Ultra Precision Reference LTZ1000

[ap]

Well, the true ground point should be at the output of the unit. Now for a PCB, one could have that at pin 7, for a unit in a box with e.g. binding posts, things look different. It should be at the negative output binding post. It needs to be set in a way that the change in current through the related wire from pin 7 to the output (load effects, temperature changes / related copper wire drifts and current accross) is sufficently low. The output impedance of a 732B e.g. is 0.1mOhm, as a side comment. So 10mA load at the output causes 1ppm change in voltage. It musst be assured, if that e.g. is the goal, that any return current from that binding post to pin 7 is not generating a ground shift.
The filter capacitor should be a lowest leakage foil capacitor, and the input resistor could be higher, provided the bias current (change again) of the opamp is low, the capacitor will anyway filter resistor noise.

[Micke]

I did at least not drive the output current via pin 7 of LTZ1000... I do have built in the PCB in a metal box with 2 pair of binding posts (Buffered 7V with LTC2057 and 10V with LTC2057) , the output negatives are done with separate 0,75mm² cables directly to incoming ground on the PCB. I have done load tests, for example loading the 10V output with 10mA, the output drops 7,6ppm... not great result, but most likely I will not load the outputs that much   
But I realize now I should have done like your good explanation, have the output ground as the common ground point... 

[mimmus78]

What you think about using LT1037 to buffer LTZ1000 output with a LTC2057 to zero out offset (same circuit as shown in LTC2057 datasheet).
I'd like to use it in my 4 TLZ1000 reference as "Ultralow Noise Composite Amplifier"

The problem I see is that LT1037 has a little much input bias current than just LTC2057.
This should be not much a problem as this input bias/offset current seems to be constant from 20°C and up ... I think you will get some nA difference from 25° to 45°c that is the internal temperature of the reference.
What will be the influence on the LTZ1000 of this nA bias current offset?

Also LT1028 is interesting, but input bias current is not linear over the temperature range.

[Kleinstein]

The standard LTZ1000 circuit can deliver quite some current. The OP is delivering the current and the zener connection is used for sensing. The variants with an extra transistor as emitter follower can even deliver higher currents if needed. This goes up to the point of providing force/sense connections to compensate wire resistance.

The problem why sometimes buffering for the output is desired is, that if the load to the output is so high that the voltage drops, this shifts the temperature set point to a higher value and thus can easily cause the heater to turn on all the way. This thermal stress can cause a shift in observed voltage, e.g. due to thermal hysteresis. In theory one could add an extra protection against this: so limit or turn of the heater, if the reference voltage drops by more than something in the mV range  (enough to allow a cold start).

So there is no really problem for the ref circuit to drive even a much higher input current in the mA range, if this internal. Current spikes from the LTC2057 would be the greater concern. So it is a good idea to have an RC (maybe LC) filter in series, at least for that OP. Without extra filtering, the white noise of the LTZ1000 is way higher than that of the LT1037 - so if would make sense to have at least some filtering here as well.

The LT1037 is not unity stable, the LT1007 is the unity gain stable version.

[mimmus78]

Thanks Kleinstein, I understand this now.
By this buffer I want to obtain short circuit proof circuit, and a more easier way to sum up the four cells without much interferences and current loops one each other.
LTC2057 will be only used to zero out the offset. His spark noise should not came out directly and I was considering RC filtering both at the inputs that at the output of it.

What op amp to use as current cancelling?
For this I'm considering again LTC2057, this time it will be tied up to a very low impedance to ground node, I was thinking in this case I can ignore those sparks.

[Kleinstein]

Current canceling usually should not use an AZ OP - definitely not a high current noise one like the LT2057. More like a normal precision one, like LT1013 (if single supply is needed), OP07, LT1097 or maybe OPA141. In some areas even an TLC271 or LM358 might do the job. Having a small DC error here is not that critical. More of the error will likely be due to the resistors anyway.

Just for averaging, there is no need to have a buffer. The LTZ1000 circuit has no problem driving the rather constant maybe 10-100 µA needed for a resistive combiner. An important principle in a precision circuit is not to use many precision parts, but to have a circuit that does not critically depend on many parts. So it is more like having a minimum number of amplifiers in the signal path. So one would rarely have something like only a buffer - so it is a little like going back and learning from old times when an OP was expensive and not that good. No OP can beat a piece of copper wire when it comes to drift and noise. There can be still quite a few parts in not so critical areas, like protection, current canceling and the power supply or monitoring.

[ap]

Why would the current cancelling Opamp's Vos drift not have the same impact on output drift as in the buffer amp (amplification factor, if any, left aside)?

[Kleinstein]

There are different ways of doing current canceling / combining the negative side. It depends on the solution chosen of cause:
Using a kind of buffer or virtual ground for the whole circuit would have a large impact of the OPs offset. In this case a really good OP is needed, and the LT2057 might be acceptable.

A second, similar option is using resistors to combine the low side voltage too and than use a buffer with optional force/sense function. Here too a good OP is needed.

If one uses classical current canceling - thus sending a constant current of the right size to the GND point, the OPs offset is even less important. It is attenuated by a factor set by the resistor used to set the current and the resistance of the ground loop - so usually something around 100-1000, maybe more. However the resistors used in the compensation circuit also contribute, likely more than the OP.

An other option is using a virtual ground only for the main current ("120 Ohms " resistor) and keep the transistor to sense the ground. In this case the drift / noise of the OP used is attenuated by a factor of about 10. So drift is less important by that factor. One can combine this with current canceling for the 100 µA at the transistor if really needed.

So my idea was about using the last option of current canceling by current sources, as this should give the best results.

[d-smes]

The standard LTZ1000 circuit can deliver quite some current. The OP is delivering the current and the zener connection is used for sensing. The variants with an extra transistor as emitter follower can even deliver higher currents if needed. This goes up to the point of providing force/sense connections to compensate wire resistance.  <clip>

If you pull several mA from the Pos output, where/how do you set up a return current path such that it does not corrupt the output with a I*R voltage drop?  Do you guys buffer the ground and use force/sense connections?  Or just have one massively heavy-duty ground connection at R5, R1, Pin 7 super-node?

[Kleinstein]

With an additional current source to provide the approximate current needed by LTZ1000 Zener and the divider for the temperature set point, the positive side of the zener can act like a low input-current sense connection. The OP (usually LT1013) will act as the forcing output.

This is similar to return current compensation on the negative side. However the negative side might be a little easier with only about 100 µA from the transistor used for sensing.

However the more normal situation is to have an only very moderate load current, so that just a single ground point and low impedance connection. The really troublesome part is only the possible change in load current. The more predictable part of the load current could be added locally at the load.

[chris_11]

Did anyone try to force the LTZ1000 PCB temperature with a TEC? The TEC would be linear controlled to avoid switching noise. My plan is to use a PCB temperature setting close to typical ambient I.e. 22 centigrade and use a LTZ1000A with only about 45-50 dice temp. The thermal voltage between the dice and Kovar leads and again between the Kovar leads and the PCB copper is a contributor to the stability. That's the main reason for the hood over the LTZ1000 in the HP3458A. The only cure is to keep the differential temps at the Kovar to copper interface at the PCB as stable as possible.

Christian

[Andreas]

Hello,

and how do you keep humidity from condensing at 22 deg C?

with best regards

Andreas

[chris_11]

Keep it sealed and put a dryer bag inside. For condesation on a 22 centigrade surface the humidity and temp would need to be high. Definitely not metrology grade environment.
Ovenized only means temp higher than worst case ambient. Then the LTZ1000 has to run even higher with higher aging/drift. And the oven draws always max. power which needs big batteries for a transport standard. TEC should be relativ efficient when the environment is close to the set temp.
My idea is to run everything from a 12V SLA without any DC-DC. For transport with cigarette lighter plug in for recharge.   

[Andreas]

Hmm,

to keep the PCB of the LTZ down to 22 deg C the TEC has to be lower than 22 deg C.
Assume 17 deg C. -> In my "Lab" a dew point temperature of 17 deg C is not unusual in summer.

With best regards

Andreas

[Echo88]

@MisterDiodes, Edwin Pettis: What would you suggest as a costeffective and good LTZ1000 -> 10V schematic which can be adjusted for drift?

I would assume the circuit from the Linear AN86 page 46 (LTC1150 + LT1010) with a PWW-resistor-network from Mr. Pettis + low resistance Bourns 3250W trimmpot/Hex-switch which connects PTF56-resistors. The whole thing, including the LTZ1000 of course, temperature stabilized by a PID-controller like the ADN8834.
But maybe there are better solutions?
 

[MisterDiodes]

Circuit for stable 7V to 10V conversion?  That all depends on your final application.  The first question to ask:  Do you really need 10V?  Why?  How quiet does that Vref  need to be?  If you're trying for <5ppm / yr drift, do you even have a way to measure for that accurately?  If you do, that would mean you already have multiple calibrated references on hand.

7V to 10V accurate conversion is not a trivial problem, and is harder than building the LTZ circuit - for instance:

A 732a/b comes to mind as a good design example... depending on how much drift and how often you need to adjust.  That's basically what you're talking about, at an extreme level.  If you need a very stable 10V, just buy a 732'.... You won't make anything near as good as that for the price (used or new), especially if your time is worth anything.  Depending on the definition of "good" of course.

A PWW pot with PWW resistors will make an acceptable, economical boost unit for an LTZ.  Talk to Edwin, and he'll make you a set of ratio resistors.  Use a quality Bourns PWW pot for adjustment.  How well you build it from there is up to you, and that could take an infinite number of forms - depending on what you need.

It also depends on what load you're driving - how many mA, op-amp selection, output sourcing + sinking requirements, current limiting,  hysterisis, temp. span power cycling, vibration, thermal flow, etc etc. etc.  The design parameters list is very, very long and is impossible to answer here.

You also have to decide:  If you're driving an ADC Vref input, pay attention to the drive current required.  That's a gotcha on newer fast ADC's.

I wouldn't worry about ovenizing if this your first project.  Build your device to not need an oven first, and see if you really need one.  Add the oven later and see where it degrades performance.   If you don't have a way to really measure down to low PPM accurately, you won't even see if the oven is doing anything for you.

Those PWW wrapped around an amp will get you started on a simple 10V boost circuit.  It won't be 732a/b level, but maybe you don't need that?

If it were me, I'd attenuate the LTZ output down to 5V or 2.5V...much more useful for ADC / DAC work, and you're not gaining up the noise. For an accurate, stable, very low drift 10V do it right and get some working 731's or 732's first... otherwise you won't have a good way to verify a DIY circuit.

EDIT: Somewhere here Dr. Frank has has a published design for his 7V to 10V Vboost circuit.  I don't have time right now to look up the link but that will get you started.   Also:  If you're going to all the work to make a 10V converter, just have Edwin make you an LTZ resistor set - further reducing the need for a noisy, complex oven system. That will make for a quieter LTZ circuit.

[Echo88]

Lets be honest here: the nut in voltnut has its reason  I frankly dont need an ultrastable reference, im just interested in the technology/necessary skill to build them. My problem is indeed that i dont own a 3458A to even judge my selfbuild reference (>3,5k€) and 732-references are likely >1k€ here in germany, if they even appear on ebay. My 6.5 digit DMM 34465A is at the moment all i have, but of course i aim for better equipment.

At the moment i have a LTZ-reference based on TiN KX-board, connected to a zener-to-10V-converterboard based on Andreas suggestion: PTF56-resistors, LTC2057, 12-Bit-DAC to trimm it nicely to 10V with about 5µVpp noise, according to the 34465A. But since the 7V-10V-converter consists of a lot of those resistors i dont have great confidence in its drift and want to change the design to a 2-resistor-OP Amp-design like you mentioned. I will search for the design approach of Dr. Frank like you mentioned and have a look at it.
Im not really interested in using the reference with an ADC, since discrete ADCs like the LT2400 or LTC2508-32 dont come even close to the specifications of the Multi-Slope-Designs used in 8.5 Digit DMMs...also i simply lack the necessary amount of skill and experience to get the reference-voltage to the ADC without introducing those nasty errors because of the ADC-input-current-peaks (like mentioned in the LTC2508-32-DS).

The 732B-oven-controller really isnt that special, im wondering why you say that such an oven contributes to the noise? I thought the only downside of oven-stabilized-designs are the additional thermo-emf-voltages caused by the temperature-differences between oven and measuring-DMM at ambient temperature?

All in all, i would still have to build 3 or more references for intercomparison reasons, to get the necessary amount of confidence and the ability to judge inidividual drift and errors.

[ap]

Well, if this is mostly due to educational purposes, why dont you try different approaches with the three samples you want to build. Use different techniques/parts, options have been described here. To judge the results though, while intercomparison will give you a clue whats going on, in the end you need some type of starting point and end point reference, everything inbetween could still be relative, that way you limit the effort to have external comparision to precise reference sources (as you dont have your own calibrated one) but still know what you get.

[rigrunner]

I've had the parts to build this for a while and have finally found a little time to put it together.

This board was courtesy of Svangool - thanks Sjef 

I'll find somewhere safe to leave it for a while and let it burn in.

[mimmus78]

I've had the parts to build this for a while and have finally found a little time to put it together.

Any reason why you thermally coupled also the 70K resistors other than the divider?

[rigrunner]

Any reason why you thermally coupled also the 70K resistors other than the divider?

No great rationale other than they are in parallel feeding q1 and q2 on the LTZ and i didn't think it would hurt to couple them.

[lars]

After reading the thread “How do you measure drifts of the order of 1ppm/year?
  https://www.eevblog.com/forum/metrology/how-do-you-measure-drifts-of-the-order-of-1ppmyear/msg1198288/#msg1198288   I am curious what maximum long-term stability the LTZ1000 really have. The datasheet just say some typical value. If I buy an LTZ1000 from Linear or Digikey what can I expect as maximum drift (with reasonable confidence) if I go for the design in the datasheet but set for 45C with 12k:1k? Assume perfect other components.  Say that I let it be powered for 3 months and calibrate it after that. What will be a reasonable uncertainty spec for 1 year? 1-2-5-10ppm? I can´t find a lot of data for a not selected LTZ1000. DCV standards like Datron 4910 and Fluke 7000 series probably have selected LTZ. Assuming I don´t have the possibility to select my LTZ it would have been nice to have more data from others on non-selected units. Spreadbury´s paper seems to be from a batch of 24 non-selected LTZ but probably from the same batch (and 28 years ago).  As Spreadbury says it seems that every 10 degrees C seems to halve the drift but for me the statistics are to low and I am not sure this rule applies here as little as for voltage references from Analog Devices. If this was true, a reference at 95C would have 32 times higher drift than a reference at 45C. Even from Spreadbury´s paper the variance seems to be high and his data are extrapolated as I understand it. The other paper Dr.Frank mentions by Pickering, Fluke, Metron, etc. I can´t find on the internet.

Another paper I have found and that confuses me due to the large drift of the Fluke 7000 series: In Andrei Pokalitov´s very interesting thesis “Development of National Standard for Voltage Unit Based on Solid-State References” he has followed 10 Fluke 7000-series modules and from what I see the average drift the first year is -2.3ppm (between -1.4 and -3.8ppm/year for a single module) that is far more than the -0.8ppm/year I found in the user manual from Fluke and also the figure Dr.Frank probably refer to in the thread for the LTZ1000 drift. Also the stability specification in the manual is 1.8ppm for 1 year. Even after 3 years the drift Pokatilov saw was in average -1.3ppm/year and the single modules varied between -1.0 and -2.2ppm/year.
After about 1 year one module (Z5) of the ten makes a jump that shifts the average about 0.3ppm so my conclusion is that the LTZ in Z5 made a jump of about 3ppm! In another chart, another module (Z12) seems to make a jump and also shifts drift rate.

So what to believe about LTZ1000 reasonable maximum stability over a year at 45C internal?
What conclusion can I do if I intercompare three newly bought LTZ1000?

Of course I know you can do a lot of other mistakes in the LTZ design but consider only the new bought LTZ. From the above could 5ppm for a year be a reasonable spec (with the same confidence as I see from eg DMM manufacturers) from the above or am I totally wrong?

Lars

Ps. Ok sending out this question only hoping to get a good discussion.

[Andreas]

Hello Lars,

in the datasheet there is no upper limit for the drift.
And so it is not predictable for a single device.

They cannot look at the purity / crystal defects of every
single device which might influence the ageing.

As I observe increasing popcorn noise on newer devices (references)
it might also be that after the tsunami in Japan (Tepco)
the puirty of silicon is no longer the same.
It might of course also be that I now have better instruments to detect popcorn noise.

And final even when you have built the LTZ you never know if the drift itself will be stable or even decreasing.
Each "event" like (temperature) shock or short cirquit might start a new ageing cycle with different ageing speed.

with best regards

Andreas

[Dr. Frank]

Andreas, most of the wafer silicon comes from Wacker Chemie, very probably outside Japan. Wafer production also should not have any conjunction to the Japanese Tsunami and Earthquakes.
LTZ1000 production is in California, so also no connection.
It's always possible, that one batch is bad, but that's old, large analogue structures, so not that probable.

Frank

[Andreas]

Hello Frank,

afaik there are only 3 large scale producers on the world for single crystal silicon.
one in Japan
one in the US
and of course Wacker Chemie in Burghausen.

with best regards

Andreas