliance, just a "before" and "after" if an adjustment is involved). 

You had better put your game-face on. Calibrations at NIST start around $10k. And all they do is hook your machine up, hit the go button, and tell you how far you are off. Forget having time to tweak it.

I'm reminded of the book "Foundations of Mechanical Accuracy" in which the story of replicating the meter standard in the 1970's was told. The Moore Tool Co would send a rep to France with a briefcase that contained what they though was some fraction of a meter. The guy hops on a transatlantic flight, sets up the standard next to the one in Sèvres, waits how ever many hours/days you need to make sure everything's at the right temperature, and finally shines a laser at the ends of each to figure out how much they are off. The guy gets the report, flies back home, sets up the standard in a room inside a room in the sub-basement of the Moore Tool Co. and they take like a month to file off their best guess at another nanometer. All this just to fly back to France and try again and again until they've gotten "close enough" or they've gone too far.

[Dr. Frank]

.. Well, after 2 weeks of this torture, the readings in the 3458A suddenly dropped 0.25ppm (when it had been solid, with no discernible drift, for the first part of the year). ...

 and the 3458A dropped another 0.25ppm!  [The 3458A has never recovered-- even though the humidity has been around 8% for months now, it is still low by 0.5ppm-- so I think it was permanently affected].  ...

 My Fluke 732B did not suffer at all through this (as far as I can tell-- I don't yet own a JJA)-- the important resistors in the 732B are all hermetically sealed.

Hi,

If you don't have an JJ, how did you distinguish, or decide, that the 3458A was drifting absolutely and not your 732B?

I assume you noticed the drift by comparing the 3458A against the 732B?

You know, what I mean, the "man with the 1 clock, 2 clocks, or 3 clocks" problem.

Anyhow, if it was really the 3458A, there are 2 possibilities of a permanent shift.

First, the sudden high humidity caused a sudden shift , i.e. an irreversible ageing, of one or two of the bulk metal foil resistors, about 25ppm in total.

Second, the humidity caused the temperature stabilization to fail, heating the LTZ even more than 95°C, where it is supposed to be normally.
Maybe you did not log that event.

This might have caused a reversible but permanent  hysteresis, resulting in a smaller reference value, so the 3458A reading should go up afterwards.

Would be interesting what  you have observed, and how you argue about the 3458A/732B comparison.

And I am interested, how you basically monitor the 3458As stability , against that single 732B only?
Do you leave the 3458A always on?
What if you switch it off, does it return to its former value?

Frank

[Dr. Frank]

I don't know where in the country you are, but Sandia Labs has a Cal-Lab that is humidity controlled (one of the few in the USA), and their uncertainty for Zener reference calibrations is 0.017ppm !!! 

No big deal, if they own a JJA and a decent nV voltmeter. Comparisons to the level of <= 0.001 ppm should be possible.

But which Zener based references have that level of stability?

Frank

[robrenz]

I'm reminded of the book "Foundations of Mechanical Accuracy"

One of the most treasured books in my library. It inspired me to buy (3) 18" square surface plates and take them to 15µinch flatness using the 3 plate method back when I was about 22. I studied that book like the bible and the skills I learned from it served me well thru my mechanical career. I ended up buying a Moore #3 measuring machine where I worked and many trips to the Moore special tool co.

[SeanB]

If you are going to have very precise references you will note the accuracy and cal specs require them to be in a temperature and humidity stable location at all times, just for the ageing. Every cal lab will have all of the precision equipment in a room with a hermetic seal on all the doors and no windows aside from a small one in the door, with airconditioning just for the room and a heat exchanger on the room ventilation. This will be to keep the standards at constant temperature and humidity. The standards lab near me has the mass standards in glass cases with felt linings, all in a room with thick brick walls inside the middle of the building. There is a balance in there as well to check the transfer standards against them. Those then go into the room next door where they can bring the equipment in that they are calibrating.

[CaptnYellowShirt]

I did not think about the hysteresis problem-- is there any way of determining if that is the cause, and if so, is there a way to fix it (put the reference board in the freezer or something)?

I just had an image of a kegerator-style DMM flash into my mind.  Mmmm... 8 1/2 digits of frothy precision.

[CaptnYellowShirt]

I have an idea to fix this calibration cost problem for us "Volt-Nuts".  If it turns out to be feasible, I will announce the solution here and on the "Volt-Nuts" list.

I've been thinking that rather than calling and asking nicely, we (the collective group here) should publish a paper on our experiences. If we could get it into Metrologia, I'm sure Sandia (or whomever) would be much more receptive to returning your calls.

[hgg]

Do you know what is the initial %accuracy of the LTZ1000?
I cannot find any info in the data sheet.  Only its stability of 0.05ppm/C is emphasized. 

I've bought a MAX6325 1ppm/C 2.5V reference with a 0.02% initial accuracy.
How does it compare with the LTZ1000 in terms of initial accuracy?

[TiN]

It's first item in electrical characteristics on page 3. Anything from 7.0 to 7.5V with Iz 5mA, or from 6.9 to 7.45 with Iz = 1mA.