Ultra Precision Reference LTZ1000

[Dr. Frank]

FEA simulations usually exaggerate displayed deformations so it can be visualized.

Sure, but ±0,5mm is at least 1mm deformation!

The graphic exaggeration is not what bothers me..
Such strain by temperature always creates very small dislocations (µm) on the PCB, but the forces this generates on the components may be extremely high nevertheless, but only on rigid bodies, i.e. SMD parts only.


Frank

[robrenz]

FEA simulations usually exaggerate displayed deformations so it can be visualized.

What about the strain gauge effects (resistance change) of the conductors as being the primary cause? Either circuit board traces, chip leads, or chip die level.

Which resistors do you mean?

I thought of such an effect in first instance also, but there is no such resistor element, perhaps SMD, as all precision resistors are leaded parts also.

Frank

What about the circuit board traces that are connecting the precision resistors to other elements of the circuit. Those traces are part of the circuit resistance and will change resistance with strain. The cross sectional area of a typical trace will be much greater than a strain gage but the effect will still be there. It might not be large enough to have any appreciable influence. I am just thinking out loud.  You guys are way past my knowledge on this but I find it interesting anyhow.

[Dr. Frank]

What about the circuit board traces that are connecting the precision resistors to other elements of the circuit. Those traces are part of the circuit resistance and will change resistance with strain. The cross sectional area of a typical trace will be much greater than a strain gage but the effect will still be there. It might not be large enough to have any appreciable influence. I am just thinking out loud.  You guys are way past my knowledge on this but I find it interesting anyhow.

Hey, that's perfectly okay!

Brainstorming is always a good thing, especially, when nobody else has an idea..

The trace resistance should not make a problem for the output, as the voltage is measured by a high impedance instrument.

The Zener feeding current might be affected, but this has to be calculated. (I feel, that this effect would be too small, but guess what?)

Frank

[babysitter]

Hi there,

there is a lot going on today!

Even as I don't expect a big piezoelectric - even less a triboelectric - effect on FR4 PCBs, I can offer some little experiment:

At work I have little PCBs with a "interdigital" or entwined finger-like electrodes on a single side. I can connect each electrode to a 34401A and give them a really harsh bend or hit, either in the direction of the fingers or perpendicular to it.

Ceramics might be something different, it just has this "smell of piezo". Also, Bob Pease wrote somewhere that he found isolated wires to have a piezoelectric effect, but only certain type of isolating material (piezo-material PVDF, Polyvinylidene fluoride, comes to mind immediately, have to look up the source.) A reason to keep the wires outside the test gear fixed during measurement ?

BR Hendrik

[robrenz]

The trace resistance should not make a problem for the output, as the voltage is measured by a high impedance instrument.
.....
The Zener feeding current might be affected, but this has to be calculated. (I feel, that this effect would be too small, but guess what?)

Frank

I know right now the thread is focused on the affects of the reference only, but what about the trace resistance variation from strain on the pcb traces connecting the precision resistors when looking at the entire circuit?

[robrenz]

I just happen to have a excellent micro ohm meter that is offset compensated and gives very stable single digit micro ohm readings. It will be easy for me to measure if there is an appreciable resistance change of a trace. Even if it is significant, wouldn't it be a simple matter to increase trace widths or copper thickness to reduce it to insignificant levels?  When I get a chance I will post some values on the delta R for various trace widths, lengths, and deflections.

[Robert763]

Hi,
This is my first post here. I suggest you research strain gauges. This is a known science (and art). Vishay are a big player in the market. I understand, but have no reference, that the Vishay foil resistors were a side benefit of their research into low TC strain gauge materials. This is another area where "RF like" PCB layout can help. If you have opposite arms of a bridge or divider as identical  tracks opposite each other on two sides of a PCB (like a stripline) the PCB bending induced stress changes cancel out.

Robert.
Semi volt nut, DMM's to 7.5 digits, couple of LTZ1000 references, LM399's, Diff. Voltmeters, JRL VDR-107 KVD.

[branadic]

I have also IR-videos from the heat-up phase with long and short leads, but no webspace to put them on. On the other hand I don't won't to register at youtube. Someone out there with some webspace for the videos?

[babysitter]

To prevent messing up this server, I suggest my Youtube-Account and some of my Webspace to embed it at.
What camera did you use ?

[branadic]

What camera did you use ?

It's a VarioCam hr:

http://www.infratec.de/de/thermografie/waermebildkameras/variocamr-hr-head-600-serie.html

[babysitter]

Just throwing Ideas at you right as you requested, not really thought thru

• Every LTZ1000 or -A device should be able to take care of itself. I would not touch it in any way for control of an outer oven. Just set its temp. Point sufficiently above its ambient, whatever that is. If your outer oven is good to 0.1K, use a setpoint that is 0.2K plus the waste heat from the LTZ. Its own temp. signal tells nothing about the outer oven besides that it is out of range, right ?
• The difference between- A and no-A is the isolated die attach, the higher temp setpoint for the -A only accounts for the waste heat of the rest of the circuit.
• Replace the the outer cooler by putting it in a freezer for the first experiments before you invest into the outer cooler. Put fans in the freezer to spread the heat evenly.
• Consider the use of a peltier element as heat pump + resistive heater instead of only resisive heater, even when extracting heat from the cooled outside sounds crazy at first.
• Take measures against dew
• Check if you can get a good Opamp in TO case or SMD with thermal pad, glue and wire it on top of the LTZ so the reference temp stabilizes the Opamp too ?
• Observe and measure. 

[Andreas]

But, what if I'm wrong?  Would it be possible to use the regular LTZ1000 (not the 'A' version), and just use the on-chip temperature sensing transistor to control the Peltier device and still maintain +/-0.001K die temperature, or would the control function be too unruly?  Before i run out and spend a small fortune experimenting on this, does anyone have thoughts on this?

Hello,

for the first: I´d never try to cool down something because I fear that the condensing humidity would change my output values. But perhaps you have the perfect rooms with humidity control.

Of course the thermal regulation will have to be much slower with a large thermal mass than only with the chip alone.
Otherwise the controller will oscillate.
For my thermal chamber I use a 2 stage concept to speed up the heat up times:
One sensor is mounted to the middle of the PCB with the references. (this would be the LTZ1000 sensor).
The other sensor is mounted directly to the aluminium heat spreader plate where the heater foil is mounted.

Since there is some self heating of the references the PCB is usually 4 degrees celsius warmer than the heater setpoint.
For a temperature of 50 degrees on the PCB the heater has to be kept around 46 degrees in steady state.

So I have 2 control loops.
A fast P (+D part for heat up) control loop for the heat spreader with a setpoint around 46 degrees
and a slow I part for the PCB with a setpoint of 50 degrees.

The I part is managed by correcting the setpoint (46 degrees) of the outer control loop.
For this the temperature on the pcb is measured. If the pcb temperature is only changing slowly (steady state) then the difference between actual temperature (e.g. 50.3 degrees) and setpoint (50 degrees) is subtracted (with a time constant factor e.g. 0.33/minute) from the setpoint. In this case in the first minute the setpoint is corrected to 49.9 degrees. (0.3 degrees times 0.33/minute).

I don´t know if its possible to reach 0.001 degrees stability with this method. A good thermal management (keeping air currents away from the chip, and stable environment temperature) will be necessary to get a stable regulation loop.

With best regards

Andreas

[babysitter]

(Advertising for sheet metal boxes and feedthrough capacitors again to realize small spaces with pretty good temp and humidity control, and also excellent for electromagnetic compatibility resons )

[Dr. Frank]

OK, I'm looking for some additional ideas on my Peltier-cooling idea.  I originally was thinking that there is no way that the Peltier device could maintain the die temperature of the LTZ1000 at +/-0.001-deg-K (in the same way that the on-chip heater can with proper external circuitry).  So, I thought that I would have to use an LTZ1000A, then chill that down to (say) 10 or 15 degrees-C lower than the chosen die temperature, then use the on-chip heater to heat the die back up to where you want it.  This is going to use a lot more power though-- more power to chill it down further (and to fight the on-chip heater), then add even more power for the on-chip heater.  This is not good for battery operation.

But, what if I'm wrong?  Would it be possible to use the regular LTZ1000 (not the 'A' versio n), and just use the on-chip temperature sensing transistor to control the Peltier device and still maintain +/-0.001K die temperature, or would the control function be too unruly?  Before i run out and spend a small fortune experimenting on this, does anyone have thoughts on this?

Forget all about that:
In an oven assembly, you always need high thermal gain, especially when you strive for 1mK stability.
That's always coupled with a small-as-possible thermal resistance  between heater/cooler and the sensor.

The thermal resistance of the LTZ1000 is about 80K/W, the A version even worse for that purpose, 400K/W, and both much too high to make a stable, well regulated oven.

If you simply use another sensor, outside the LTZ1000 case, then the case itself may be temperature regulated, but not the reference amplifier, again due to the high thermal resistance. The self heating will always create a temperature difference, and te temperature of the reference amplifier will be free to fluctuate strongly.

Only if the reference amplifier is also thermally coupled tightly to the sensor and heater, this might work.

Frank

PS: Best read are the articles of Richard Karlquist, one of the most experienced designers of the HP OCXO oven technology.
eg.: www.karlquist.com/oven.pdf

[cosmos]

Just a thought..
Dealing with fast external variations one might have a look at what some heatpumps for houses do (those are the ones I know).
They look at the house as balanced energy system, where the goal is to generate exactly the correct amount of energy on the inside to balance the energy leaking out.
This assumes the heat leakage out of the house is constant (should be true for a oven too).
Tuning it is a matter of selecting a gain variable matching the heat leakage and multiplying that with the temp difference inside to out to get a target energy level.
The leakage trough the walls change near instantly with temperature changes.
Stored energy inside a house creates a delay that messes with the regulation when it is only based on inside temperature.

If the mass inside the oven has limited ability to store heat I guess the above has little effect and that measuring temperature directly is the simplest way.
Using the temp difference to control the gain of the regulation loop might still improve the precision?

[babysitter]

Well, there are uncounable ways of optimizing, but you get into the "crazy" area quickly:

[li}Several shells of thermal mass for equalisation and thermal insulation, [/li]
[li]Spread Temp Sensors everywhere, log, look for patterns, use programmable heater control to play compensation games like "start decreasing inner temp when you realize the outer ambient is heating up even before the decrease appears outside[/li]
[/list]

Most thoughts in this field are immediately related to your mechanical construction.

[Dr. Frank]

I wonder if there's a hybrid approach-- something with the main temperature sensor close to the temperature driver (heater or chiller)-- and this would control the temperature, but then we mix-in the [conditioned] signal from the LTZ1000's on-chip temperature sensor, in order to steer the final die temperature closer to the ideal.  To me, this seems at least possible, because the internal burden heat from the Zener current of the LTZ1000 would be constant.  Just thinking out loud here...

Have a look on the tear downs of the 732B.. a hybrid is used with the heater and all the aligned circuitry on the ceramic PCB. 
I do not remember, if the LTFLU is used as a bare die, or housed... Bare die would be bad, because it would be prone to humidty/oxygen.

But this design comes near to what you propose..

The Fluke 5720 calibrator also has got a hybrid with two stacked LTFLU.

Anyhow, to my opinion, the stability can be improved, w/o such big effort.

Simply select the most stable external components( e.g. VHP202Z) and chose 45°C for the LTZ1000, and that's it.

Frank

[eurofox]

I wonder if there's a hybrid approach-- something with the main temperature sensor close to the temperature driver (heater or chiller)-- and this would control the temperature, but then we mix-in the [conditioned] signal from the LTZ1000's on-chip temperature sensor, in order to steer the final die temperature closer to the ideal.  To me, this seems at least possible, because the internal burden heat from the Zener current of the LTZ1000 would be constant.  Just thinking out loud here...

Have a look on the tear downs of the 732B.. a hybrid is used with the heater and all the aligned circuitry on the ceramic PCB. 
I do not remember, if the LTFLU is used as a bare die, or housed... Bare die would be bad, because it would be prone to humidty/oxygen.

But this design comes near to what you propose..

The Fluke 5720 calibrator also has got a hybrid with two stacked LTFLU.

Anyhow, to my opinion, the stability can be improved, w/o such big effort.

Simply select the most stable external components( e.g. VHP202Z) and chose 45°C for the LTZ1000, and that's it.

Frank

If you use 5 LTZ1000 with all perfect conditions specified in the different posts and make the average it will be even better 

eurofox

[eurofox]

I wonder if there's a hybrid approach-- something with the main temperature sensor close to the temperature driver (heater or chiller)-- and this would control the temperature, but then we mix-in the [conditioned] signal from the LTZ1000's on-chip temperature sensor, in order to steer the final die temperature closer to the ideal.  To me, this seems at least possible, because the internal burden heat from the Zener current of the LTZ1000 would be constant.  Just thinking out loud here...

Have a look on the tear downs of the 732B.. a hybrid is used with the heater and all the aligned circuitry on the ceramic PCB. 
I do not remember, if the LTFLU is used as a bare die, or housed... Bare die would be bad, because it would be prone to humidty/oxygen.

But this design comes near to what you propose..

The Fluke 5720 calibrator also has got a hybrid with two stacked LTFLU.

Anyhow, to my opinion, the stability can be improved, w/o such big effort.

Simply select the most stable external components( e.g. VHP202Z) and chose 45°C for the LTZ1000, and that's it.

Frank

If you use 5 LTZ1000 with all perfect conditions specified in the different posts and make the average it will be even better 

eurofox

I didn't mean a "hybrid integrated circuit"-- I meant a hybrid system approach to temperature control (as opposed to an oven inside of a refrigerator).

As far as averaging multiple LTZ1000's-- yes, this will reduce noise by the square-root of the number of references you average (and more than 4 would be cost prohibitive).  The problem with long term drift is that the LTZ1000's all seem to drift in one direction (IIRC, in the "down" direction)-- it's just that some drift more than others.  A better idea would be to select the best references that drift the least, and throw the rest away (or sell them on eBay or something).

What I want to do is get better long term stability than the standard circuit can provide.  Let's say the best you can get with the standard LTZ1000 circuit is 0.3ppm/year-- well I want better than that.  0.1ppm/year would even be better-- and I want to be able to recreate this multiple times without having to make 1000's of references, and throw away all but the few that meet my spec.  That's why I'm looking into cooling the reference as opposed to heating it (to maintain a stable die temperature).

Cooling should be easy with a peltier element and add some control with PID to keep the temperature accurate to 0,1 °C
I suppose aging of the LTZ1000 is worst with high temperature, so by keeping it "cold" it should drift less.
Why not averaging LTZ1000 with another one from another supplier with similar tolerance.

Just another idea but I have no experience in this field just to input maybe new ideas, use precision frequency to voltage controller, based on a rubidium standard you can have an extreme stable frequency, of course your converter will drift as well with aging and temperature.

eurofox

[Andreas]

As announced by Andreas we wanted to find out what is best: "Using a solid pcb or spend some slots around the heat controlled voltage reference?" and furthermore: "Keep the leads as long or as short as possible?".

So let's start the discussion

Ok, I have finally evaluated the measurements = .csv tables of Branadic from the LM399 thermograpic measurement.

Long leads without slot:   around 5 degres stray within pad  + 1,7 degrees from pad to pad
Long leads with slots:      around 5-6 degres stray within pad + 2,3 degrees from pad to pad
Short leads without slot: around 9-11 degres stray within pad + 3,1 degrees from pad to pad
short leads with slots:     around 12-15 degres stray within pad + 6,6 degrees from pad to pad

So from the measurement the conclusion is that with long leads the stray (=thermoelectric voltage) within pad is lower than with short legs. And the thermoelectric voltage between different pads is lower without slot.
If there should be mechanical issues the slots should be at least around 15 mm from the reference to give the pads the possibility to equalize the temperature.

2 simple conclusions:

- short leads will increase the temperature of the PCB, long leads will give more thermal isolation between component and PCB => long leads preferred
- slots in the PCB give higher thermal isolation between solder joint area and outer PCB area, thereby reducing cooling of the solder joints through the PCB, and in turn increasing the local heating of the solder joints. Higher solder joint temperatures might cause temperature differences more easily and  will increase thermo voltages.

I'm now very confident that 45°C and a PCB without slots is the most stable solution.

So the conclusions of Frank are now quantified.

Before the measurements I planned to make slots near the reference and short leads. Now I will go for long leads and at least keeping the slots far away from the reference. (e.g. at the edges of my inner metal shield).

@branadic
When regarding the videos (the flimmering of the heat) I had the idea that the result should be also be visible by noise measurements. So its a pity that we did not measure the noise voltages of the references in parallel to the thermographic measurements.

With best regards

Andreas

============================== evaluation in detail ==============================

Long leads without slot
=======================

pin upper left:

min   27,64
max   32,81
avg   29,59356083
stddev   1,144305683
max-min   5,17

pin upper right:

min   28,02
max   33,15
avg   30,25453333
stddev   1,000260225
max-min   5,13

pin lower left:

min   27,8
max   33,18
avg   29,33930796
stddev   1,031510593
max-min   5,38

pin lower right:

min   28,87
max   33,48
avg   31,08652597
stddev   0,855853162
max-min   4,61


all in all: around 5 degres stray within pad
1,7 degrees from pad to pad


Long leads with slots
=====================

pin upper left:

min   30,1
max   35,46
avg   32,78116883
stddev   1,173468631
max-min   5,36

pin upper right:

min   28,44
max   33,54
avg   30,48031128
stddev   1,148254067
max-min   5,1


pin lower left:

min   29,24
max   34,46
avg   31,80304
stddev   1,342351664
max-min   5,22

pin lower right:

min   29,3
max   35,7
avg   31,30611111
stddev   1,183186142
max-min   6,4

all in all: around 5-6 degres stray within pad
2,3 degrees from pad to pad


short leads without slot
========================

pin upper left:

min   36,649
max   47,527
avg   42,58773745
stddev   3,146954606
max-min   10,878

pin upper right:

min   38,599
max   49,453
avg   43,66859498
stddev   2,401555908
max-min   10,854

pin lower left:

min   37,141
max   45,726
avg   41,45009881
stddev   2,115676062
max-min   8,585

pin lower right:

min   38,552
max   48,064
avg   44,64691221
stddev   2,080904881
max-min   9,512

all in all: around 9-11 degres stray within pad
3.1 degrees from pad to pad

short leads with slots
======================

pin upper left:

min   41,186
max   55,277
avg   46,75962041
stddev   3,418151728
max-min   14,091

pin upper right:

min   39,226
max   50,78
avg   44,88190295
stddev   3,051374117
max-min   11,554


pin lower left:

min   42,81
max   58,327
avg   50,25318644
stddev   3,742872955
max-min   15,517


pin lower right:

min   40,915
max   56,608
avg   48,35819758
stddev   4,061728375
max-min   15,693

all in all: around 12-15 degres stray within pad
6,6 degrees from pad to pad

[Mickle T.]

I think it will be less stable.

[Galaxyrise]

The problem with long term drift is that the LTZ1000's all seem to drift in one direction (IIRC, in the "down" direction)-- it's just that some drift more than others.

Looks like the LM399 drifts upwards. If the two references were summed with the LTZ1000 given a 4:1 weight, perhaps the net stability would be better than the LTZ1000 by itself?  And it would be cheaper than a second LTZ1000, even with some good resistors at the summing point

[babysitter]

Off to Weinheim. See some of you tomorrow !

[BiOzZ]

i have used these before but i had to buy 20 of them from digikey and that was a real bitch to write off but i can see they have been all but one used up!

[quantumvolt]

I wonder if there's a hybrid approach-- something with the main temperature sensor close to the temperature driver (heater or chiller)-- and this would control the temperature, but then we mix-in the [conditioned] signal from the LTZ1000's on-chip temperature sensor, in order to steer the final die temperature closer to the ideal.  To me, this seems at least possible, because the internal burden heat from the Zener current of the LTZ1000 would be constant.  Just thinking out loud here...

Have a look on the tear downs of the 732B.. a hybrid is used with the heater and all the aligned circuitry on the ceramic PCB. 
I do not remember, if the LTFLU is used as a bare die, or housed... Bare die would be bad, because it would be prone to humidty/oxygen.

But this design comes near to what you propose..

The Fluke 5720 calibrator also has got a hybrid with two stacked LTFLU.

Anyhow, to my opinion, the stability can be improved, w/o such big effort.

Simply select the most stable external components( e.g. VHP202Z) and chose 45°C for the LTZ1000, and that's it.

Frank

Do you have have schematic for the stacked LTFLUs. I looked in http://assets.fluke.com/manuals/5720a___smeng0200.pdf (576 pages!), but the ref board on p. 522 seems not to have a schematic.