Ultra Precision Reference LTZ1000

[EEVblog]

Below is a link to the HP Journal that highlighted the 3458A which would have appeared soon after product introduction. Again it makes no claim for designing the LTZ.
http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1989-04.pdf   ... “

Wow, tons of cool stuff in that, and lots of names 

[MisterDiodes]

I can also vouch for the LTZ being available to everyone before 3458a was even on the market - I was there on the fab line.  Our company was working with LT on some die production issues of the LTZ and similar references, and I remember HP and other manufacture's feedback on a higher insulating epoxy bond attach, better wire bond attach and other improvements that made the "A" version of the LTZ a much better device for production than the first non-"A" version.  HP was building on experience gained from the 3456a and the LM399, as was LT.  The LTZ was absolutely NOT HP's design, but HP's (and other's) feedback did help to improve the device later.  That's when there was collaboration going on between the designers, various suppliers and the manufacturing fab lines.
'

[d-smes]

2.   It uses Vishay epoxy S102 resistors (!). HP used their TaN technology for the resistors elsewhere in the 3458A and I believe that they would have recognised that only one resistor in the LTZ circuit needed to have absolute stability, the others are all ratio defined which would be excellent built with their TaN film.

I don't recall this subtlety of the LTZ circuit being discussed before.  i.e. the ratio of R3 to R2.   One of my LTZ builds used 68.1K in series with 2K to form the 70K resistors for R2 and R3.  Wondering how critical the exact resistance had to be, I shorted out the 2K portion of R3.  Output voltage increased 150 uV from 7.11835V to 7.11850V.  I removed the short across R3's 2K and shorted out the 2K portion of R2.  Output voltage increased 850 uV.  So far, so good. My calculated sensitivity of output voltage WRT R3 is -0.074 ppm voltage per +100 ppm change in R3 and -0.42 ppm voltage per +100 ppm change in R2 which is consistent with JanAF's (and others) sensitivity of -0.07 and -0.4 ppm/100ppm for R3 and R2 respectively.

Here's the surprise-  When I shorted out both 2K portions of R2 and R3, the voltage returned to the original 7.11835V.  In other words, decreasing both R2 and R3 by the same 2.85% had no effect on output voltage.   This implies the ratio of R2 to R3 to be far more important than absolute value.   Comments?

[Andreas]

Hello,

interesting finding: so you should keep both collector currents at the same value.

with best regards

Andreas

[lars]

I had to do a quick check on my prototype LTZ1000 box that still have jumpers to lower the original R2 and R3 69.8kohm by paralleling 6.8M.

What I see is:
a rise of 320uV for paralleling R2
a rise of 50uV for paralleling R3
a rise of 370uV for paralleling both R2 and R3

Buffered 7V output quickly measured with a Keithley 2000.

Lars

[Andreas]

Hmm,

this result sounds more logical.

By the way Lars: you know how to display one more digit on the K2000?
https://www.eevblog.com/forum/metrology/getting-one-more-digit-from-a-6-5-digit-meter-without-using-gpib/msg1300529/#msg1300529

with best regards

Andreas

[lars]

Hmm,

this result sounds more logical.

By the way Lars: you know how to display one more digit on the K2000?
https://www.eevblog.com/forum/metrology/getting-one-more-digit-from-a-6-5-digit-meter-without-using-gpib/msg1300529/#msg1300529

with best regards

Andreas

Yes I have seen that. But normally, if I need an extra digit I use GPIB and averaging in software. For this test I didn´t care.

Lars

[d-smes]

I had to do a quick check on my prototype LTZ1000 box that still have jumpers to lower the original R2 and R3 69.8kohm by paralleling 6.8M.

What I see is:
a rise of 320uV for paralleling R2
a rise of 50uV for paralleling R3
a rise of 370uV for paralleling both R2 and R3

Buffered 7V output quickly measured with a Keithley 2000.

Lars

Thanks for checking this.  I did measurement on a different unit (nominal 7.13692V) and got:
    R3 (70.1K --> 68.1K)   +100 uV
    R2 (70.1K --> 68.1K)   +880 uV
    Both decreased            +990 uV
Conclusion- I screwed something up in my initial test.  Will have to take that one apart and try again...

@Flinstone-  What did you mean by "the others are all ratio defined"?  I assume R1 is the "only one resistor in the LTZ circuit needed to have absolute stability" and clearly the R4/R5 ratio defines the operating temperature (absolute value not important).  So what's the other ratio if it's not R2/R3?

[mimmus78]

The one resistor is the current setting resistor.

The temperature setting resistor con be calibrated.
Just measure the 7V zener out with a 6.5 multimeter and what exit from the divider. If you keep this ratio constant with a pot you are done, you don't need super extra stable resistor.

The 70K resistor have very high suppression ratio and even higher if you use 100K ohm.

So you end up with one resitor.

Hope I'm not wrong.

@Flinstone-  What did you mean by "the others are all ratio defined"?  I assume R1 is the "only one resistor in the LTZ circuit needed to have absolute stability" and clearly the R4/R5 ratio defines the operating temperature (absolute value not important).  So what's the other ratio if it's not R2/R3?

That's secret proprietary information...  I'm certain Flinstone has been asked not to repeat this...  It is very similar to my design-- only *one* high stability resistor (in terms of absolute value) is needed.  There are enough clues in this thread to figure this out on your own...

[Dr. Frank]

I have assembled and tested my 5 new LTZ1000 reference boards.

They all are based on Andreas schematic, providing additional EMI filtering for great immunity against RF and mains disturbances.
The single sided layout, gathers all possible thermo-couples on the bottom side, and accepts through hole components only, PWW as well as MBF precision resistors.
Star points for +Uref and - Uref separate the different currents of the reference amplifier and the supply currents. All capacitors are WIMA foil type.

All modules are set to nominal 45°C oven temperature, by a 12k/1k divider.

4 of the circuits (LTZ #1 .. #4) are completely mounted inside a tuner box, and an interior styrofoam isolation, latter being intended mostly for cancelling any air draughts, and equalizing the interior temperature. It is planned to mount these 4 modules inside an outer metal box, including isolated PSUs. Inside the small compartment of the tuner boxes, there's room for a small and simple 12V stabilizer.

The fifth box (LTZ #5) is built on a PCB from Andreas (thanks again), which contains a tuner box on the top side, and is enclosed in an outer aluminum case. Maybe later I like to add a battery supply for use as a Travelling Standard.

These non-A type LTZ1000 allow a trimming of the temperature coefficient by an appropriate choice of R9.

Both the reference voltage, and the amplified 10.000V are fed to CuTe jacks (Pomona), so that the 10.000V can be re-adjusted at any time by measuring the known ratio between both potentials. Any good 6 1/2 digit DMM should be good for a few ppm precise 10.000V output.

These two blue resistors which provide the 7.2 -> 10V stepUp, are BMF type from AE, and the precision NTC measures their temperature in situ, for a possible correction of the 10V / Uref ratio.


The T.C. measurement of the 10V/Uref ratio just of today reveals a value of < 0.1ppm/K, although I did not (yet) determine and match the BMFs T.C.s in any way.

Edit 21.4.2018: Update schematic for correct OpAmp LTC1052.
C7, C8 chopper capacitors are tied to GND for the 1052.
The MAX420 and the 7650 require to connect them to Cret, pin 5.

[Dr. Frank]

The new boards are an improved and miniaturized version of my prototype board, which actually went live in 2005 already.
The prototype layout contained several errors, and I reversed the heater supply, pin 1+2, so I discovered these substrate diodes, and LT changed their datasheet for the first time, afterwards.

The first measurements showed noise of  1ppm peak to peak, mainly caused by the 34401A I used.
But there were many spikes and also longer glitches.
The spikes were picked up from the noisy environment, and some of them caused the oven regulator to step out, which caused these ~ 30sec long glitches.

I made a lot of experiments especially on REF_1 (7.176..V), including accidently shorting the direct output, which caused big permanent shift effects. I could remove this hysteresis by temperature cycling.

After 12 years of nearly permanent operation, these references read 7.176 212V and 7.147 957V, so that's an average annual drift of < 0.8ppm / year. Currently they both show about - 0.6ppm/year.

I later moved my whole metrology equipment to our basement, where temperature changes during the experiments are as low as a few tenths of a °C, and the EMI disturbance is quite low,
as all switch mode power supplies are banned from this room, apart from the supplies in the measuring P.C.s
Recently, I replaced a 'stinker', inside one of the P.C.s.
At least the snubber network inside its PSU failed, creating a lot of disturbance.

The output voltages of the new modules now show no spikes, nor any glitches any more, as the 28h stability measurement (out of the box) on LTZ #5 demonstrates, thanks to better shielding, and due to the additional blocking capacitors, designed in by Andreas.
The measured noise is about 10 times lower than with the 34401A and the prototypes, between 0.025 and 0.050 ppm (rms).

This LTZ #5 is already trimmed to about 0.020ppm/K, so all apparent voltage changes are most probably caused by the 3458A only.

The warm-up time of the whole LTZ assembly to < 1ppm is about 15 minutes.

The change of the internal temperature of the 3458A is clearly visible in the measurement.
The big dip of -0.5ppm after 21.5 hours was caused by my wife when she opened the window, and so unintentionally determined the T.C. of my 3458A, i.e. +0.4ppm/K.

I measured the oven temperature of the prototype REF_1, by fast sampling of the base-emitter voltage of the transistor inside the reference amplifier.
As its collector current is constant, this U(BE) is a direct measure of the oven temperature, it's changing by quite precisely -2.1mV/K (see AoE).
The first reading of U(be) after powering on represents room temperature (about 22°C), the last and stable reading represents the oven temperature.

The U(BE) of the oven regulation transistor cannot be used for that purpose, as this voltage is fixed by the 12k/1k divider.

The diagram shows the self heating effect. As the LTZ1000 is very well insulated, upside down in a double styrofoam enclosure, this 10K temperature rise would be typical for the A version also.

The stabilized oven temperature is about 51°C, and between 50..54°C for the new modules.

The U(BE) of the new modules were measured with the 34465A, sampling each 1 msec.

I also measured the collector-emitter voltage of the oven sensor transistor, which shows the typical 7ms time constant of the 100nF/70kOhm in the beginning and about 500mV/K sensitivity of the sensor output.

The compilation of the reference amplifier change over temperature shows a nice linear T.C. of +55ppm/K

[Dr. Frank]

How to measure the T.C. of the modules on the order of 0.05ppm/K , when the reference DMM, the 3458A itself has a 10 times higher T.C.?

The bootstrap solution is to first keep the temperature of the 3458A to within 0.1°C constant, and to monitor its internal temperature. Then the modules temperature will be changed by 10°C by forced heating or cooling, inside an isolated (beer) box.
The difference of the interior and the environmental temperature is constantly at about 6.5 ... 7.5°C.

This reverses the situation of the confidence in these measurements, so that the change on the module output voltage is now 10 time higher than possible changes caused by the 3458A.

This will give figures like the initial measurement on the LTZ #5.
During forced cooling, there occurs a shift of the output voltage, which is most probably due to thermal imbalances, causing air drafts and thermal e.m.f.s
As soon as the cooling / heating source is remove, all these inhomogeneity vanishes, and the output shows the regular T.C. change on the equilibrium path.

This is definitely no sign of hysteresis, as start and end point clearly converge into the same output value.

The first measurement with no R9 assembled, gives about -0.060ppm/K.
All modules have the LTZ1000 with its legs left long, and all module show an initial negative T.C.
I determined the individual T.C.s of all the precision resistors R1 .. R5, and matched the 5 sets to about equal resulting T.C. of -0.05ppm/K. I used the attenuation factors for each resistor from 5 different experiments, as given in the table.
The resulting T.C.s of the complete modules were higher than this caluculated T.C., up to -0.3ppm/K.
R9 will always add a positive T.C., so it is possible to trim the T.C.s to near zero, as shown for LTZ#1, using the box method.
The thermal cycle loops look more complicated, but in the end also deliver a nearly linear T.C. behavior.

[Dr. Frank]

In the last (bootstrap) step, I used an already trimmed module to determine the T.C.s of the other modules with better stability and confidence.
I used the DCV RATIO mode of the 3458A to measure the relative changes of each of two modules.
This eliminates the T.C. dependency of the 3458A.
Now, the reference module is kept constant to a few tenths of °C, and the DUT module is again changed by 10°C, which gives a confidence ratio of 100.
The measurement of all parameters is displayed in the first diagram. The resulting T.C. (box) for
LTZ#5 is about 0.020ppm/K, and U-shaped, by averaging away the reference noise, which is much bigger in magnitude, than the change due to the T.C..

It's really not practical to trim the T.C. further. Over a 18..28°C laboratory temperature range the absolute change is about +/-0.1ppm, completely sufficient for a travelling standard also.

The summary for all 5 modules is given in the table, which also includes their noise, measured against the 3458A at NPLC 100, over 1h, therefore including short- and midterm variations.
LTZ #1 is remarkably low-noise, whereas LTZ #4 has nearly double the noise.


In retrospect, the 3458A simplifies these delicate T.C. measurements, by its own low noise level.
Its own T.C. is not that much better, than other bench DMMs, so a stable environmental temperature is required with all these instruments, or an already trimmed reference module.


Next step will be, to monitor the timely drift of all the modules.

LTZ #5 already is running since March, and showing quite high -2ppm of drift since then.
This will decrease by next year, and hopefully I'll find several modules with lower drift, perhaps with a positive one also.

Frank

[TiN]

Yey. Unveil the tempco on us 

[dr.diesel]

Digesting, thanks for write-up Dr.Frank!

[TiN]

I would exercise caution with box method on tempco match, as I see not always linear behavior (granted that I run bit wider DUT exercise from +20C to +55C).

Did you do any selection for resistors to determine similar direction TCR, or you just base on LTZ attenuation factors for resistance stability? What are their spec, 3ppm/K?

Thank you for detailed write up, appreciate the effort.

I think we getting enough participants to do a big round robin for LTZ refs in 2018? I'll definitely have new design travel ref for it (10V).

[Dr. Frank]

I would exercise caution with box method on tempco match, as I see not always linear behavior (granted that I run bit wider DUT exercise from +20C to +55C).

Did you do any selection for resistors to determine similar direction TCR, or you just base on LTZ attenuation factors for resistance stability? What are their spec, 3ppm/K?

Thank you for detailed write up, appreciate the effort.

I think we getting enough participants to do a big round robin for LTZ refs in 2018? I'll definitely have new design travel ref for it (10V).

The LTZ #5 shows a slight U-shape behavior, whereas the other 4 references, #1 ... #4 are mostly linear, like the LTZ #1 I have shown.
The T.C. is buried so deeply in the zener noise, that it's pretty hard to make any reasonable T.C. determination below 0.02ppm/K.. therefore I find the box method quite appropriate.
It's also sufficient, if one assumes an absolute drift goal over 18..28°C.
In this case, the LTZ#5 would be within +/-0.1ppm, and this is also pretty darn close to the physical limits of analogue DCV measurements.

All commercial references claim a T.C. of minimum 0.04ppm/K, if I remember correctly, so 0.02ppm/k is ridiculously low.

The TCs of all PWW resistors are written down in the table, including the sign of their T.C.s

As I simply bought these resistors (econistors from G.R.) from the stock, w/o any further specification limits, or special matching, I had to use, whatever was delivered.

These were specified 3ppm/k typical, 5ppm/K maximum, similar to Vishays BMFs.
The 120 Ohm resistors showed some strange hysteretic behavior, as reported elsewhere in the forum, but all others were mostly linear and well below the maximum value.

As discussed quite often, the attenuation factors take care that the resistor stabilities are not so important, and it additionally turned out, that the intrinsic T.C. of the LTZ1000 seems to have superior influence.
In the end my design goals were completely fulfilled, that is an annual drift rate below 1ppm, and a low T.C. so that this does not carry no weight in uncertainty, compared to the timely drift, and compared to reasonable limits in DCV measurements in an amateur metrology lab.

Frankly speaking, I designed the modules in first instance to a good cost/performance relationship (about 100€ BOM cost), avoiding any over-engineering, like using Vishay hermetical VHPs (I could barely resist!), or voodoo-stuff w/o any proven benefit (e.g. these slits).

Yes, we may plan a chain comparison on a high metrological level.. I assume thsee modules also deliver sufficient uncertainty w/o battery backup.. and I might throw in one of my VHP202Z 10k resistors.

Let's see, what happens in the upcoming year, looking forward to the new S.I., kg, and Volt / Ohm.

Frank

[pitagoras]

Hello,
I'm entering the game with a standard +7V design.   
Only difference is separated heater/ref power (as A9 does).
Resistors are Rhopoint econistors for this first build. A second board will probably wear Edwins. Both will use ACH.
For R4 I could only get 2x25k so pcb has this provision for two parallel resistors.
While pcb is being made, I'll be probably building a tec box and some controller.
I'll be reporting news!

[mimmus78]

Did someone tested LTZ1000 at 20mA for long term consequences?
I suppose it needs also a higher temperature set point than usual ... maybe at lest 13K/1K resitor for the non A version.

[TiN]

Why would you want that? Drift rate will be horrible in long-term.

[mimmus78]

Well I'm just thinking at this message of few posts ago. If it does not accelerate burn in and he tell us that "You can safely bias the LTZ1000(A) with 20mA" I suppose it does not harm too much too. So I'm asking if someone tested at 20mA.

Hi forum members,

Bob Dobkin answered one of my follow-up questions:

[Me]:

About the Zener current of the LTZ1000-- I think you said it can take more than 5mA, but is it helpful to use more Zener current during the burn-in or would I end up with more long term drift in this case, and if more current is helpful, then how much is safe?

[Bob Dobkin]:

You can safely bias the LTZ1000(A) with 20mA. However, I do not think that will further accelerate the burn-in process.

I think three things take effect during the burn in. The first, is annealing of the aluminum metal. With this annealing, there is a resistance change in the aluminum. Secondly, the interface between the aluminum and the silicon changes. Also there are stress changes that anneal out.

None of these is particularly current sensitive.

====================================

It took him a while to answer this, but I'm glad he did... Wow, he must be super-busy!

Later,
Ken

[Kleinstein]

There is a slight chance that aging might be a little faster at a high current. One will definitely need a higher temperature set-point at a higher current as there will be more self heating. So not a good idea with the A version.

At higher current, the resistance on the AL traces on the chip, the pins and the board get more important. So chances are drift and TC will be worse. Higher power will also give more driving force for convection type thermal noise.

The main possible advantage with a higher current would be slightly lower noise, about 1/2 the noise voltage at 4 times the current would be a first guess. However the low frequency part can be different with less advantage.

[mimmus78]

So Kleinstein maybe the 5mA is just the balance of all the factors: self heating, TC and internal resistance.

[Kleinstein]

The main reason for using a higher current is the lower noise level, and maybe to a certain degree a lower zener resistance. Most other aspects like self heating, influence of lead resistance, internal resistance would prefer a lower current. Also the unheated TC of the reference is expected to get slightly better at lower current, though the effect is likely small - the TC zero crossing for the unheated chip would be at rather low currents (my guess would be in the 0.1-1 mA range, but also depends on the "70 K" resistors).

So it is a little like low current for good long term stability and higher current for lower noise.

[Dr. Frank]

As promised, a quick look at the performance of one of my new LTZ1000s, it's named LTZ #3.

This module is programmed for output of the LTZ1000 voltage, and also trimmed to 10V, label +Ref_buf and 10.00000V in the schematic.
The 20h stability of the LTZ direct output (7.131364V) is shown in the first diagram.


Main contribution for drift in the beginning comes from the 3458A, which needed 5h to stabilize completely. The LTZ #3 mostly stays within a window of +/- 0.1ppm, otherwise.

The 1h noise @ NPLC50 is about 0.045ppm rms.



These diagrams, and all which will follow, are prepared directly from the raw data, so there is not a single dip or glitch, thanks to the tuner box, and Andreas additional caps.

The T.C. of the LTZ1000 is about -0.025ppm.
In the window, I have marked the thermal equilibrium parts of the curve, again.



Then I've measured the T.C. of the divider, that are 4k over 10k BMF resistors from Alpha Electronics.
I just used the DCV RATIO function of the 3458A, measuring both voltage outputs.
I had some luck, the resistors seem to have a very close T.C., so the overall ratio T.C. is about +0.06ppm/K.




In the end, the 14h stability of the 10V output, vs. the 3458A is demonstrated. Latter was stable to +/- 0.1°C.
The module at first was sitting directly on top of the 3458A (~25°C), but I put a box underneath, so it cooled down a few degrees. So, there's again a non-equilibrium step visible, which quickly (20min.) returns to the initial value.
These frequent dips in temperature are probably caused by the thermostat in the basement, as we are heading winter, and the heating is already running, until about 11 p.m., when these dips vanish.



1h noise is  again about 0,043ppm.

So, these relatively cheap BMFs perform very well, as I already found out with my precision Hamon divider.

Frank