Ultra Precision Reference LTZ1000

[Andreas]

There was a classical document (1) by Fluke, where they monitored many different 732B over years.. and there was no sign of decrease of the ageing rates. The specifications are also quite realistic..
No, in contrary, for the 7001, even a linear drift prediction was specified, which might improve the uncertainty of this reference.

Only heavily drifting references like the one in the HP3458A may calm down, but only because they are operating far away from an equilibrium state.

Frank

(1) "Predictability of Solid State Zener References", David Deaver, Fluke Corp.

Hello Frank,

perhaps I have overlooked something. But I cannot find any statement for a linear drift in the document (1).

As far as I understand they are simply comparing all references which where calibrated at Fluke during a certain time against the predicted drift. And not monitoring several specific devices over time.
The only monitoring for drift over time is made for 1 single device where the drift is interpolated linear and nonlinear with the result that for the nonlinear drift model the limits are halved against the linear model for this particular reference.

http://www.vishaypg.com/doc?63003

I do not recommend a burn-In on Z201 resistors, neither does Vishay!

I will do it anyway just to be shure having done all what I could do.
Vishay offers PMO services to stabilize "foil resistors"

http://www.vishaypg.com/doc?49789

The only recommendation that I see is that PMO cannot be used for wire wound, and film resistors.
Of course it is mentioned especially on the hermetically VHP resistors.

And yes, you are right. The soldering shock after the pre-ageing will shift further the molded resistors with some hysteresis.
But since I am doing at least 1-2 gentle thermal cyclings (15-40 degrees centigrade environment) for measuring the tempco of the whole reference the introduced hysteresis should be mostly removed.

With best regards

Andreas

[robrenz]

@ Rufus
That is an interesting take on the situation that I had not considered. It makes sense that that solder blob has effectively changed the "alloy" therfore the Seebeck coeficient of the wire with the blob is different now.  You have inspired me to do some experiments myself.

@ quantumvolt
So we are both picking the same internet info to believe about thermocouple theory.

@ All interested
I welcome your comments on my current opinions: That a plating on a connector can have no appreciable affect on a connection because it is too thin and intimately connected to the connector base metal to have any significant thermal gradient that would produce a thermal emf.   Likewise that the mere fact of soldering a good mechanicaly coupled copper to copper joint with lead tin solder is not going to create a thermocouple with the charts 5µv/degC thermal emf.  So here are some tests that hopefuly illustrate my point.

Twisted tinned copper wire soldered with thermocouple next to it.


Loop of copper wire only and then the twisted wire that was soldered. In both cases the heat was on the junction/bend only to heat the junction to as uniform a temp as possible


Heat on one leg only of the twisted soldered copper to get a thermal gradient across the juntion.  Full scale is 1.78µV delta V for the 85degC rise was 1.424µV = 17nV/degC.  The up down ramp effect of the heated section is from moving closer to and further away from the junction itself on the one leg.


Then I made this very small area soldered junction from swaged tined copper wire to have minimal solder volume.


I strapped my thermocouple to my heat gun for a fixed position and set the temp to be stable at 115C. Then I tried the three positions heat on each side of the juntion then centered on the junction itself. 


Straight line section is ambient temperature then one side of junction heated then the other side of juntion heated then natural cooling to the center then repeat both sides of junction and then heat the juntion itself.  The heated juntion is basicaly zero output. The minor undulations are minor variatons in centering the heat on the junction. Full scale is 3.543µV and the delta V for 115degC rise was 1.5945µV = 14nV/degC.


Here is one leg of tinned copper sodered to one leg of flux core solder as a thermocouple. The delta for a 55degC rise is 294.8µV which equals 5.36µV/degC.  That is very close to the published 5µV/degC for copper vs lead tin solder.

[Rufus]

@ Rufus
That is an interesting take on the situation that I had not considered. It makes sense that that solder blob has effectively changed the "alloy" therfore the Seebeck coeficient of the wire with the blob is different now.  You have inspired me to do some experiments myself.

Good work. I don't find any of your results surprising but nice to see them all the same.

On changing 'alloy' I think it more like two conductors shorted together along their length. The emf you get must depend on the output impedance of the seebeck effect. I don't have a clue what that output impedance is, but, I guess it has to be inversely proportional to the conductor cross sectional area. Then the resistance of the conductors doing the shorting might be significant or perhaps not.

If you twisted equal gauge solder and copper together to make one side of a thermocouple do you think you would get about half the 'notional' solder/copper junction emf?

[quantumvolt]

Here is an easy-to-understand scholarly article that might describe the 'blob'-thing. http://www.msm.cam.ac.uk/utc/thermocouple/pages/Drift.html

"Drift (permanent change) occurs because of metallurgical changes of the thermoelements ...."

"... a change in Seebeck coefficient is a necessary condition to have drift, but it is not a sufficient condition for it: the change in Seebeck coefficient needs to occur in a region of temperature gradient ..."

"... changes at the junction of the thermocouple usually does not play any role in drift, as it can be assumed that the junction is at constant temperature ..."

I suggest you take a nice clean piece of wire and shape it as a U (no sharp bends). Heat it locally at the bottom of the U. Versions (apply a thin layer of solder on):

1. Half of the curved part of the U. Thermocouple effect.

2. The bottom part of the U (the entire curved part of the wire).  No thermocouple effect.

3. A small blob at the far end of one of the legs of the U. If the blob is not heated too much, little or no thermocouple effect.


BTW The statement "... a change in Seebeck coefficient is a necessary condition to have drift, but it is not a sufficient condition for it: the change in Seebeck coefficient needs to occur in a region of temperature gradient ..." explains why there not neccesarily is a  thermocouple effect on every contact point of different materials - the connection is at thermal equilibrium (and having no or small current) ...

[Dr. Frank]

Hello Frank,

perhaps I have overlooked something. But I cannot find any statement for a linear drift in the document (1).

As far as I understand they are simply comparing all references which where calibrated at Fluke during a certain time against the predicted drift. And not monitoring several specific devices over time.
The only monitoring for drift over time is made for 1 single device where the drift is interpolated linear and nonlinear with the result that for the nonlinear drift model the limits are halved against the linear model for this particular reference.



I will do it anyway just to be shure having done all what I could do.
Vishay offers PMO services to stabilize "foil resistors"

http://www.vishaypg.com/doc?49789

The only recommendation that I see is that PMO cannot be used for wire wound, and film resistors.
Of course it is mentioned especially on the hermetically VHP resistors.

And yes, you are right. The soldering shock after the pre-ageing will shift further the molded resistors with some hysteresis.
But since I am doing at least 1-2 gentle thermal cyclings (15-40 degrees centigrade environment) for measuring the tempco of the whole reference the introduced hysteresis should be mostly removed.

With best regards

Andreas

Hello Andreas,

I have never spoken about linear drift concerning the Fluke study of the 732A/B!

In this document, they analyse the drift prediction of up to 7 measurements on the same device.
The real behavior (either linear or other function over time) is not described in detail, that's right.

But one can see, that the specification of say 3ppm/yr. for the 732B, is not very conservative, as many devices (with SZA263) are on the edge of this spec, and will also not improve noteworthy after some years.

It's only a myth, that such devices get very stable, if one waits long enough.
That might happen on a few devices, but on many others, the ageing rate might increase again.

The study shows that, indirectly, because the drift prediction is quite uncertain.

For the 7000 reference only, Fluke specifies a typical drift rate of -0.7ppm/yr, and also an ucertainty of the prediction.
This parameter is specified as being linear, isn't it?


Vishay offers an PMO for the VHP101, and the details disclose, that this PMO is definitely no Burn-In!

Burn-In means: Storage for an extended period of time at constant high temperature.

PMO includes temperature cycling, in form of temperature shocks.

All metal foil resistors show hysteresis, therefore for real precision applications, a careful "degaussing-like" cycling is necessary.


I have shown my own T.C. measurements over a range of 10...35°C, and there you can hardly see any hysteretic effects.. that's only for greater temperature ranges.. >30°C in one direction, and therefore, 'gentle' cycling won't bring them back to zero.
You'll need trips to -23°C, up to +60°..80°C, whatever you have available in your home..

Anyhow, soldering ultra stable VHP resistors is always done with a heat transfer pincer, so that there is no heating of the resistive element.

Frank

[babysitter]

to protect Parts from the thermal solder shock, i use hemostat clamps (the latching clamps for blood vessels) on the wires quite often to dissipate heat away fm the Part.

[Andreas]

Part 2 LTZ1000A of Andreas

Mechanical

All is placed in a aluminium Euro-Card case EG2 of ProMa with dimension 168*103*56 mm.
http://www.reichelt.de/Proma-Gehaeuse/GEH-EG-2/3//index.html?ACTION=3&GROUPID=5199&ARTICLE=50424&SHOW=1&START=0&OFFSET=16&



Top side from left to right:

D-Sub connector for the main output with neighboured pins for the Zener output.
Next to the D-Sub is the auxiliary connector for a daughter board


Below the D-Sub you can see one of the auxliary 4 mm banana plug connectors (Hirschmann).
Both connectors are placed intentionally on the same height of the reference.
I cannot understand why on most instruments the positive (red) plug is always higher
(and hotter) than the negative (black) plug which is then cooler. Especially when having
a device with a power consumption that makes it necessary to use a fan.

In the middle the reference section with the parts around LTZ1000A. They are all placed into a TEKO 3710 metal shield with top cover.
http://www.reichelt.de/Teko-Stahlblech-Gehaeuse/TEKO-3710/3//index.html?ACTION=3&GROUPID=5202&ARTICLE=34042&SHOW=1&START=0&OFFSET=16&

The teko shield has the intention to equalize thermal gradients from outside to the reference section.
And calm down air currents.

All resistors (pre-aged before use) are in the upper half of the shield.
The LTZ1000 is thermally shielded with some foam.
Since the PCB is not plated through I could not settle the LTZ1000A directly on the PCB.
So the legs of the LTZ are surrounded by additional foam to keep air currents away.
This can be better seen on the closeup (IMG_4671.JPG) which I posted May 29, 2013 (page 12) in this thread.
The BF245C FET for the current source is above the LT1013.
The NTC for temperature sensing of the board temperature is near the LTZ
(between the FET and the LTZ).



The air gap between the top plate of the TEKO and the rest of the cirquit
is filled with additional polystyrene foam material (plate with 10mm thickness).

The right half of the aluminium case is filled with 12 AA cells in 2 battery holders
which are fed by a simple constant current charger during charging.

Between the TEKO inner shield and the battery holder some parts of the voltage regulator can be seen.



Bottom side:

On the bottom side you can see that the zener voltage is Kelvin sensed at the LTZ1000A
and going directly to the neighboured pins of the D-Sub connector only being connected
to the output filter capacitor.

There are many additional capacitors and some cirquit modifications which where not planned
from the beginning but which are already included in the cirquit diagram.
The base emitter capacitors at the LTZ are connected as close as possible to the LTZ1000 pins.

On the right side lower the low noise 14V power supply around the LT1763 is built.

Unfortunately the Aluminium case has not much place on the side below the pcb.
So there can be no metal plate for the LTZ-Section on the lower side.
There is only room for a 4 mm polystyrene foam sheet for thermal isolation.

By the way: for first firing of the cirquit I did not use the LTZ1000A but replaced the
zener section with a self built "refamp" consisting of a ordinary zener and a transistor.
Just to look whether the current regulation will do it's job properly, especially
when powering up/down.


Shielding and guarding:

ince I use a battery powered design where all components are within the aluminium case,
and no mains line can introduce any common mode noise to the cirquit,
the topic of shielding and guarding can be much simplified:

I have only a guard and dont need a outer shield (connected to eart ground)
All parts are "inguard".
A 2mm banana plug connected directly to the EG2 aluminium case can be used to connect
to the guard pin (or if not available to the negative pin) of the multimeter or calibrator.

Of course I have a large metal plate connected to earth ground on my desk which I use as outer shield
during my measurements. The guard of the LTZ is isolated against shield by some bumpers.

problems observed with the cirquit:

A short ciruit to the (unbuffered) output will set the heater setpoint to a large value.
This shifts the output voltage of the LTZ1000A. (Hysteresis probably due to the die attach).
Fortunately I could remove the hysteresis by simply power cycling the reference for several times.

When measuring immediately after charging Im observing some shift to the output voltage.
Probably this is due to thermal gradients going across the PCB due to the "hot" AA-cells.

todo list will follow ...

[Dr. Frank]

Part 2 LTZ1000A of Andreas

...

Nice, compact design, Andreas.. like it.

I will later show some LTZ1000 hysteresis measurements, I have encountered similar trips to extreme temperatures and hysteresis on one of my references also, but was able to reset it.

See Pickering patent.

Frank

[branadic]

For a future design, there are feed through capacitors available by the ham guys, that are affordable:

http://server3.gs-shop.de/200/cgi-bin/shop.dll?AnbieterID=9187&Seite=frameset.htm&PKEY=9D07

They are made of a zylindrical ceramic with inner and outer metallization. In the inner on a wire is soldered, the outer side can be directly soldered into a hole of your galvanically tinned metal case.

So you are able to fully encapsulate your bord with your reference, the buffer circuit and the current regulator. You are then also able to fill the case with what ever you want (hydrogene, nitrogene etc.) and hermetically seal it by soldering the lid to the rest of the case.

[babysitter]

Exactly what i talked about. You can solder in tubes for filling and sealing in advance. The feedthru caps are fine. I need to come out as being a ham

[branadic]

What about long-term stability of multiple paralleled references? Does it to average decrease? All I could read about is improved noise, sure by sqrt(N), but nothing more. Any experience about that?

We all know that it needs at least three references to find which one drifts in which direction and by what value, so wouldn't it be worth building a 3ref-ADC-DAC system regulating the voltage output after initial calibration?

[SeanB]

5 devices would be better for long term drift, as you can use it to check which ones have the most variation from the average.

[babysitter]

With heated refs having a bunch might help spreading the thermal aging. Also, averaging groups was described by pease, using 4x4 lm399.

[quarks]

Inspired by the Datron/Wavetek 4910, I thought 4 would be perfect, that is why I ordered 4 LTZ1000A.

[Andreas]

Inspired by the Datron/Wavetek 4910, I thought 4 would be perfect, that is why I ordered 4 LTZ1000A.

Yes 4 is the minimum for maintaining the "volt". At least 3 are on your site and you can find out a defective one.
The fourth you can send for calibration and import the calibrated volt to the group.

With best regards

Andreas

[branadic]

Inspired by the Datron/Wavetek 4910, I thought 4 would be perfect, that is why I ordered 4 LTZ1000A.

Are they paralleled or compared against each other?

[quarks]

Yes 4 is the minimum for maintaining the "volt". At least 3 are on your site and you can find out a defective one.
The fourth you can send for calibration and import the calibrated volt to the group.

@Andreas, for this the 4910 would not be the right choice (but Fluke 732 and 7000 would be see picture)

Are they paralleled or compared against each other?

@branadic, see 4910 Front Panel with 4 individual cells (10V each) plus average (see schematic)

[babysitter]

Sorry I was a bit handy-capped by using only my mobilephone which has no english dictionary, so entering those twitterlike short posts was hard work
My bit of thermal management was

* a "starlike" layout at the LTZ footprint, same-thickness, same-length traces on both sides of the PCB, if they are needed or not. Kept the "equalisation" star part free of other traces.

* cooper pour used as GUARD, will help spread out heat to reduce gradients.

* small bottle cap over the LTZ keeping regional airflow off

* Tinned metal sheet (heat-spreading) inside ABS outer enclosure (insulating).

Regarding DIY hermetically housing (close soldered box, feedthru capacitors):

Putting the feedthrus in close proximity will give only small temperature gradient. Symmetrical construction will help cancel out thermal EMFs. Putting all feedthrus on one side would allow using a socket or PCB.

BR
Hendrik

[Andreas]

* a "starlike" layout at the LTZ footprint, same-thickness, same-length traces on both sides of the PCB, if they are needed or not. Kept the "equalisation" star part free of other traces.

Regarding DIY hermetically housing (close soldered box, feedthru capacitors):

Putting the feedthrus in close proximity will give only small temperature gradient. Symmetrical construction will help cancel out thermal EMFs. Putting all feedthrus on one side would allow using a socket or PCB.

So if I understand it right: up to now you have no slots or cutouts within the PCB.

The feedthrus from "Geist" have copper leads as inner lead according to data sheet.
So on one side you will have lesser problems with thermoelectric effects. On the other side
I fear that the hermetical tightness is only a question of time.
How do you plan manage the barometric / temperature pressure relief with your close soldered box?

Has anyone a Idea how we can test the quality of a thermal design.
What is better: having long or short legs. Slotted or non slotted PCB.

With best regards

Andreas

[branadic]

Has anyone a Idea how we can test the quality of a thermal design.
What is better: having long or short legs. Slotted or non slotted PCB.

I think that is something you can only answer by measuring the heat spread with a thermal cam at an exemplar and as you expect I have access to one. Maybe one of those questions we can answer at one weekend?

BTW: Does anyone have the "ultimate" LM399 circuit that is worth building? Have build the "10V Buffered Reference" circuit with LT1001ACJ8 shown in the datasheet, but there are a few LM399 left, waiting for something useful.

[Andreas]

I think that is something you can only answer by measuring the heat spread with a thermal cam at an exemplar and as you expect I have access to one. Maybe one of those questions we can answer at one weekend?

BTW: Does anyone have the "ultimate" LM399 circuit that is worth building? Have build the "10V Buffered Reference" circuit with LT1001ACJ8 shown in the datasheet, but there are a few LM399 left, waiting for something useful.

Sounds good. This would be a good possibility to verify the design.

Mhm, why not use some LM399 to verify the thermal design. Principally they have a similar layout. But are much cheaper. And possibly the higher chip temperature gives an advantage in seeing temperature differences.

My ultimate idea (not already a complete design) would be building some kind of "calibrator" from 0-10V with a LM399 as basis. Either with a PWM divider (EPN cirquit idea) (cheap but noisy) or 2 interleaved 16 Bit DACs giving about 28 Bits resulting (more expensive but less noise) resolution. When having 2 independent calibrators of this sort and a 24 bit low noise ADC with some relays or MUX switches you could build a "self calibrating" system which calibrates out the non-linearities below around 1ppm. Similar to Franks 10V/1V divider but with a binary adjusting scheme.

For e.g. Set the first output to 10V the second to 5V compare the output voltages as 0V + 5V and 10V - 5V and re-adjust the 5V output to get the exact PWM/DAC values for half of the range. Store the values and determine + store the ADC INL error for half of the range. Then measure the difference of both outputs and adjust the first output to exact the 5V value of the 2nd output. Repeat the same for 5V and 2.5V and the 7.5V points. And so on.... until the linearity error of the DAC + ADC (e.g. 4-8 ppm) are well below 1ppm. I guess that about 32 - 256 measurement points should be sufficient as a basis.

By the way Frank: How often is it necessary to re-calibrate the 10V/1V divider to maintain the 0.1ppm divider accuracy. Which drift over time did you observe after calibration?

With best regards

Andreas

[tinhead]

BTW: Does anyone have the "ultimate" LM399 circuit that is worth building?

no, but i'm thinking about REF102CP with the QH40A crystal heater attached to it.

At 40° there is almost no drift at all, see attached picture (QH40A temp drift is between two blue lines)

[Dr. Frank]

By the way Frank: How often is it necessary to re-calibrate the 10V/1V divider to maintain the 0.1ppm divider accuracy. Which drift over time did you observe after calibration?

With best regards

Andreas

Andreas,

this kind of divider (720A, K.V.) is 0.1ppm of input, i.e. 1ppm of output @ 1V!

This is stable for 3 months or more, I guess.
I did not calibrate it more often.

I monitored the 10/7 output only, and that was accurate to ~ < 0.5ppm for 1 year, or so.

If you need more precise 10:1 division, have a look on the Hammon type divider, it's also much easier to switch and to calibrate!

Also have a look on the Fluke 5440A self calibration scheme, it has a nice 0.1ppm linear DAC ( 2x 1bit) and automatic  range calibration, uncertain to < 0.5ppm, or better.

Frank

[Andreas]

At 40° there is almost no drift at all, see attached picture (QH40A temp drift is between two blue lines)

I have measured the tempco of many references. But no one of them had ever seen the "idealized" curve out of the data sheet. In reality you will have to select many references to find one which has a maximum or minimum in the range of 40 degrees. Further the maximum/minimum will probably be shifted by hysteresis effects.

With best regards

Andreas

[babysitter]

I have not yet calculated the forces due to pressure on the DIY hermetic can, as I dont have decided for a size and know very little about the pressure behaviour of the gas yet. But my boss as a physicist will Nitrogen could be a good match as a filling gas, as most of air is it. Pressure exchange without mixing of internal gas and ambient air could be done with a flexible membrane, say, a hole in the close-soldered can, with a brass ring soldered in the inside and a piece of copper foil soldered on the ring. The cooper foil might be shaped to improve movement, cold-forming concentric rings like done with pressure sensor membranes comes to mind.

Man, I didn't want to go for gimmicks...

Lets see if our trainee is free for building the LM399 voltage source today...

EDIT: Have a look at those: http://www.servometer.com/products/metal-bellows/standard-bellows/
These might be a off-the-shelf practical membrane for pressure relief. Just solder it to the case with one end and solder it close on the other end.