The LTFLU (aka SZA263) reference zener diode circuit

[Dr. Frank]

If i assume about -2 mV/K as the Ube contribution of that transistor, this amounts to -0.002 V / 7 V * 1E-6 = -286 ppm/K. Is that correct?

This terrible TC is there even after compensation by a positive TC of the zener of similar size.

No, that overall T.C. is indeed reduced to about +50ppm/K for the LTZ1000. In some special applications, it's reduced to < 5ppm/K by adding a resistor in series with the zener.
LTFLU can be easily reduced to  ~ 0ppm/K by selection of the collector current @ 3mA zener current.

So to get the reference down to TC < 0.1 ppm/K, one needs a factor 3000. If the TC compensation is adjusted to 1 %, the oven still needs an ambient temperature regulation of 1/30, that is  < 30 mK/K(ambient) or so.

That's proven to be no problem, the factor is indeed 500 only, and the regulation is even much better.

I know the LTZ1000 thermostat is very fast and oscillates at about 50 Hz if you have enough gain. So speed may be important. On the other hand i remember reading in this forum someone was using water to attach "thermal mass", with very interesting results. Arroyo TEC controllers appear to reach stability levels of some mK, too.

Really? never heard of this..especially no oscillation problems. That measure seems to be counter productive, as the normal circuit w/o these artifical measures works just fine, giving near zero overall T.C. by trimming.
Anyhow, I always try to add some thermal around the whole assembly, so that all affected components stay on the same temperature.

What is a practical solution for the LTFLU?

Regards, Dieter

?? LTFLU always requires an external oven, and its own T.C. is already near zero, so requirements on the oven are for sure relieved.
Main advantage of external oven is the stabilization of the step-up resistors.

[dietert1]

The 50 ppm/K of the LTZ1000 are the final result of a compensation that involves the sense transistor with it's  300 ppm/K, right? During all my professional life i learned that compensations to better than 1 % or so aren't very reliable. More like pieces of art instead of solid engineering. So in the LTZ1000 the compensation is about 50 / 300 = 16 % and it relies on an almost perfect thermostat instead. Will have to run the LTZ1000 without thermostat to have a look.
Of course nobody wants to have an oscillating thermostat and there were no "special measures". I just observed oscillations when building the circuit first without damping capacitors. I had noticed before that many schematics proposed in this forum and on the web were fairly chaotic concerning those bandwidth limiting parts. So i added/tuned the caps later observing the dynamic behaviour of the circuit.

For the LTFLU i remember now we have some old, non-working FEI rubidium oscillator. Maybe that one can donate its thermostat for the LTFLU reference.

Regards, Dieter

[Kleinstein]

In the LTZ and LTFLU the compensation in temperature drift is with transistor and zener diode directly on the same die. So there is no problem to get a good compensation and both parts are made to be very stable. So the "art" is already in the chip. The external part is not that critical:
Doubling the current through the transistor increases the voltage by some 20 mV and gives a change in the TC of some 66 µV/K or about 10 ppm/K of the total reference voltage. So to adjust the overall TC to 1 ppm/K the collector current only need to be within about 10% of the ideal value. The adjustment is quite predictable and the oven could be used to change the temperature, so that it does not take many iterations to get a good setting.

[dietert1]

While glueing a PT1000 temperature sensor to the bottom of a LTFLU i had the idea that the transistor inside the LTFLU should make a good temperature sensor, too. In order to measure its Ube one can split the 1K3 resistor that generates the zener current into a voltage divider 6:1, e.g. 1080 + 220 Ohm. After that the scheme should become somewhat similar to a LTZ1000, except the heater is external. Current variations in the transistor will probably not disturb temperature measurement, if the 10 V step up OpAmp is of good quality.
Need to work it out, though. So, next thing is to glue a heater transistor to the bottom of a LTFLU, for constant voltage heating (linear!).

Regards, Dieter

[branadic]

Here is what I'm currently working on. It's a Fluke style solution, thus a 20mm x 40mm ceramic board with LTFLU circuit and NTC on it. I use LT1006 for +12V power rail only and NOMCA16035001 resistor network for what I call R7 divider.
I'm designing a LTZ1047 like mainboard, lets call it LTFLU1047, with BMON (battery monitor and charging circuitry) and with a modified copy of F732B oven circuit, but use AD822 instead of TL062 for the oven, so again +12V rail only. The ceramic board will be thermally glued to a BPR10101 resistor, with the heater resistor perpendicular mounted to the mainboard. It will have some thermal shielding around it, that's for sure.

Single sided ceramic boards of that size with 12µm thick silver traces (silkscreen printing + burning --> thickfilm technology) can be bought commercially rather cheap. Lowest offer is $150 for tooling and $0,6 for each board.
Couldn't find any 100R heater resistor by now fitting the same size as my ceramic board, but will give it a try.

-branadic-

[jaromir]

The ceramic board will be thermally glued to a BPR10101 resistor, with the heater resistor perpendicular mounted to the mainboard. It will have some thermal shielding around it, that's for sure.

I thought of using aluminium substrate PCB for projects like this.
Did you consider this? For this particular project, what are advantages of ceramic PCB over aluminium one?

Lowest offer is $150 for tooling and $0,6 for each board.

Aluminium substrate PCB 40x20mm, 1,6mm thick, 6/6mil goes for ~50USD per 5 pieces (DHL shipping included) from random first asian vendor. Or ~60USD per 100 pieces.

[branadic]

Aluminium is not the best material because of it's thermal capacity and worse thermal conductivity compared to ceramic. Furthermore you have additional thermal resistances (aluminium --> glue --> foil) which worsens the oven functionality, since the temperature sensor is mounted at the LTFLU circuit.

-branadic-

[Kleinstein]

Another problem with an aluminum core PCB is the different thermal expasion. It is much higher than that of SMD resistors.
There is a chance the glues could show some creeping / drift.

Heat capacity should not be that much different to make a difference and also thermal conductivity is good with an aluminum core.

The DMM7510 shows that an oven is even possible with just FR4. After all with a tuned TC the oven is much less critical.

[branadic]

It is heavily different... please make the experiment, take a plate of aluminium and same size Al2O3 ceramic, heat both to say 350°C, wait half a minute and touch the aluminium with left hand and ceramic with right hand. Let me know which hand hurts.

I'm sure you will agree that DMM7510 is not a metrology grade meter and that it's more due to cost aficionados to make such a piss pour oven, nothing really serious.

-branadic-

[jaromir]

Just a few comments:

Aluminium has 5-10 times better thermal conductivity than Al2O3 - depending on exact composition and manufacturing process of both materials. Aluminium spreads heat from resistor better than ceramic substrate, creating lower thermal gradient on back side of board. Aluminium nitride ceramics is much better than Al2O3, berrylium oxide is even better than aluminium itself, but those are out of question, probably.

Typical thermal conductive dielectric for aluminium PCBs has one order of magnitude worse conductivity than Al2O3, accounting for its low thickness it may mean somehow worse conductivity from back side of board to front side of board. Question is how much it matters in well insulated system.
Thermal regulation via the PCB with its thermal mass and thermal resistance (which is suboptimal for both Al2O3 or Al substrates anyway) may need to be adjusted for somehow slower response - in comparison to Al2O3 board. That may be problematic if you need to power up reference from cold state to full specs fast, but this is not the usual use case of high-class references.
One may consider placing the temperature sensor to aluminium substrate directly; here would aluminium (with its better thermal conductivity compared to alumina) fit better and make thermal feedback loop tighter (that is always desirable for temperature regulators).

Thermal expansion of Al2O3 is usually stated 5-10 E-6/K, while aluminium itself 12-14 E-6/K. FR4 PCB has two orders of magnitude higher thermal expansion, so I expect resistors designed for PCB mounting being OK with both aluminium or AL2O3 substrates.


Bottom line - alumina ceramics is suitable material for your project and fortunately material science and cheap mass production is bringing new suitable material possibilities to the table. Decades since original Fluke references inception we have more options to choose from.

[Magnificent Bastard]

The new Fluke 732C uses a ceramic PCB, and they claim that this is the reason that the 732C no longer has seasonal variations (due to humidity) as earlier versions of the 732 have.

@branadic:
Can you say who is the vendor you found that has such good prices?

[Kleinstein]

The expansion of FR4 is a little complicated: it has a "normal" thermal expansion part that is not that bad (some 5-10 ppm/K), humidity effect and a delayed part from the epoxy part creeping relative to the glass-fiber part, especially at higher temperatures like 70-100 C. When coming from higher than some 150 C one can also have some structural relaxations.

Aluminum CTE is normally quite a bit higher, more like 22-23 ppm/K.

So the ceramic carrier parts are likely better of with ceramic or FR4 than an aluminum based board.
For the OPs in a SOIC8 case, the thermal expansion of the plastic and copper carrier is likely better matched to aluminum. 

Anyway for a reference oven the temperature is likely relatively constant and power consumption is constant. So thermal conductivity and heat capacity are not that important.
What hurts, when touching a hot piece of aluminum is the high thermal conductivity - lower conductivity can ease things quite a bit.

For the DMM7510, I don't think the oven temperature control is the real problem, the weak point is more the FR4 board and using numerical corrections for temperature effects (which has some acceptance problems). The other point is the Keithley typical extra noise for longer integration  (which likely is a software/statistics problem taken over from the old days when they had the brown cases).

[jaromir]

The new Fluke 732C uses a ceramic PCB, and they claim that this is the reason that the 732C no longer has seasonal variations (due to humidity) as earlier versions of the 732 have.

Humidity changes do influence board itself, as well as components on the board.
For example folks at Fluke are using custom made resistor network, rather than resistor array in plastic SOIC package.

[branadic]

@branadic:
Can you say who is the vendor you found that has such good prices?

www.xdpcba.com / www.xdxpcb.com

You can't compare single material only, but the complete board setup. So having an aluminium core board also includes the insulator between aluminium and copper and additional thermal resistances. Also keep in mind the different thickness, e.g. 1mm? for aluminum core board and 0.5mm for alumina. Humidity influence was also mentioned, almost no problem on ceramic, but a factor on FR4 and the insulator of aluminium core board. Same for CTE, FR4 with 13 - 17ppm/K in x and y (good match to copper with 16ppm/K), but 80ppm/K in z-direction. On aluminium core board you mix up materials with different CTEs.
So you can use whatever you want, I decided for ceramic.

-branadic-

[dietert1]

Recently i got four LTFLUs of the "LTFLU-1ACH 0625" version mentioned before. Basically they were tested OK.

In the standard application circuit, the first one exhibited a TC = - 40 ppm/K at 0,1 mA collector current of the sense transistor and TC = - 25 ppm/K at 0,04 mA. This means the negative TC of the transistor  Ube overcompensates. To tune the TC to zero, collector current would have to be further reduced. Apparently this LTFLU is out of specs.
Now i implemented the idea i mentioned above to use the LTFLU transistor as a temperature sensor. A second OpAmp outputs -200 mV/K.
While this was first meant for the implementation of an oven, i found that with a resistor R24/R25 i could tune TC to less than 1ppm/K without modifying the sense transistor collector current. This method works similar to the 400K resistor in the LTZ1000 circuit.
Already got some promising measurements under ambient temperature variations. Apparently nonlinear TC (including HP3456a TC) is so small that a +/- 0,5 °C oven is enough to keep the reference voltage within 0,1 ppm. Need to put the circuit into a aluminum box to use as oven.

Regards, Dieter

[Kleinstein]

If the TC is too negative (would need to little collector current for the transistor,  a little like with the LTZ) one could try reducing the zener current. This should also result in a more positive TC. The zener current is a second point to trim, though less predictable.

Another option would to use a diode in series to the resistor that sets the zener current. This is found in some circuits too.

[dietert1]

As i wrote, TC compensation was a lucky by-product until the oven is ready. I am not even sure i will need that once the oven works.
I think it may be interesting how the LTFLU transistor can be used as an on-chip temperature sensor. Haven't seen that before.

Regards, Dieter

[Kleinstein]

I don't think it would really take AZ-OPs in the control loop around the LTFLU. They tend to cause more trouble due to EMI issues going out from the OPs. RF signals originating from the OPs could effect the reference reading and the level of interference could be effected in a hard to predict way by thinks like attached cables.

At least for the OP for the voltage loop an normal OP (like OP07, LT1097,ADA4077) should be OK, as only something like 1/100 the OPs offset would appear at the output.

For the temperature reading this is less clear. Still 1 mK corresponds to about 2 µV at the sensor. So some drift may not be that critical.

However the temperature compensation part could be rather sensitive to the OP. R22/R20 gives a gain of some 100, while R7/(R24+R25) give an attenuation of some 700. So about 1/7 the OPs drift would appear at the output, which is not a good ratio, though probably just acceptable.
 
Another possible problem could be resistor drift from R20,R21,R22,R24+R25, especially if R20/R21 are not adjusted to keep the current through R22 and R24 low. The high values resistors tend to be more drifty than others. Chances are R20 and R21 also would need to be low drift, not as important as R7, but still with quite some effect.

I would try adjusting the zener current, if less transistor current is no longer an option. The Zener current should also effect the TC, though less predictable.

[dietert1]

Yes, the temperature measurement should be as precise as possible. If we want to get down from 40 ppm to 0,1 ppm, that's a stability requirement of 1/400. I think this is realistic with standard metal film resistors, if they are inside the oven. This is why i gave up on splitting R12 and ordered some foil resistors for that one. And the 100 gain is an example, maybe a -20 ppm/K output is as good. Then the MegOhm resistors will be 100K or so. On the OpAmps: I used what i had around, will certainly try others.

By the way, the schematic is missing some startup helper. Can be made with a zener, like in HP 3456A reference.

Regards, Dieter

[dietert1]

The statement i made above that a LTFLU with -40 ppm/K was out of specs was premature. Today i tested another one of those "LTFLU-1ACH 0625" and its TC is the same within 1 ppm/K in the same circuit.
Also i noticed the 8842A reference schematic in the beginning of this thread showing a 164K resistor instead of the 16K resistor in Dr. Franks schematic, which means running the LTFLU sense transistor at 10 uA instead of 85 uA. Maybe LTFLUs could be configured for different operation conditions. I mean we have seen those fuses in the die images presented above. Interesting enough, in many circuits above that resistor value is marked as configurable with no value given.

Regards, Dieter

[Dr. Frank]

The LTFLU is intended to have zero T.C.between 25 .. 45°C, i.e. operation at room temperature, inside 731B or 334A, 8842A, or similar, or inside a 45°C oven.

Both used cases obviously are different, concerning different setpoints, reflected in the 2nd requirement:

At 3mA zener current, this zero T.C. has to be obtained for collector currents between 20...200µA.

Therefore, if your LTFLUs do not achieve a zero T.C. within these limits, that might be the reason, why they showed up on the black market.

Frank

[dietert1]

How did you determine the current limits 20 .. 200 uA? Where did you get that from? Do we have to vary both temperature and current or can we find the zero TC point at any given temperature within the given range 20 .. 45 °C?

If the difference between LTFLU-1CH und LTFLU-1ACH is that one of them is meant for oven use and the other one for ambient temperature use, i would like to know that.

Regards, Dieter

[branadic]

Dieter,

I'm refering to this circuit:


By changing R13 you can set the temperature at which you have zero t.c. For my device I found it to be:
R13 = 25k343 --> ~30°C
R13 = 24k --> ~35°C
R13 = 22k --> 45°C

with zener current set to 3mA (R12=1k3) and the R7 divider set to 5k : ~11k, but without R8, R9, R10.

You should be able to reproduce similar results.

-branadic-

[Kleinstein]

The resistor to set the transistor current would be my primary choice to adjust the TC, as the effect should be relative predictable. There is a range of current that is feasible: some 10 µA may be the lower limit with maybe noise increasing to go up a little. However this would mainly be higher frequency noise, not so much the often more critical low frequency noise. An upper limit may be at some 200 µA as than the base current and current noise at the divider inputs get increasingly higher.

Another point to adjust the TC would be the zener current, with more zener current usually leading to a more positive TC. Here the effect could be unit dependent, so its not so clear how much effect a 50% higher current would have. There is no need to stay at 3 mA - if needed some 2, 5 mA or 7 mA may be acceptable.

A third point would be a possible diode in series to the resistor (especially R12 to set the zener current). A diode in series should also give a more positive TC, as the zener current would than go up with temperature. This could also effect the 2 nd order TC and thus the curvature, that could be important for a non heated reference.
Only if the 3 point's above don't give a suitable working point, I would consider adding the extra TC compensation.
The collector current is the most predictable way and thus likely the choice for fine adjustment.

There is no absolute need to adjust the TC zero to the exact oven temperature. It is more like adjusting the TC at the oven temperature to a low enough (e.g. < 3 ppm/K) value. So no need for large range temperature scans. Just 2 temperatures (e.g. +-5 K) near the paned temperature should be good enough.

I don't think there are official specs available to others than Fluke - so hard to tell if out of unknown specs.

[Dr. Frank]

https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/msg2519910/#msg2519910

branadic has demonstrated, that the zero T.C. point is only an extremum point over temperature, on a parabola, so for ambient temperature applications, it's necessary to trim the collector current.

All known oven application, beginning with the T.I. chip inside the 332/335 calibrators also have a trimmed collector resistor, or even have a label indicating the zero T.C. collector current.
Therefore, even oven applications need to be trimmed to zero T.C. for best performance.

I would also use the most simple examples of circuit, like inside the 5440 or 537x calibrators, I.e. w/o any additional diodes, or so.

Frank