Compensating the temperature coefficient of a LTZ1000 voltage reference?

[imisaac]

Thanks Kleinstein for the prompt information.

How are the resistor values selected for the "control loop"? It seems that the output voltage only maintains at 6.9 V if R5<<R6.

[iMo]

The only parts which contribute to the Vref's tempco shape are the R1, R2 and R3 resistors with "tc", the zener diode, transistor and the opamp (when not "ideal").

The components in the opamp's loop do not contribute (the output transistor and the 10k and 25ohm) as the opamp does compensate the changes. Basically "anything" you put between the opamp's output and Vref node is compensated off.

The output transistor buffers the 5mA zener current, as I think 5mA is too much for a precision opamp to stay cool.

The TRS1 and TRS2 are the intrinsic params of the zener diode and cannot be changed from outside, afaik. The TRS values used in my simulation are the examples from 3roomlab.

[imisaac]

Super, imo.

I added one extra op amp at the buffer stage to scale the output up to ~10 V, a convenient voltage for a DAC.

[imisaac]

Does anyone have experience in which operational amplifier is the best for the stable voltage reference circuit?

Three possibilities are listed below to begin with....

OpAmp	|VOSTC(nV/oC)	|IOS(fA)	|VOS(μV)	|Noise(nV/√Hz)
ADA4530-1	|130	|1	|40	|14
LT1013	|300	|150	|150	|22
LT1006	|200	|120	|20	|23

Thanks in advance.

[Kleinstein]

The L1013 is well good enough for the job. So no real need for experiments. As the transistor inside the LTZ1000 already provide a gain of about 200, the demands one the OP are not that high. So neither offset, drift or noise are that important. So in theory a LM358 might be good enough - but possibly problems with phase reversal. But why experiment on a $2 OP if a $50 reference is at risk ?

The LT1006 is essentially the single version to the LT1013.
Just be careful with the odd pin-out of the SO8 version of the LT1013.
If there would be no LT1013 available for some reason my next best guess would be TLE2022.
Lots of discussion on the OPs is in the long LTZ1000 thread.

[imisaac]

Thanks, Kleinstein.

If there is an offset voltage at the input sides of the OP, they will show up at the output even if the gain is small, I think. The OP at the 10 V boosting stage is the one that will affect the output more strongly. In particular, its tempco (plus the tempco of the resistor divider) will directly determine the tempco of the 10V voltage reference. After all, the resistor trimming tricks for zero tempco discussed so far is done prior to the boosting stage.

It could be true that the requirements (noise, tempco, aging, offset...) for the OP on the zener side is not so critical. However, the same conclusion may not be true for the boosting OP, right?

ADA4530-1 appears very interesting since it outperforms LT1013/LT1006 in the tempco by ~3 times.....

[Kleinstein]

The OP in the LTZ1000 circuit itself is less critical as the extra gain reduces the effect of errors (offset, drift and noise) by about a factor of 200, at least for the low frequency (some 1 KHz) part. Similar the resistors in the circuit have a reduced effect by something like a factor of 100 or more. So 1 ppm of change in the output would correspond to some 100 ppm (the exact numbe depends on the resistor) change in resistor vales.

A possible boost stage is much more critical. The offset is directly added to the 7 V reference level and the resistors are responsible for some 1/3. So 3 ppm resistor ratio change to give an 1 ppm output effect.
The ADA4530 is no the right choice here - this is a super low bias OP, but with high noise. LT1013 is also not that good here.
The more practical candidates would be LTC2057, ADA4077, OP177 - so more OPs made for a lower impedance.

[imisaac]

Thanks Kleinstein. I re-tabulated the (boosting stage) op-amp selection table based on your suggested criteria. In particular, all input bias parameters are removed from the table. All noise related parameters are inserted.

Here is an update.

OpAmp	|VOSTC(nV/oC)	|0.1~10 Hz Vpp(nV)	|Inoise(fA/√Hz)	|Vnoise(nV/√Hz)
ADA4625-1	|200	|150	|4.5	|3.3
ADA4077-1	|250	|250	|200	|7
LTC2057	|15	|200	|170	|11
OP177	|100	|?	|?	|10

If the tempco is of the most importance, then LTC2057 seems to be the best (as the boosting op amp).
If the input voltage noise is of the most importance, then ADA4625-1 seems to be the best (as the boosting op amp).

Is the above understanding of your suggestion correct?

[imisaac]

Just a brief clarification.

To ensure what the "boosting stage" meant in the previous message is clear, I drew a hypothetical boosting stage consisting of two op amps. Of course, the tempco of this boosting stage should be comparable to the tempco of the LTZ1000 voltage reference (the circuit that produces 6.9 V).

It seems that a low bias op amp may produce a more accurate 10 V output in comparison to low noise op amps because of the small offsets.....

Is it still true that the noise consideration outweigh the input bias when choosing both op amps for this circuit?

[Kleinstein]

The circuit with the extra boosting stage is not good: it adds the errors of the 2nd OP. So normally the divider should go from the output. So the current boosting buffer would be inside the loop. In this case there is no real need to use an OP, even just a transistor is enough, as the divider is already plenty of load / biasing current.

The divider is using rather low resistor values. This would produce quite some heat and related problems. I would more consider resistor values about 10 times higher.

Unless noise is really high, the noise is not that critical - it should just be considerably lower than the reference noise. If at all it is the low frequency noise that can really be a problem.
Bias of the OP is also no that important - the bias is usually relatively constant and only the variable part would really be a problem. The really tricky parts are more like offset drift with time and possible offset hysteresis - these parameters are not often specified.
Especially when using an AZ OP like the LTC2057 it would be a good idea to have some filtering (RC like 5 K and 1 µF) before the amplifier, to keep chopper spikes away from the actual reference circuit. The high frequency components can cause hard to predict effects, a little like EMI effects. For a BJT based OP the filter could also provide source impedance matching the divider.

[iMo]

For example

[imisaac]

Thanks, Kleinstein and imo for the wonderful insights as always been.

1. The first half of the "boosting stage" is needed to scale the 6.9 V LTZ reference circuit voltage up to 10 V. In the example circuits that imo designed, U1 serves this purpose as well. Therefore, this voltage boost step appears to be inevitable if one needs a voltage that is different from the default 6.9 V.

2. When a transistor is used to supply the current to the load, the voltage noise of the +15V power supply directly connected to the collector will be converted to a voltage noise at the emitter and hence producing a voltage output noise. The capacitor C1 (in the imo's circuit) may be able to alleviate the problem by shorting the high frequency noise to the ground. However, is this noise suppression scheme as good as having a high PSRR when an op amp is used?

[Kleinstein]

The transistor as an emitter follower would be inside the control loop. So the OP would compensate for any change coming from the collector side.  Anyway there is not much coupling through the early effect, as V_BE for a given current does not depend much on V_CE. So not sure an OP would be better. The main advantage of using an OP as current boost stage would be that most OPs have a current limited output - so they could survive a short.

[imisaac]

If I understand you correctly, you are saying that the 10 V voltage output on the emitter side is inside the control loop. Therefore, any deviation from the 6.9 V will be compensated by the op amp that is comparing the 6.9 V and the voltage divided 10 V signal.

This compensation is also fast enough to act as a low pass filter to get rid of the high frequency noises coming from the power supply.

In this case, the bandwidth of this op amp (the one inside the control loop and scaling 6.9 V to 10 V) becomes an important factor?





 

[Kleinstein]

The higher frequency part is usually not that critical, as it can be filtered. The difficult part is the low frequency part. There should not be much higher frequency noise in the supply, so there is not much need for a fast OP in a DC reference.

The really critical part are the resistors to do the 7 to 10 V amplification. These are much more (e.g. 30 times) critical than those at the LTZ1000. Today the amplifier part is more on the easy side.

[imisaac]

Thanks, Kleinstein. You are absolutely right that the voltage divider for the 7 V to 10 V amplification is definitely critical.

For that divider network, I am thinking about using the DSMZ Z-foil voltage divider (R1/R2 ratio of 2) from Vishay. The tempco of the ratio (TCR tracking) for this resistor network is < 0.1 ppm/^oC (tempco spec attached).

The output will not be exactly 10 V, however, with this off-the-shelf ratio.

Does anyone have an experience with this resistor network or perhaps have a better resistor option in terms of stability (tempco and aging)?

Thanks in advance.

[TiN]

I think all this should be in original LTZ1000 thread. Feeling like a dejavu already, as many of such topic discovered in that thread already .

I've trimmed multiple LTZ refs with "400K" resistor to tempco <0.05 ppm/K. Issue to get better TC often not with LTZ circuit itself but in thermal management and layout of the parts. Also detecting such low tempco especially on limited Trange is difficult and not cheap. I have to use modified multiple 3458As for that to be reliable (meters have zero TC for 20-30c range, down in own noise). No amount of simulation will help to find "receipe" that works in real PCB. I can only suggest to build few boards and start experimenting with real hardware.

imisaac, also don't believe marketing posters about 0.0 ppm/K resistors, there is no such thing unless you buy hundred resistors and cherry pick one golden sample out of them. I've tested hundreds of various resistors, from $2 USD a piece to $240 USD a piece and still didn't find magical 0.0 ppm/K To be fair - not that you need that for LTZ reference anyway, even 2ppm/K will be fine and performance of the reference is limited by design/LTZ chip itself.

[imisaac]

Thanks, TiN for the references! The -0.03 ppm/^oC tempco of your 10 V reference is super impressive!

1. In your circuit (screenshot attached), you used a ADA4522 chopper amplifier. Its Vos tempco is 4 nV/^oC, seeming like a good choice. However, you used two of them and then added them up at the output. Was the idea to average down the chopper noise?

2. You used VHD200 as your (7V->10V) voltage divider network. It has got the same TCR tracking (0.1 ppm/^oC) as the DSMZ divider network. Did you choose VHD200 because it was hermetically sealed and hence had a smaller ageing?

3. A minor comment about the well established LTZ1000 thread. That thread has reached 105 pages, with perhaps more than 2000 posts. Many of the discussion topics are beyond my current electronics level, I think. It is a bit difficult for me to dig the specific answer out of it. In addition, the tempco compensation solution that imo/Gyro/Kleinstein/Andreas/3roomlab suggested in this thread appears a bit different from most existing LTZ1000 reference circuits in the sense that the internal heater on the LTZ1000 is not utilized to reduce the tempco, but with a special resistor trimming approach. It is interesting to see how this topology performs in comparison to the established LTZ1000 reference circuit.

4. Do you have an estimate on the annual drift of your 10V FX reference?

Thanks.
 

[TiN]

VHD network used only in prototype. Final units including the one I linked use unobtanium laser-trimmed hermetic Fluke film resistor (used in Fluke 8846 DMM).

Output stage opamps used for both force and return drive bipolar ability. This way you can abuse output with caps, shorts, reasonable reverse backdrive and there is no major risk to kill expensive LTZ.

As of annual stability, I will know answer to that question next month, currently in progress to obtain SI Volt from Josephson reference via transfer by Fluke 732.

Existing LTZ thread have lots of data, so time spent to read it thru can answer some of your questions here already. Either way LTZ designs aren't what many can call "beginner" level

[Andreas]

Does anyone have an experience with this resistor network or perhaps have a better resistor option in terms of stability (tempco and aging)?

Beginning from here 2 T.C. measurements: (sorry no ageing data)

https://www.eevblog.com/forum/metrology/t-c-measurements-on-precision-resistors/msg521240/#msg521240

like all plastic/epoxy packages (absorbing humidity and swelling with time constants of some days) the T.C. is very dependant on humidity history.
0.1 ppm (typical ?) tracking could not be reproduced on my 2 samples.

with best regards

Andreas

[Gyro]

3. A minor comment about the well established LTZ1000 thread. That thread has reached 105 pages, with perhaps more than 2000 posts. Many of the discussion topics are beyond my current electronics level, I think. It is a bit difficult for me to dig the specific answer out of it. In addition, the tempco compensation solution that imo/Gyro/Kleinstein/Andreas suggested in this thread appears a bit different from most existing LTZ1000 reference circuits in the sense that the internal heater on the LTZ1000 is not utilized to reduce the tempco, but with a special resistor trimming approach. It is interesting to see how this topology performs in comparison to the established LTZ1000 reference circuit.

Agreed, the other thread is mainly heated LTZ.

This thread definitely needs to be moved from the Beginners section to Metrology though.

[iMo]

..Bias of the OP is also no that important - the bias is usually relatively constant and only the variable part would really be a problem. The really tricky parts are more like offset drift with time and possible offset hysteresis - these parameters are not often specified.. For a BJT based OP the filter could also provide source impedance matching the divider.

FYI - While messing with my junkbox I discovered a pack with OP07AY in CerDIP14 packages, 1979-85 by PMI and "M". After an intensive search I found the datasheet with package.
Enclosed the shot from the datasheet (PMI 1982) as the cdip14 pin layout it is not easy to find.
Hopefully the chips could be used with 399/LT1021/REF01 or LTZ.

[Gyro]

A very nice hermetic collection, especially being A spec. I've never seen a 14 pin OP07 in the flesh.

[imisaac]

VHD network used only in prototype. Final units including the one I linked use unobtanium laser-trimmed hermetic Fluke film resistor (used in Fluke 8846 DMM).

Output stage opamps used for both force and return drive bipolar ability. This way you can abuse output with caps, shorts, reasonable reverse backdrive and there is no major risk to kill expensive LTZ.

As of annual stability, I will know answer to that question next month, currently in progress to obtain SI Volt from Josephson reference via transfer by Fluke 732.

Existing LTZ thread have lots of data, so time spent to read it thru can answer some of your questions here already. Either way LTZ designs aren't what many can call "beginner" level

Thanks, TiN. I have another question related to the "output stage opamps" in your circuit. If the two BJT transistors are replaced by two buffer opamps (see figure attached), would you expect it to worsen the voltage reference stability/noise performance? I am also not sure myself if isolating the ground with two opamps is any better than just using the ground pin directly.

Secondly, there are two stages of low pass filter at the output of the 7 V reference. Wouldn't you add more 1/f noise, due to capacitors of the filter, to the 7 V reference then without using it? It may be true that the high frequency noise is reduced, but 1/f noise may be the more important one here in this part of the circuit. If I move the low pass filter to the output of the ADA4522 (in the "High" portion), it may work better?

Last but not least, how do I move this thread to the "Metrology" section? Done already.

Thanks in advance.

 

[CalMachine]

Secondly, there are two stages of low pass filter at the output of the 7 V reference. Wouldn't you add more 1/f noise, due to capacitors of the filter, to the 7 V reference then without using it? It may be true that the high frequency noise is reduced, but 1/f noise may be the more important one here in this part of the circuit. If I move the low pass filter to the output of the ADA4522 (in the "High" portion), it may work better?

That LPF is put there to protect the LTZ from current spikes of the chopper