Compensating the temperature coefficient of a LTZ1000 voltage reference?

[imisaac]

The LTZ1000 voltage reference has a temperature dependent drift of about +0.35 μV/^oC. This is a positive temperature coefficient.

The temperature coefficient of a BJT transistor, according to the datasheet of ON Semi 2N4401 (tempco plot attached), is negative and is adjustable based on its collector current.

Has anyone tried to combine the LTZ1000 voltage reference with a BJT to minimize the overall temperature coefficient further? Thanks in advance.

[iMo]

Imagine the Vbe tempco of the transistor would be -2000uV/K (an example).
How would you transform that into a "control loop" such you compensate the LTZ?
I mean what parts would you use? What should be the tempco of those parts?

[Andreas]

The LTZ1000 voltage reference has a temperature dependent drift of about +0.35 μV/^oC. This is a positive temperature coefficient.

Has anyone tried to combine the LTZ1000 voltage reference with a BJT to minimize the overall temperature coefficient further? Thanks in advance.

to a) this is dependant from device to device. So did you measure it for your specific reference?
to b) why should one use a external transistor to compensate the internal temperature coefficient? this will lead to time lag and other problems.

The 400 K resistor does a much better compensation because it uses the amplified internal temperature sensor.
You only have to adjust the 400K resistor to your specific T.C.
See the long LTZ1000 thread.

with best regards

Andreas

[imisaac]

Thanks imo and Andreas.

I think you are referring to the 400 kΩ temperature compensation resistor in the Typical Application section of the LTZ1000 datasheet? Yes, that could lead to the 0.05 ppm/^oC tempco as specified .

However, is it possible to use LTZ1000 in a typical negative feedback series voltage regulator circuit such that even the remaining 0.05 ppm/^oC tempco is nullified?

This surely depends if a transistor that is stable enough exists. I am not sure. For resistors, I think <0.05 ppm/^oC are indeed available.

[Kleinstein]

The "400 K" resistor from the LTZ1000 standard circuit can be adjusted to get the TC down to nearly zero over a small temperature range. A second other temperature compensation would also need individual adjustment - so one could as well adjust the "400 K" resistor.
The main problem with the TC compensation from the DS circuit is that the compensation is proportional to the heater current and thus more like a square root function, especially if the temperature set point is relatively low.

If one really needs very low TC my idea would be more an additional level of temperature regulation.

[imisaac]

Thanks, Kleinstein.

Yes, an additional oven for the LTZ1000 voltage reference is inevitable for the ultimate performance.

However, it is unclear if trimming the 400 kΩ primary resistor is the same effectiveness as trimming the collector resistor (in the case of the transistor based tempco compensation). How much change in "400k" is needed to change the tempco of LTZ1000 by say 1 ppm?

One potential advantage of the transistor-based tempco compensation, I hope, would be that the voltage reference output is relatively immune to the load variation. The extra current needed the compensate the load variation will come from the transistor, rather than from the LTZ1000 itself. This help stabilising the primary LTZ1000 circuit.

[Kleinstein]

Any load to the reference circuit would come from the OP (usually LT1013) anyway. Depending on the implementation there might be a transistor of JFET to buffer the OPs output anyway.

For the residual TC I see different possible sources:
1) the internal oven has non perfect regulation and possible thermal gradients on the chip, in the case . These errors should scale with the heater power.
2) the TC of the external resistors. There TC is attenuated by a factor of some 100 but not perfect.
 With reasonably quality (e.g. TC < 10 ppm/K)  resistors this should not be a problem.
   Similar the offset drift of the OP is attenuated - so this is usually not a problem.
3) Thermal EMF due to changing temperature gradients. Like the 1st. point heater power is expected to be the driving force.

In a limited temperature range and with not to low power, the heater current is a good approximation to the heater power. It is a little non linear (especially at low power), but any correction tends to be temperature sensitive. So the original circuit's 400 K resistor is already good in correcting effect proportional to the heater power (or current).

To make use if a very stable reference one kind of has to use a stable environment anyway. So I don't see the importance of the TC that high. The more limiting factor is long term drift and other external influences like air pressure, humidity (via the resistors).

[imisaac]

Thanks again, Kleinstein for the insight.

It seems that two of the three main temco contributions of the LTZ1000 reference circuit you mentioned comes from the internal heater related issues. Namely, the thermal gradient across the circuit and the temporal change of this thermal gradient.

In addition, the aging of the voltage reference, hence the annual drift in the accuracy, could also come from the prolonged heating of the circuit.
I could be wrong, but I think the humidity and the pressure effects are more pronounced for the resistive elements and to a lesser extent for the solid-state based zener reference.

Therefore, it may be possible that without using the heater function (i.e. turning off the internal oven) and opt for the external temperature stabilisation across the whole board (e.g. stabilize at slightly below 23 ^oC) may achieve a better overall stability?

For example, a typical room temperature change in a climatized room is about 1 ^oC. Combining this with a state-of-the-art LTZ1000 reference circuit tempco of 0.05 ppm/^oC gives 0.05 ppm stability. However, if there is a tiny oven that can stabilize the temperature to 0.05 ^oC, then a 1 ppm/^oC voltage reference can do the job just as well.

Does anyone know the tempco of LTZ1000 without the heater function?

Last but not least, regarding the long-term drift of resistive elements due to pressure, humidity...etc, it feels like the only remedy is to calibrate them as frequent as needed against primary standards.

[Kleinstein]

The LTZ1000 TC without the heater is at some +50 ppm/K (relatively little variations), so not really good. So to get a 0.05 ppm/K TC with the oven, the oven would need to reduce temperature variations by a factor of 1000. This is already quite demanding, both for the regulator and the stability of the temperature sensor.

The big advantage of the unit internal oven (heater + sensor) is that it can be fast and can thus also compensate external variations. In addition the small oven just for the reference also uses less power. If the whole circuit is stabilized it would need considerable more power, as the needed oven power kind of scales with the power of the circuit that is stabilized. So an outer oven would need something like 5 times the power.

Humidity is mainly a issue with the external resistors, if not very expensive hermetically sealed. Even than, there can be swelling of the board and thus an indirect effect due to stress. Due to the attenuation the effect is not that large. For the pressure there is a small direct effect on the metal case of the LTZ1000.

[imisaac]

It appears worthy to check how much the +50 ppm/^oC tempco of the raw LTZ1000 can be compensated to without using the internal oven. If +1 ppm/^oC tempco can be achieved, then a <0.05 ppm overall stability may still be possible when an external oven is added (assuming the power consumption is not an issue). The point here is to see limiting factors of this approach.

The pressure and humidity related drift of the voltage reference is also very interesting for the metrological point of view.....I may need to find a climate chamber to study them after the reference is built.

The issue with the swelling of the PCB, however, is a bit challenging. I don't have an idea how to apply a controlled amount of stress to the pins of LTZ1000 yet. In addition, I suspect the stress effect is somewhat anisotropic as well. That complicates things further.... 

Thanks for the tips, Kleinstein.

[Kleinstein]

The LTZ is not that sensitive to board stress. It is more resistors that can react to board stress, especially SMD form factor. The problem with humidity effects is that they tend to be slow and with different time constants. So some resistors may react to the average humidity of the last few days, the board itself may need weeks for the full area or a day for the edge and other resistors could need months.

There is a known way to reduce the intrinsic TC by adding a series resistance to the positive side of the zener. However as a downside this adds another rather critical resistor.

The pressure effect is mainly important if he pressure really changing, e.g. if one sends a precision reference to NIST in boulder at a rather high altitude (my guess is some 1600 m like Denver). The normal variations with the weather should not be a problem.

[imisaac]

Regarding the stress related issues of SMD components, I think the remedy is to use PCB substrates with low deformation such as ceramics or glasses.

Regarding the pressure related issues of the voltage reference, a pressure regulated thermostat could help.

Regarding the tempco compensation of a zener, there are indeed several techniques found. But I am not sure which one would work the best yet without further evaluation.....

[imisaac]

In an early eevblog post, several members indeed have attempted to trim the temperature coefficient of a zener to nearly zero. However, the main part used there was LTFLU.

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

In one message (https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/msg1285987/#msg1285987), one member (Dr. Frank) mentioned that:

...Temperature stability is achieved by this zener / Ube combination, as the zener has a positive, Ube a negative characteristic.
The LTFLU / SZA obviously is better in this aspect, as  the LTZ1000, due to a smaller zener voltage, which matches better with the T.C. of a pn structure, which in the end allows for a really precise zero T.C. trimming.
The LTZ1000 sucks in this aspect, as it always has about +50ppm/°C w/o oven, and is not really trimmable to zero T.C., any further....

Does anyone know the reason why the +50 ppm/°C tempco of LTZ1000 w/o oven can not be trimmed further? For a 7 V output, this tempco corresponds to +0.35 mV/°C. Is he implying that finding a component with the negative tempco (i.e. -0.35 mV/°C) would be difficult?

Thanks in advance.

[Gyro]

Does anyone know the reason why the +50 ppm/°C tempco of LTZ1000 w/o oven can not be trimmed further? For a 7 V output, this tempco corresponds to +0.35 mV/°C. Is he implying that finding a component with the negative tempco (i.e. -0.35 mV/°C) would be difficult?

Thanks in advance.

There's an app circuit for unstabilized temperature in the LTZ1000 datasheet if you want to have a play. It doesn't specify the degree of temperature compensation that you might achieve though...

[iMo]

I think fine trimming with the 70k in above schematics may work too..

[Kleinstein]

The 70 K resistor is only suitable for very fine trim. Reducing the current to half gives about some 20 mV lower voltage and some + 65 nV/K to the TC. So to get near zero TC one would need some 2-3 MOhms and thus so little current that noise and amplifier bias becomes a problem. In addition resistors in the > 1 M range tend to be less stable and leakage currents become a problem.
The more suitable way is the resistor at the positive side of the zener. Better not use a trimmer there, but find a suitable resistor (e.g. some 20 Ohms) and than maybe do the fine trim with the "70K". However this extra 20 ohms resistor would be rather sensitive ( e.g. like 1 ppm change in output voltage from some 50-100 ppm change in this resistor) - so not much saved from not needing the divider for the set temperature. The initial TC of the LTZ is just a little to high to make it worth it.

[imisaac]

Thanks Gyro, imo and Kleinstein for the circuit suggestion!

Looking at the left most "arm", the output voltage V_out = V_200Ω+V_zener+V_BE.

Looking at the middle arm, V_out = V_70Ω+V_CE.

The operational amplifier will make the two input pins equal to each other, i.e. V_CE=V_BE. Therefore, the following conservation relation should hold true always:
V_200Ω+V_zener = V_70Ω.

Adjusting the V_200Ω ever so slightly by changing its resistance value away from 200Ω will change the zener voltage in order to balance out V_70Ω, and vice versa.

Because the positive tempco of a zener depends on its operating voltage, the tuning of 200Ω is a means to tune the zener tempco.

Tuning the 70Ω collector resistor, however, may have two consequences. Firstly, it will change the collector current and hence the negative tempco of the BJT transistor. Secondly, it will change the positive tempco of the zener according the the conservation relation. The two effects somewhat cancels each other out. This is the reason that it is only used for fine tuning of the system tempco.

The 1N4148 diode at the op amp output is only for protection, and the 0.022 μF is for noise filtering.

Is the above interpretation of your circuit correct?

Thanks in advance.

[iMo]

I've found a model of a Zener with a tempco, and I've created the simulation where the tweaking with the serial Zener resistor works.
After some tweaking I've found the point where the 4 degC (25-29degC) makes 2.38uV Vref change.
You may iterate closer to zero if interested
With 200ohm you get 10mV, with 5ohm 1.8mV.
See below.

[imisaac]

Super! Thanks imo for the simulation.

It clearly shows the temperature dependence of the output voltage under different series resistance values.

[iMo]

And the sweep with the 70k finetuning resistor.
(Fixed the pictures type to .png above).

[Kleinstein]

Thanks Gyro, imo and Kleinstein for the circuit suggestion!

Looking at the left most "arm", the output voltage V_out = V_200Ω+V_zener+V_BE.

Looking at the middle arm, V_out = V_70Ω+V_CE.

The operational amplifier will make the two input pins equal to each other, i.e. V_CE=V_BE. Therefore, the following conservation relation should hold true always:
V_200Ω+V_zener = V_70Ω.

Adjusting the V_200Ω ever so slightly by changing its resistance value away from 200Ω will change the zener voltage in order to balance out V_70Ω, and vice versa.

Because the positive tempco of a zener depends on its operating voltage, the tuning of 200Ω is a means to tune the zener tempco.

Tuning the 70Ω collector resistor, however, may have two consequences. Firstly, it will change the collector current and hence the negative tempco of the BJT transistor. Secondly, it will change the positive tempco of the zener according the the conservation relation. The two effects somewhat cancels each other out. This is the reason that it is only used for fine tuning of the system tempco.

The 1N4148 diode at the op amp output is only for protection, and the 0.022 μF is for noise filtering.

Is the above interpretation of your circuit correct?

Thanks in advance.

The diode is for protection and maybe start up - it is not really needed.
The 22 nF cap is for frequency compensation of the control loop - the transistor adds extra gain to the loop and thus some extra compensation is needed, so the extra gain is gone when the OP's loop is closed. For the DC calculation one can ignore it.
One can see the resistor to set the zener current and the 200 ohms pot as a kind of "amplifier / divider": they see both the same current. So the extra resistor adds a fraction of V_BE on top of the zener.

Changing the 70 K resistor mainly changes V_BE, the effect on the Zener current is small and does not change the TC very much. Usually less current for a zener gives a more negative TC - so the 2 effects don't compensate but go in the same direction.  The problem is just that it takes a much lower current to get enough TC from V_BE.

[iMo]

Changes of the 70k resistor.

R      TC=0
65k    34C
70k    28C
75k    22C

[iMo]

Made for fun in simulation.
You may try in HW, sure
Would require an opamp and resistors with low TCs and the opamp capable working with 0.5V at its inputs (in that wiring). Also good thermal zener-transistor coupling.

What would be a long term stability of a cheapo Zener in glass? 
PS: a first help for the owners of an LTZ1000 with broken heater

[Kleinstein]

The cheap zeners in glass case could be stable (though likely not really good), but they are generally rather high noise, so it is difficult to measure. The noise is not only a little higher but more like a factor of 100.
There are 1N82x zeners which are not cheap, but they can be reasonably long time stable and the noise may be acceptable. However it may need some selection and luck to find a good one.

[iMo]

Sure, it is rather a simulation "experiment".

To the above tempco - based on the definition of the box method (for example the OP177 picture above):

from 16C to 34C is (Vmax-Vmin) / 18C = 8.9e-7V/C,  with Vref=6.99569V.

That is 0.13ppm/C, isn't it?