Ultra Precision Reference LTZ1000

[Echo88]

Regarding paralleled LTZ1000 for lower noise:
Im working on a lower noise ADC-board variant for my HPM7177-version.
For that im thinking of using 4x LTZ1000 -> ~0.5µVpp or if i ever get these: 4x ADR1000 -> 0.25µVpp.
Question is how to best achieve the 4x averaging without errors considering the GND-sense-concept of the HPM7177.
What would be the best way:
-Average raw 7V and GND-sense of every ref with resistors, without buffer
-Average raw 7V and GND-sense of every ref with resistors following a OP-buffer like ADA4522 which has 117nVpp and is already on the BOM?

Im also unsure about startup and interference between refs due to the GND-sense connections.

I attached schematics that showcase the ideas with and without buffers.
So far i know only of the QVR from TiN which is a quad-LTZ/ADR-reference board, are there other known publications of paralleled LTZ/ADR-boards?

https://xdevs.com/article/qvref/ 
https://www.analog.com/media/en/technical-documentation/data-sheets/ada4522-1_4522-2_4522-4.pdf

[dietert1]

In february i started making a 6x LTZ1000 array and opened a separate thread.
https://www.eevblog.com/forum/metrology/making-an-array-of-ltz1000/msg4771946/#msg4771946
I tried to make the cells of the array as simple as possible and put everything onto a very small board that can later be used as one reference.
I omitted LTZ1000 heater control. This can work well inside an external oven after fine-tuning each cell for zero TC at the same oven temperature. Fine-tuning the TCs required a temperature chamber with programmable temperature and development of a reliable tuning scheme.

Regards, Dieter

[Dr. Frank]

I wanted to find out about the T.C. behavior of my HP3458A, i.e. what's the influence stemming from the T.C. of the reference board, and which one from the A/D converter assembly U180.
The specification "with ACAL" assigns 0.15ppm/K to the reference board, and "without ACAL" are 0.5ppm/K for the overall T.C., i.e. the sum of both.
This overall T.C. makes manual transfer and stability measurements over a longer time difficult, especially when measuring at sub ppm level.



I pulled the reference board out of my 3458A, and assembled it into an aluminum box, with 18V supply for heater and reference circuit, reference output to KS34465A DMM, and a precision NTC, tied to the 75k resistors on the board, and to be measured by an HP34401A.
The oven temperature is lowered to about 70°C, by the 100k parallel to the 15k resistor.
Also notice, that the T.C.-correction resistor R417 = 200k is assembled, which should NOT be used for the A-version, and probably creates an unwanted T.C. of the board.



The LTZ1000A on the reference board is in upright orientation inside the aluminum box, when latter is closed and placed with its bottom plate on the electrical heater.



Such low TC measurements are quite delicate, and I was now missing another "silent" and stable DMM like the 3458A.
The KS34465A has an LM399 inside, quite noisy, and might drift too much.
So I used its RATIO function, and a FLUKE 7000 as a reference. This deletes temperature and time drifts of the DMM.



The noise of the 34465A is mitigated by taking data each 0.1K, or each 30 sec.

The aluminum box was first cooled down to an inner temperature of < 18°C, and then during the slow warm-up, the whole assembly was nearly in thermal equilibrium, so the T.C. could be determined to about +0.95ppm/K (linear regression or box method)



For the regular operation temperature range inside the 3458A, usually 35..36°C in my lab, I heated the aluminum box to about 43°C, and let it slowly cool down.
This gave a T.C. of about + 0.11 ppm/K.



So this T.C. of about +0.1ppm/K is well inside specification.

I intended to bring its T.C. to zero, by removing or changing R417, improve its long term drift by further reducing the oven temperature, and would have replaced the ceramic capacitors by film capacitors, to reduce possible noise generation.

As timely and temperature drift is very low anyhow, I finally left the reference board as it is, for now.

[Dr. Frank]

My HP3458A has shown an overall T.C. of about +0.42ppm/K, when my wife opened the basement window unintentionally:



From monitoring the CAL? 72 parameter, I determined roughly the T.C. of the U180, which is about -0.32ppm/K


So it was evident, that the T.C. of the reference board should add in the same direction, and is now confirmed to be about 0.1ppm/K

That overall T.C. of 0.42ppm/K may disturb my transfer measurements, which take about 40minutes for 13 voltages.

If I encounter a room temperature drift of only 0.5°C, due to the heat generation of the fluorescent lamp, and of the 100W from the operator, I already get a 0.2ppm shift. As the reference with 0.1ppm/K gives no real chance for improvement, compared to the 0.32ppm/K for U180, I have to take care, that the room temperature rise stays below 0.2°C over the session.

Btw.: using the known T.C. of U180, I was able to calculate the timely drift of U180, from monitoring CAL? 72 parameter.



That gives about 1.5ppm over 4 years, or 0.001ppm/day. Therefore, my U180 chip is not affected by the Service Note 18.

[aronake]

For the regular operation temperature range onside the 3458A, usually 35..36°C in my lab, I heated the aluminum box to about 43°C, and let it slowly cool down.
This gave a T.C. of about + 1.1 ppm/K.

You later wrote:
"So this T.C. of about +0.1ppm/K is well inside specification."

There seems to be something I don't understand. Assuming the graph show PPM on y-axis and degree C in x-axis, it looks like 1.1 ppm/10 degrees which is the 0.1ppm/K later mentioned.

Or just a typo?

[Dr. Frank]

For the regular operation temperature range onside the 3458A, usually 35..36°C in my lab, I heated the aluminum box to about 43°C, and let it slowly cool down.
This gave a T.C. of about + 1.1 ppm/K.

You later wrote:
"So this T.C. of about +0.1ppm/K is well inside specification."

There seems to be something I don't understand. Assuming the graph show PPM on y-axis and degree C in x-axis, it looks like 1.1 ppm/10 degrees which is the 0.1ppm/K later mentioned.

Or just a typo?

Sorry, just a typo.
That happened, because I'm struggling today with adding / inserting the pictures. The first one is always deleted.
 Something is not working properly with the posting software, or I'm too confused today.

[Dr. Frank]

My basement lab has a very constant room temperature w/o HVAC, and when the room is not occupied.
In winter, only the gas heating is running, so you see the regular night setback, and a nearly constant r.t. of about (21.2 +/- 0.2) over weeks.
This can be used to make unattended/automated stability measurements on a reference DUT with the 3458A directly.



I made transfer measurements on November 1st, 16th and December 1st. These semi manual measurements always raise the r.t. by 0.5 .. 1°C, but may reach a constant temperature later on in the session.

So I have to take this change into consideration for the 13..14 volt transfers, taking about 30..40min, which by far exceeds the transfer specification of the 3458A.
For the session of Dec. 1st I powered my equipment on the night before, at 10 p.m., which raised the r.t. to 22.4°C.



In my recording sheets, I write down date / session start time / initial room and / 3458A temperature / ACAL temperature / CAL? 72 / end time / end temperatures.



I began the first transfer session for my 7 resistor references before, at 9:56, lasting 23 min., and raising the temperature to 22.6°C, which was stable afterwards, as I didn't move out of the lab. During the following voltage transfer measurement between 10:50 - 11:25, r.t. and TEMP? of the 3458A raised by 0.1°C only, which would give about 0.05ppm for this TC induced change. I always begin with LTZ #1, and re - measure it in the end, which really gives about -0.06ppm transfer repeatability, as expected.

On Dec 16th, I made the next transfer measurement, only on my voltage references, so the temperature change was +0.3°C, giving about 0.13ppm shift.
LTZ #1 was measured +0.17ppm high compared to the start of the session, so that fits well.

As the 3458As reference has a 0.1ppm/K T.C. only, I could do an ACAL DC in between, as soon as the TEMP? rises more than 0.1°C, or I could repeat the session when the r.t. has settled.

One important aspects has to be addressed:
I have to make absolute voltage measurements by means of the 3458A for these transfers, as I have to compare voltages on very different levels, i.e. @ 7.1V and 10V.
Making differential measurements only, would nearly avoid all such T.C. related shifts, but would require a stable precision divider like the FLUKE 720A, to differentially compare all references against the FLUKE 7000. That transfer method can be done with about 0.1 .. 0.2ppm uncertainty, so it's questionable, which method is best.

[aronake]

My still not completely verified approach to separately measure TC of A3 and TC of LTZ1000 on 3458a has been to measure change in cal72 with temperature for A3 TC and change in measured voltage after ACAL for LTZ1000 TC. Technically I have done this by keeping voltage reference in TEC box and using the AC to change temperature in my "laboratory". TC of cal72 + TC of voltage measurement after ACAL has quite well matched total 3458a (with no ACAL) in my sample universe among the 5 3458a I have. Plan is to have this verified by making TC sweeps of the bare A9s as Dr Frank has done.

4 of the 3458a have a LTZ1000 TC of 0.05 to 0.12 PPM/C. The last one have 0.19 PPM/C. Idea is to dissect this out of spec A9 and measure TC on all resistors separately and then tune it to 0 TC.

[guenthert]

[..]

One important aspects has to be addressed:
I have to make absolute voltage measurements by means of the 3458A for these transfers, as I have to compare voltages on very different levels, i.e. @ 7.1V and 10V.
Making differential measurements only, would nearly avoid all such T.C. related shifts, but would require a stable precision divider like the FLUKE 720A, to differentially compare all references against the FLUKE 7000. That transfer method can be done with about 0.1 .. 0.2ppm uncertainty, so it's questionable, which method is best.

   Clearly using the DMM for direct measurement is more convenient, faster and can be automated.  If the linearity of the DMM is known or can be verified, that will then be  preferred.  Since the divider might have itself a (small) temperature dependency, I'd think it can be used to its fullest potential only in temperature controlled environments.  You are already in a better position than most of us there.

    Perhaps best in your position would be to (occasionally, say every other year or so) verify the linearity of the DMM and use that for measurements of the DUT.  I've seen others attempt such, by measuring 1,2,...9,10V from a calibrator (assumed to have better linearity than the DMM under test) or ~1V from a string of resistors fed from a stable source (and adding the individual  measurements).  I can't help by wondering, whether that is sufficient. @Kleinstein hopefully can comment on this, but afaiu, the capacitor of the integrator in the ADC of many precision DMMs, like the 3458A, is charged only to some comparably small voltage (much less than 1V) repeatedly in order to avoid nonlineratiies due to the voltage depending varying capacity.  Then it might be more appropriate to test the linearity at smaller voltages, i.e. fractions of the maximum voltage of the capacitor of the integrator.  This probably belongs  in another thread though.

[Kleinstein]

4 of the 3458a have a LTZ1000 TC of 0.05 to 0.12 PPM/C. The last one have 0.19 PPM/C. Idea is to dissect this out of spec A9 and measure TC on all resistors separately and then tune it to 0 TC.

I don't think one would need to measure the TC of all resistors separate. Unsoldering would be quite some stress to the parts. The first cadidate to look at the the R9 (200K/400K) resistors used to trim the TC, that HP populated even with the A version that does not need it. Best case one gets away with just removing the resistor and maybe add a different (likely larger) value to do a fine trim. Though only a relatively small disturbance one would still loose the absolute voltage calibration (may not be that bad if one has 5 meters at hand). In this case the question would be if one also wants set the temperature lower. It has pros and cons, e.g. limiting the max environmental temperature and a phase of more possible drift from a new settling phase.

The 3458 is normally quite good with linearity and the 7 to 10 V step is not that large. One may reduce the effect a little by using the meter with both polarities and thus do the test with negative voltages too. Actually testing the INL of the 3458 to a meaningfull level is difficult and by nature only a spot check. For the use here maybe check if the 10 V ref. is the same as the sum of 2 "halves"  (with a divider and buffer). It is more be a check if there is a defect causing way out of spec INL.
The main candidate for nonlinearity in the 3458 is not so much the integration capacitor and if at all the dielectric absorbtion there and not a voltage dependent capacitance. The main candidates are thermal effects in the resistors (self heating causing a tiny U³ contribution) and settling effects at the integrator. Much of the INL is not expected to change very much over time and would thus not matter that much for looking at the drift, if the same meter is used for all tests.

[aronake]

4 of the 3458a have a LTZ1000 TC of 0.05 to 0.12 PPM/C. The last one have 0.19 PPM/C. Idea is to dissect this out of spec A9 and measure TC on all resistors separately and then tune it to 0 TC.

I don't think one would need to measure the TC of all resistors separate.

Thanks for comments. This would absolutely not have anything with some kind of need to do, 100% out of curiosity. I suspect the individual resistors to be quite bad from a TC perspective.  But lets see. I only intend to give on A9 this treatment, so very minimal statistical input. It would not surprise me if HP/Agilent/Keysights approach to making A9 boards is to use quite mediocre resistors then temperature sweep the A9 boards and with quite relaxed tolerance, which seems to be 0.15 PPM/C or maybe a little lower, discard these that was not good enough or replace the resistors and do a new temp scan.

[julian1]

For a pcb ref board - should gnd copper fills run right up to and around the ltz1000?

The trade off -
- Having outer-layer copper fills that enclose inner signal traces, should help shield EMI a bit more.

- Alternatively the extra copper around the ref pins, means less thermal isolation to the pcb and other TC sensitive components.
Although there may already be plenty of thermal conduction with the traces.

[Kleinstein]

The normal way is having relatively thin traces going to the LTZ1000 and no extra ground plane to keep the heat loss low. If at all one could have a ring of copper directly around the ref chip (e.g. top side) to reduce thermal gradients there.  An improtant point can be getting the output and ref. ground with seprate traces. For the ground side this often means having ground current compensation that goes to the ref. chip with a separate trace.  With a more isolated reference one could have the central ground point at the reference- This however does not work that well with a more complec circuit (already a 7 to 10 V amplifier).


An unusual point in the layout is that the heater transistor is relatively close to the reference. Most plans have it at the opposite end to keep the heat away from the reference.  The transistor close to the reference may not be that bad, as the transistors heats more when it is cold and this could act has an additional heater to reduce the temperature swing seen be the reference.  As shown the transistor is connected a bit one sided. Chances are this is still good enough and would not make a difference either way.

[Andreas]

Hello,

- keep the pins (solder joints) at the same temprature (especially pin 3 + 7).
- EMI-wise: the substrate = metal can is not GND (but pin 4) so for EMI you will need a metal shield around the whole LTZ. Using GND-planes is useless.
  EMI enters by the power supply or output lines and is capacitively distributed by the metal can of the LTZ to the environment.

with best regards

Andreas

[julian1]

Kleinstein, There is a gnd current scheme, but it never clicked, that it should be injected at the pin - where lo is sensed.
Although, I oddly have done exactly that for other lesser soic-8 references.

Andreas, noted that ref emi issues can't be addressed with pcb copper.

A+ Thanks.

[floobydust]

I stumbled onto a PCN 23_0011 14-Jul-2023 related to the LTZ1000 die-attach, it appears they use 90-106 micron soda lime solid glass balls for it. Makes sense as thermal insulator and the die rides a cushion of glass balls for minimum stress I guess?
"Current Glass Sphere supplier has informed ADI that they will no longer provide this material. ADI is qualifying a replacement Glass Sphere material supplier, COSPHERIC to be used at LTZ1000 Manual Die Attach process."
Product: SLGMS-2.5 90-106u, #130803-200.

[Andreas]

Hmm,

there are some open questions:

Why is also the LTZ1000 (non A) mentioned in the PCN and not only the LTZ1000A?
in the "qualification tests" I had at least expected a comparison of thermal resistance chip to environment between old and new material.
Better would be also a comparison of ageing drift of the chip.

I fear a "visual inspection" does not help much for the relevant parameters.

with best regards

Andreas

[floobydust]

I found pictures of the glass balls in ADR1001 Richi's Lab die shots but do not see them in the LTZ1000.
Could not find die shots of the LTZ1000A die attach. I would assume it's the same so the PCN is in error