Ultra Precision Reference LTZ1000

[Kleinstein]

My estimate below shows, that a temperature gradient between the pins should not be that critical. It would take more than about 0.3 K of difference to make the error larger than the "normal" 0.05 ppm/K TC of the LTZ1000 circuit. This should not be that difficult with a symmetric layout. An extra heater would be kind of overkill. If at some extra copper to improve the coupling between the pins should be all it takes.

Today the thermal performance can be simulated if needed, and it is reasonable repeatable. This might make is even possible to replace the 400 K compensation resistor with an intentional temperature gradient. However I am not sure the needed gradient is easy to obtain - it might need a significant asymmetry.

[floobydust]

Kovar delta-T of 0.1 K gives 4uV or 0.6ppm at 7.1846089V ref. Does that seem correct?
To me it looks critical, keeping all the pins at the same temperature throughout varying ambient temperature. I hear you can cal it out as a fixed offset, but the danger is if the pin-pin temperature gradient ever changes magnitude, due to changing heat loss/gain across the PCB.

34470a having fan airflow across the ref. PCB either will either swamp that or make it worse, as I think the fan is variable speed. The voodoo slots insulate in three areas but are not symmetrical, and LTZ pin 3 is at the PCB edge.


3roomlab, from TiN's IR pic, I see radiant heat transfer from the underside of the LTZ to the PCB, so the PCB pads are heated by both (short leads) conduction and radiation from the ref. IC. The challenge seems to be keeping symmetrical heat loss around the full circle. I did not see the glass as part of your thermal model. Is it in there.

[Andreas]

My estimate below shows, that a temperature gradient between the pins should not be that critical. It would take more than about 0.3 K of difference to make the error larger than the "normal" 0.05 ppm/K TC of the LTZ1000 circuit.

I think it is much more critical than that what you can estimate.
My PSRR experiment shows for a battery voltage between 14 and 18.5V a temperature difference of 1.5 deg C measured near the voltage regulator (LT1763) which is a couple of cm away from the LTZ.

On the NTC near the LTZ1000 you can see practically no change from the voltage regulator heating. (only some ambient temperature changes).
So the temperature on the LTZ pins is nearly not affected by the voltage regulator heating.

But: the change on LTZ output is repeatable around 3uV (nearly 0.5 ppm or 0.1 ppm/V) difference between 14 and 18.5V at the voltage regulator input.

So what really counts are measurement values. (We still have not enough).

with best regards

Andreas

[floobydust]

The third heat source is the heater transistor- in your circuit, is it supplied pre or post voltage reg? I could see it's dissipation also increase with higher Vcc, if pre-Vreg.

[martinr33]

The heater has to be behind the regulator, as the LTZ1000 is very sensitive to operating temperature.

The 3uV is going to be from differential thermal EMFs. We are talking about really small differences, but they need attention. This explains why  you see slots between the power section and the LTZ on some of these boards. That new Keysight reference seems to have a lot of good ideas in it, questions around final execution notwithstanding.

At the end of the day, we are niggling over very small things. But they add up, so eliminating everything possible that we can hit through layout is a good idea.

[chickenHeadKnob]

3roomlab, from TiN's IR pic, I see radiant heat transfer from the underside of the LTZ to the PCB, so the PCB pads are heated by both (short leads) conduction and radiation from the ref. IC. The challenge seems to be keeping symmetrical heat loss around the full circle. I did not see the glass as part of your thermal model. Is it in there.

I believe that is a mirror reflection artifact from the copper ring on TiN's kx board, copper is an excellent IR reflector. Using teflon as a substitute for FR4 in simulation is suspect, as I expect pure teflon is a much better thermal insulator.

[hwj-d]

Found a nice article in
"http://www.electronicdesign.com/analog/what-s-all-ltz1000-stuff-anyway"

about LTZ1000's, TiN's KX-design, and plenty of background information and references to basic articles.

Particularly interesting for me in this context, the thermal image of the temperature antistress distribution.

Thanks again to TiN 

(edit: yes, i know, well known infos for some )

[Andreas]

The third heat source is the heater transistor- in your circuit, is it supplied pre or post voltage reg? I could see it's dissipation also increase with higher Vcc, if pre-Vreg.

the schematic is here:
https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/?action=dlattach;attach=189655

So all is supplied from the 14V voltage regulator LT1763 to reduce battery voltage dependency.

with best regards

Andreas

[3roomlab]

The third heat source is the heater transistor- in your circuit, is it supplied pre or post voltage reg? I could see it's dissipation also increase with higher Vcc, if pre-Vreg.

the schematic is here:
https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/?action=dlattach;attach=189655

So all is supplied from the 14V voltage regulator LT1763 to reduce battery voltage dependency.

with best regards

Andreas

what is the ball park power consumption (entire circuit powered) of this 14V variant 1) approx at start up? and 2) when stable?

[floobydust]

it is substitute with teflon. in FEMM there is glass from 0.8 to over 4 W/m.K. nobody know what is LTZ type 

Schott uses borosilicate glass 8250 (hard) with Kovar. Corning seems to have discontinued their semi glass line. Some data pg 22
Datasheet with thermal conductivity data for 8250 glass at 90°C = 1.2W/mK.

[TiN]

Bottom "heat" pattern is definately IR reflection on the gold plated circular shape.

I since I didn't build any LTZs lately, must not resist the urge to build few more, this time bit different, and more traditional design to discuss about. 

PNG-schematics

 

[dr.diesel]

I since I didn't build any LTZs lately, must not resist the urge to build few more, this time bit different, and more traditional design to discuss about. 

WooHoo!

[Kleinstein]

Bottom "heat" pattern is definately IR reflection on the gold plated circular shape.

I since I didn't build any LTZs lately, must not resist the urge to build few more, this time bit different, and more traditional design to discuss about. 

PNG-schematics

 

The LTZ1000 part looks very conventional. I somewhat doubt one would really need 0.2 ppm/K resistors. The important parameter on the resistors should be long term stability, not the TC. Of cause they usually come together, but not always. Just for the TC the resistors look better than needed. Due to the high current output a transistor the LT1013's output might be a good idea, though not absolutely needed.


For the voltage regular, one usually needs the capacitors just at the input and output of the regulator, even close by. So even just a ferrite bead as shown could make the regulator oscillate. Extra inductive filtering should be on the input side and maybe after the first set of capacitors at the output. If you really want the LTC2057 in the circuit, I would use a separate regulator (even if just an 78L15), as the 2057 would be a significant source of RF noise. One might also want some filtering between the LTZ reference an an AZ op at its output.

The 7 to 10 V circuit shown does not look like it would tolerate capacitive loading and could compensate for voltage drop at the GND side. Not sure with this, as the drawing is a little confusing. The resistors in the 7 to 10 V part are way more critical (e.g. a factor 100) than in the LTZ1000 part. So one should also have an direct output at the 7 V level so one could at least check for long time drift of the resistor ratio.
The divider might also need resistors for trimming, if exactly 10 V are intended.
Due to the errors expected from the resistors in the 7 to 10 V stage one might consider a good conventional OP instead of a AZ type. This might need slightly lower resistor values however.

[Andreas]

what is the ball park power consumption (entire circuit powered) of this 14V variant 1) approx at start up? and 2) when stable?

The battery current of LTZ#3-#6 at room temperature is listed here:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg874560/#msg874560

The power up duration is only several 100 ms so it is negligible. You can calculate the value from the heater resistor in the data sheet.
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg832030/#msg832030

with best regards

Andreas

[branadic]

I'm back in the LTZ club again.  I had some issue with the yellow resisitors used in the cirrcuit, the 120R resistor measure high resistance after a while. That's why I dropped all of them out of my circuit.

Within the last two days I received a bunch of goodies, a resistor from Frank to repair my LT-LTZ000 board, resistors from Rhopoint, LTZ1047B boards including batterie monitor from Andreas and some material like LTZ1000 and capacitors for repairing my Keithley 181 (which is now working again) from Digikey. A lot of stuff to assemble the next weeks. However, the resistors in my first LTZ1047 are replaced and I have to observe if the LTZ took any damage. The LT-LTZ1000 is now repaired, but needs some minor changes as it differs from datasheet circuit.

I also received the mysterious reference from Philipp, which I'm pretty much sure it is an arrangement of 4x LTx99 together with some OP07, but I'm still reverse engineering the circuit. News about it in the corresponding thread. And I received some wonderful standoffs from Emmanuel. Thank you very much, thanks to all of you guys. I really enjoy your contacts.

-branadic-

[martinr33]

On the new TiN design:

1) How about a series pass FET on the output? Good way to keep everything safe.

2) The Fluke resistor pack is a bit of a rarity. If we can get the output voltage close, how about fine-tuning it with a DIP switch and a resistor pack? Tolerance and stability is not so critical in these last bits. 8 bits might not be enough - but it seems that 16 would do. Probably 16 switches, and resistor values selected such that there's a bit of overlap.

[hwj-d]

2) The Fluke resistor pack is a bit of a rarity. If we can get the output voltage close, how about fine-tuning it with a DIP switch and a resistor pack? Tolerance and stability is not so critical in these last bits. 8 bits might not be enough - but it seems that 16 would do. Probably 16 switches, and resistor values selected such that there's a bit of overlap.

Don't know where to get that at all. That's why I would prefer the ability to use a precision resistor-cascade with low resistance and -trimmer as last stage like in Dr.Frank's design.

[TiN]

2) The Fluke resistor pack is a bit of a rarity.

Lucky for me, greed has no limits here. However it's optional, one can use "simple" VHD200 R10/R11 + trim resistor R9.

Full disclaimer before this get out of hands, this reference is one off design for specific purpose to put inside the Fluke battery pack (hence the name, "Fluke eXperimental") and use as a transportable standard for 3458A/5700A. So input protection, output protection and trimming will be implemented once I start PCB design to see how it all goes.  LTZ's resistors will have option for PTH PWWs from Edwin as well. 0.2ppm VPGs right now in schematics are just reused from old KX design.

As result, I have no means to make this design commonly repeatable like the older one, however design files will be public once performance is verified (think 2019).

[Echo88]

"Before this gets out of hand" ...i thought thats why we gathered here 

Anyway: Does anyone know if SO8-Package LT1013 contribute more drift in the long term compared to the DIP-package? Because of maybe humidity dependence and pcb-lenght-variation due to temperature/humidity-changes and therefore mechanical stress on the LT1013-package.

Also: It would be nice if every LTZ1000-board in a bank would have its isolated supply which is dead quiet, like its done in Fluke 7001-modules https://www.eevblog.com/forum/metrology/teardown-voltage-standards/msg902910/#msg902910 Are there any suggestions how to accomplish this or ready made solutions? Maybe standard transformer + LT3042 already suffices?
Interestingly Fluke 732B and 734 doesnt feature the claimed quiet isolated dc/dc-converter (Fluke 7001) which would enable mains powered Fluke 7001-modules to be directly compared to a JJA. So how are Fluke 732B/A compared to a JJA, during battery operation or mains powered? 

[TiN]

SO8 suspected not that much from humidity but from mechanical stress directly from the board due to much shorter pins.

I was happy with results on my isolated supply for nV head (X1801) using LT3439, standard CTX02 trafo and LT3042+3092 on the output. However for this current LTZ ref it's not going to be used due to extra power waste (battery life) and no real need (standard should have enough juice to work few days without mains).

[Kleinstein]

The DIP LT1013 is slightly better than SO8, mainly due to board stress. However both should be good enough, since one a small fraction (around 1/200) of the OPs dirft will appear at the output.

It is not he LTZ1000 circuit that needs a separate supply, but it can be the part around it. It can sometimes save you an extra OP for separate force sense lines.

If there are batteries in the circuit anyway for transport, one can use them for calibration of critical measurements too. So there is a limited need for a highly isolated supply.

[lars]

As far as I know the Vishay resistors TiN have in the schematic (Z202 and HZ) are not 0.2ppm/C but according to the datasheet +-2ppm/C. The 0.2ppm/C specified as typical more seems to be the variation of the temperature coefficient over the whole temperature range??

As the Z202 specified for the two most critical positions (1k and 12.5k) are epoxy I would be even more worried for humidity than temperature that to some extent can be compensated. I have not tested Z202 but the old S102 definitely, from both my own measurements and also Vishay app notes, are humidity sensitive up to about 1ppm/%RH. This is also in the same region I have measured for WW like 8E16 and also SMD0805 like RN73, ERA 6A and PCF0805. The sensitivity varies both between same value resistors and also between different values. Lower values seem to be less sensitive.

The long term drift for WW I have tested are low maybe 1-2ppm/year compared to the SMDs that are more 5-20ppm/year after the first months.

This is for no load or very light loads at normal lab temperatures and mounted on FR4 boards.

Lars

[TiN]

Updated schematics to F01.

* Power nets update, LDO output is +11V due to input power is +12V pack. R15 changed to 110K accordingly.
* Added optional C22
* Guard nets around LTZ driven by R16-R17 network
* Temp sensor IC powered from isolated +5V/IGND
* Added filter RCR R18,C24,R19 between 7V out and chopper input.
* Output gain amp replaced to ADA4522-2, with second amp used as output buffer/sense.

[Andreas]

Hello,

- C7 will make U6A instable
- C22 does not help much without a additional resistor between C22/U2A and R1.

With best regards

Andreas

[martinr33]

A couple of small comments around the output buffer.

1) The 2N3904 is a bit noisy for the output transistor (6dB specification). You could use a low-noise part, typically around 2dB, instead. Or maybe even a series pass FET, as per the Fluek 732.

2) You need to get something to deliver negative feedback into  the output amp. Otherwise, the output voltage will slam up to near rail when the output connector is disconnected. A reasonably high value resistor should do it, maybe 10k.

Also - these resistors may all need to be low-noise parts.