Ultra Precision Reference LTZ1000

[branadic]

Just for the sake of completeness, the Advantest R6581(D/T) uses a similar reference board to the A9 board in HPAK 3458A.

-branadic-

[Andreas]

Hello,

just want to remember that I have already published my design (with design decisions) for LTZ1000 References some time ago:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg249123/#msg249123

with best regards

Andreas

[MiDi]

I now added Andreas noise improvements by 5 capacitors and one resistor, see schematic.
Then I doubled the zener current by changing to a 50 Ohm resistor.

Both measures decreased the StD and therefore, the uncertainty of DCV measurements down by a factor of 1.5 .. 2.
Before, I always measured my ADR1000 reference at typically 150nV StD, but rarely below 100nV, but now, ALL consecutive measurements are well below 100nV, but rarely @ 150nV, at most.

Last, I removed the hp plastic cap and put another cap all over the reference directly on the PCB. This already improved the StD of my 34465A by a factor of 1.5, as the KS engineers made this design flaw to let the fan blow directly over the reference area.
I doubt that it'll help here, as the LTZ does not sit directly in the air flow.

Are these the mods you made to A9?



Did you measure the TC of A9 with all the mods and the overall cap?
It looks like there is quite some TC in your plot:

When making direct comparisons with this absolute method by means of an DMM, I always wondered, why I never could achieve the very same noise figures as the differential method, although both methods are essentially identical: Any DMM directly measures the difference between two (LTZ1000/LTFLU/LM399) references, and the A/D only serves as the differential instrument, contributing no additional noise.

A9 is main noise contributor in 10V range at FS, but with low noise A9 the A/D becomes the main noise contributor, still the noise in 100mV range has to be lower than in 10V.
Edit: Clarification 100mV: differential measurement of two references, 10V: direct measurement of one reference (comparison to A9). Additionally the differential measurement vs direct has lower uncertainty.

Do you have intermediate comparison to quantify which mod had which effect (Andreas mods, doubling zener current, overall cap)?
Looking forward to your Allan deviation plots and comparison to Chuck and mine.

[Andreas]

Are these the mods you made to A9?

I hope not.
The 100nF drawn parallel to C414 should be across the reference output voltage 4wire connected from output connector to zener (Pin 3 + Pin 7) to remove noise which may be introduced from output to the zener.
Btw: Without this capacitive load you would  not need the additional 10K + 100nF around U402b to keep loop stability.

Of course the addional 22nF is useless (If not 2nF are populated for C411 as being wrong in some data sheets).

The X412 and X414 capacitors have the intention to protect the BE-Diode from demodulating external noise.
So they should be drawn there where they belong (between Pin 6+7 and Pin 4+7) (with short wires see attached picture).

with best regards

Andreas

[Dr. Frank]

I now added Andreas noise improvements by 5 capacitors and one resistor, see schematic.
Then I doubled the zener current by changing to a 50 Ohm resistor.

Both measures decreased the StD and therefore, the uncertainty of DCV measurements down by a factor of 1.5 .. 2.
Before, I always measured my ADR1000 reference at typically 150nV StD, but rarely below 100nV, but now, ALL consecutive measurements are well below 100nV, but rarely @ 150nV, at most.

Last, I removed the hp plastic cap and put another cap all over the reference directly on the PCB. This already improved the StD of my 34465A by a factor of 1.5, as the KS engineers made this design flaw to let the fan blow directly over the reference area.
I doubt that it'll help here, as the LTZ does not sit directly in the air flow.

Are these the mods you made to A9?

(Attachment Link)

Did you measure the TC of A9 with all the mods and the overall cap?
It looks like there is quite some TC in your plot:

When making direct comparisons with this absolute method by means of an DMM, I always wondered, why I never could achieve the very same noise figures as the differential method, although both methods are essentially identical: Any DMM directly measures the difference between two (LTZ1000/LTFLU/LM399) references, and the A/D only serves as the differential instrument, contributing no additional noise.

A9 is main noise contributor in 10V range at FS, but with low noise A9 the A/D becomes the main noise contributor, still the noise in 100mV range has to be lower than in 10V.
Edit: Clarification 100mV: differential measurement of two references, 10V: direct measurement of one reference (comparison to A9). Additionally the differential measurement vs direct has lower uncertainty.

Do you have intermediate comparison to quantify which mod had which effect (Andreas mods, doubling zener current, overall cap)?
Looking forward to your Allan deviation plots and comparison to Chuck and mine.

Hello MiDi,
yes, these are the changes I've made, basically. Thank you Andreas for your additional hints!

I also think, that this change will only be effective into the nV noise floor, when the other left noise source, i.e. mainly the two +/-12V amplifier are replaced as well, like by some OPAxxx types. So I do not expect yet a spectactular improvement in the Allan deviation, as the noise of the LT1009 opamps might be in the same order of magnitude as an ADR1000.

We'll see, a much more temperature stable measurement is just done over night.

I did not iterate all the changes, so I cannot tell, which was the most effective one.
Frank

Yes, there is a slight temperature variation visible in my graph, as I've done T.C. determination in the same experiment. 

[aronake]

Hello,

just want to remember that I have already published my design (with design decisions) for LTZ1000 References some time ago:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg249123/#msg249123

with best regards

Andreas

With 12 years since these modifications was made, and you probably accumulated lots of more knowledge on LTZ1000 circuits, would you likely do the same today or something different? Have you somehow been able to evaluate the modifications either all together, or better, part of the modifications separateley?

[Kleinstein]

One can not add the extra capacitors one at a time - at least not all. One still has to keep the loop around the OP-amp with the extra gain of the transistor stable. So there are limits to having only part of the modification. The capacitor from base to emtter is likely the main improvement (less EMI sensitive), but requires much of the other modifications to avoid oscillation.

[Dr. Frank]

Hello MiDi,

here's the Allan Deviation.
I still got no stable measurement.

The first two hours the 3458A had to warm up again, after I had to open it to mount the ADR1000 chip.
The ADR1000 obviously always needs a long time after switching it off to recover to its original value.
The usual noise minimum (20ppb) at tau = 20s is overwhelmed by this recovery effect of the ADR.

I already have observed this effect, when I switched it off for weeks and months, and the +5ppm shift took about 2 weeks to recover.
Here's a further confirmation of this effect, but did not expect at all, that the ADR1000 is that sensitive against interruption, and would take so long even after very short ones.
Again, this characteristics make the ADR1000 useless for DMM, or you need to let it powered always on.

I also have two more promising values, i.e. for NPLC100 the noise is about 9ppb, and 6ppb for NPLC 1000. Still not good enough.

I have now mathematically subtracted the drift effects, but still there is no remarkable improvement visible.
This minimum is now at about tau = 100s, at about 10ppb, instead of 20s/20ppb.
This measurement ran for 12h.

I have to investigate further, I guess.
 
Frank 

[Andreas]

With 12 years since these modifications was made, and you probably accumulated lots of more knowledge on LTZ1000 circuits, would you likely do the same today or something different? Have you somehow been able to evaluate the modifications either all together, or better, part of the modifications separateley?

I made a systematic check on the modifications on my "EMI victim LTZ#9"
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2758840/#msg2758840

So finally I will do in a further rework:
Use ADA4522-1 as output buffer. (with RF coupling capacitor from output to (separate) power ground at the OP-Amp)
Use a additional common mode choke for the output.
either self built with a NiZn Ferrite or a 51uH "CAN-bus" choke.

With best regards

Andreas

[MiDi]

Wow, your unit is really low noise! (if the input is 10V)
Allan deviation ~15nV vs ~65nV (my modified unit) - input short has ~13nV @100s for comparison
Mod. AD: mine ~55nV, Chucks ~40nV (input short: 10nV vs. 13nV @100s)
Would be interesting to get AD for short for 10V range of your unit too.

For my A9 ADR1000 there is little shift for short downtimes (down into uncertainty of ±0.1ppm).
For a 2 week downtime there is little shift ~0.5ppm visible and mostly recovered in the last 4 days - have no comparison how original A9 behaves.

[Dr. Frank]

Hello Midi,
sorry, the ADR1000 inside the 3458A @ constant temperature keeps on creeping upwards, even after 2 days.
I will later publish my conclusions about this behaviour of the ADR, which makes it completely useless for use in DMMs.

Frank 

[Dr. Frank]

Hello Midi,
I've removed the residual drift and the step from 26000sec onwards.
This Allan deviation looks much, much better, as expected.
The minimum StD is about 8ppb @ 256sec aperture.
You need to reduce on the single references noise by 1/SQRT(2), so it's 6ppb for the DMM.
that's the final proof of concept, and OpAmp noise must be much lower.
Therefore, TiNs 4 fold ADR with even twice as high zener current (20mA,, if I remember correctly) will lower this to 2ppb, i.e. we have here a genuine 9 Digit DMM!
Please let me know, what TiNs  group on XDEVS thinks, and if somebody already confiirmed my tests.
THX Frank

PS: I will remove again the ADR1000 from my 3458A, until I've built a battery backup, or added an additional transformer for the ADR

[Dr. Frank]

On 29th June, 0:30h I noticed a very near Thunder Strike with immediate Thunder. It really struck into the Earth only 2km away, By the EMP, or via mains, it permanently shifted 3 of my LTZ and my ADR references by several ppm. As well, one LM317 regulator suffered a latch-up and sat at zero volt output
Both my old prototypes, which didn't feature Andreas EMC capacitors yet, were affected the most.
I built them 20 years ago, as a starting point of all volt-nuttery.
I will add TVS diodes.

This was the perfect occasion to finally move the references including all precision resistors to proper PCBs with protecting output amplifier. This moved the output voltages by only  another + and -1ppm each.
To avoid any unnecessary drift, I cut them off the old PCB and have inserted them in golden precision sockets.
I had built the first decade of an 720A Kelvin Varley divider with these green molded VHP202Z, to have a 1V output, and also to calibrate the linearity of an old 332B. These VHP202Z performed excellent with low 0.2ppm/K and neglectable drift. 7V => 10V only had to be re calibrated 3 times in 25 years.

Here's my finished Tower of Precision. Please take notice of the guest references, the famous M69, which is modified to an ADR100 by His Excellency, John R. Pickering, and Adrian Bernhards ultra stable RS2-10k, based on the Vishay 10k VHP101T. Latter has drifted by only 2.2ppm in 7 years.

So I finally ordered 2 of them at DigiKey, which already have replaced my old VHP202Z resistors, which were a fail, as Vishay evidently made an error in production. Their TC were too high (old spec problem), but also they came to me having already drifted outside of their 5ppm tolerance. They also showed a mediocre -1ppm/yr. drift performance.

Frank

[Dr. Frank]

While I was changing my old Ref_1 and Ref_2 references, I reasoned, what's the essence and Progress in metrology of my original design Goals, which in the volt-nuts community is commonly known as the Dr. Frank Reference.

Andreas owns the Merit of having made a groundbreaking improvement on the EMC ruggedization of the LTZ1000 circuit, which in the end leads to the discovery of the noise barrier in 8digit DMMs, making it 9 digits from now on.

My design was driven by my work in the Automotive supplier Industry as a component Manager.
I reduced the whole circuit to the least necessary and cost effective components, i.e. single sided PCB, usage of Standard precision resistors (w/o any overengineered Special parts), stripping off any esoteric and unmeasurable oddities, like Slots, circular rings, cap, multilayer PCB.
The single sided PCB might have a positive physical effect that all solder Joints are on one layer, so that no temperature differences arise, apart from the warming over the wires. In practice, it has no measurable positive effect as well.

My design foresees a magnetic inner Shield, and a 2nd outer Shield providing proper Ground and Guard for improved emc Performance. It also foresees an inner thermal Shield so that all components stay on equal temperature. In DMMs, I also discovered the noise susceptibility created by the fan air draught, so my thermal design deletes this effect as well.

I initially had the Overall 1ppm design Goal for my lab: TC << 1ppm/, stability < 0.5ppm/yr, uncertainty < 1ppm
so I reversed back to the Weston cell era, where volt metrology was done on an odd 1.018V value, and using a KV divider to create 10V out of it.

My design implemented the first decade of a KV exactly for this purpose, so to completely eliminate the timely drift of the resistors, and making full use of the inertial physical stability of the Zener reference.

My design for the volt-nuts community includes a simple resistor divider w/o any special expensive parts, and making the raw 7.2V available at the outside.
This way, you could either use the 7.2V voltage directly with its ultimate stability, or with equivalent stability the 10.000V output for convenience by using any 6 digit DMM for ratio transfer, giving less than 0.5ppm uncertainty for 10V, due to the common low INL of any DMM, see also my original design goal of 1ppm uncertainty.
Later on, it turned out that these simple precision resistors already inherit low enough T.C. matching and timely drift from scratch, so that even this ratio transfer is obsolete.

My design also includes the observation, that no pre aging process is required. You simply bring up the LTZ1000 circuit, let it run for 3 months, and from then on it will only drift on its specific drift rate of typically -0.5ppm/yr. This drift rate is very predictable, to an absolute uncertainty of about 0.1ppm/year, as can be seen from my comparisons over 20 years!

That implies now, that everybody can simply build his own reference group for the BOM cost only, e.g. 4 x 200$, make a one time calibration to the recent 0.2ppm uncertainty, which I transferred from Philipp, and nobody needs to buy these expensive FLUKE references any more. All my DIY references can now be specified as:

TC < 0.02ppm/K
Stability < 0.5ppm/yr.
Uncertainty 0.2ppm   

You get for free a 3 fold stability number compared to Flukes 732C.

If you determine the drift rate of your specific LTZ1000, it will absolutely be predictable for its whole Lifetime, with an uncertainty of absolute 0.1ppm for these observed small drift variations.

So, after a one time calibration and a one time drift rate measurement, its specification can be completed to
Uncertainty 0.2 + 0.1ppm over Lifetime

If you have such an Ensemble of 4 references, comparing them against each other, i.e. man with 3 Clocks principle, you will be able to detect any irregular anomaly, and as in my Long termed Observation, it would even be possible to detect and mitigate this said variation of drift rate, probably by another order of magnitude.

This now implies, that in future there is no need any more for any regular Volt calibration for any Instrument!   

The zener references are from a Solid State Physics perspective intrinsically stable and predictable, and the metrologists have overlooked and wasted this outstanding property.

I will present  my observations to Mr. Palafox from the PTB. He should please assess for correct metrological practice and confirm the one time absolute traceability of my discovery.


A last philosophic view and why we all do this Metrology volt-nuttery stuff.

Everybody strives for an absolute fix point in his life, that is in most cases the partner, where you hopefully have a stable, long-termed relationship.
This is the case for me, as I'm still living together with my wife since 1990, coincidentally the very same year when I acquired my first HP3458A!

From my youth onwards, I always also strived for this absolute fixed point in voltage metrology, i.e. I always wanted to reach out for the absolutely stable voltage references.

So metrology has the same significance for me like this happy relationship to my wife.
This was always a striking argument when I asked my wife whether she allowed to buy a new instrument.
Maybe this is as well another solution for the volt-nuts community 

Anyhow, every development has come to an end, and this is obviously the end of my travel into volt-nuttery.
Let's see, what's next.     

[Dr. Frank]

Regarding this drift process which is constant over time:
In case of resistors it's always a decay process which will decrease or end in time.
Therefore it's true for resistors only, that they get better when they age.
This was never true for Zener references. Very early I observed, that they show a constant drift rate, and this old rumor , that old references in the 3458A would get better and all would be as good as the selected 2ppm/yrs ones, was a false interpretation of the known pre aging / annealing process for such decay mechanisms as of resistors.
Due to the extremely small crystal defects in the todays silicon technology, there will be no such decay process for zener diodes.

The zener drift is caused by it's pn diffusion process, which is absolutely constant in time, due to the very huge concentration difference of the dopants.
The rate is depending on the doping technology and the geometric structure, and maybe on one or two other parameters, which are all absolutely constant.

The zener voltage is absolutely fixed during the doping process, and given by its geometry. The diffusion changes constantly this geometry, therefore , changing this zener voltage.

It's similar to constants of nature, i.e. it's a constant of the nature of pn junctions.

In these highly doped zener structures, there is as well a certain portion of decaying processes. These explain this residual change in drift rate on the order of 0.1ppm/20 years.
Anyhow, the constant diffusion drift effect predominates latter by far.
Even these 0.1ppm/20years would not change this metrology finding qualitatively, but would only set a lower limit for the absolute uncertainty.

Frank

[ap]

Hi Frank,

many thanks for this! The observation re. noise on the LTZ being vastly reduced by those caps is extremely interesting. I would guess the same applies to the ADR, I will see...

Your above statement re. the constant drift rate absolutely makes sense to me, but my knowledge about semiconductor physics is limited, so I leave that to those who can better judge.
I would think however, that this applies after some initial settlement at a higher drift rate as is usually reported, caused by additional effects (die stress stabilization, ...). That would also make sense in the context of KS and its older statements re. low drift (2ppm) options of the 3458A reference, by just watching the initial changes over a shorter period of time and selecting those for that option that faster reach the stable drift rate.
This initially higher drift rate also seems to be the case for the ADR you use and reported about somewhere above.


cheers

Adrian

[Dr. Frank]

Hello Adrian,
many thanks for your valuable feedback.
You're the first and only one to react.

This is a pity, and I'd like to ask the community to frankly send your critical comments, hints, opinions, etc.

Thank you
Frank

[branadic]

Hello Frank,

because you asked for it, here is my opinion. I'm an electrical engineer with background in semiconductor fabrication, not a chemist or physicist.

I cannot comment on A9, because I do not have a 3458A myself, but I might do the experiment on my R6581D and/or Prema 6048 (white).
You have mentioned, that the reference in a 3458A is close to an oscillator and that some of the rf gets rectified to near DC, for which the caps might help. I would expect, that this results in a constant, repeatable oscillation in the DMM readings, but that is not what you have observed, but instead you see a sudden instability in the readings without the additional capacitors after some time x of continuous readings. How do you explain that?

What I have noticed in my experiments with the Prema 6048 (black) though, one of the early units with separate add on LTZ board, was quite some instability/oscillation of the reference board when removed from the unit and measured directly on a meter, something I could not solve even with a series resistor and/or a buffer at the zener output. With respect to the manual, the reference circuits is the basic datasheet implementation of the LTZ.

My random thoughts about drift:

1. It is known that creeping of polymeres (die attach) never stopps. However, given an environmental equilibrium, polymers will settle towards this equilibrium and will start to creep again, once that equilibrium is broken.
2. The die attach, a polymere, produces strain on the die. Although designed as a zener diode and not a strain gauge, this strain is still a rather prominent and measureable effect. Bending a die will cause piezoresistive effects as well, hence why silicon pressure transducers exist.
3. Compared to the LTZ, the doping profile in ADR is sharper, at least that is what Eric Modica mentioned at MM2021. From what I've learned, a doping profile leads to diffusion over time, that increases in speed with increasing temperature. The steeper the doping profile, the stronger the initial diffusion process, but diffusion will never end and is just slower at reduced temperatures due to lower activation energy.
4. Looking at the periodic table we see, that phosphor is in the 5th group and 3rd period, hence it is a larger atom compared to silicon (4th group and 3rd period), while boron is in the 3rd group and 2nd period, hence a way smaller atom compared to silicon and phosphor. So, we can expect that diffusion also leads to a change in strain distribution within the crystal lattice over time.
Note: The effect of a larger atom within a crystal latice resulting in strain is e.g. very well known from nitrogen-vacancy (NV) centers in diamonds - compare HPHT diamonds with CVD diamonds. A larger NV concentration produces more stain and leads to broader resonance peaks in the ODMR spectrum. Although we operate a zener reference at "near DC", I think the picture is applicable to some extend to a zener diode.

On first power on of an oven-controlled, temperature compensated zener, all the aforementioned effects are superimposed, with some of them settling towards an equilibrium, which might be interpreted as a decay, but some of them going on forever. But as far as I understand decay (vanish forever) != settling to equilibrium.
Once equilbrium is reached (oven temperature not changed anymore, polymeres have relaxed, reference constantly powered, ...), some of the aformentioned effects will still remain. I would expect that diffusion e.g. is a non-linear effect nevertheless, because the doping potential decreased over time (very very long timescales though), although the reservoir can be expected to be infinite on short timescales. But I would assume, "saturating" the space-charge region between the different dopants also changes on very long timescales.
However, on short timescales (years, decades, lifetime of a human), these non-linear effects might appear linear, but looking at very long timescales (centuries, millennia), they might still be exponential.

What I still struggle with to explain is random walk. Maybe that is some sort of "charge effect", that "discharges" if a certain threashold is reached.
On the other hand, the more effects one compensates - and we always apply very low order polynomials only, hence there is always some residue left - the more "not seen before" effects show up that were initially hidden. But in the end random walk (non-Gaussian behavior) could also be just the sum of all the residues coming from the compensations being applied.

-branadic-

[iMo]

3. Compared to the LTZ, the doping profile in ADR is sharper, at least that is what Eric Modica mentioned at MM2021. From what I've learned, a doping profile leads the diffusion over time, that increases in speed with increasing temperature. The steeper the doping profile, the stronger the initial diffusion process, but diffusion will never end, it's just slower at reduced temperatures..

The LTZ1000 might be produced with a more classic diffusion based process, while with the ADR1000 (ie the "steeper sharper profiles") they may use more modern ion-implantation based process. The ion-implantation is rather destructive process where you shoot the dopants into the lattice with high energies. You have to anneal the lattice afterwards at high temperatures. So there is a certain probability the ADR1000's lattice is not that fit compared to the LTZ1000's one.
But that might be a speculation as the processes used are not publicly known..

[tatel]

While I was changing my old Ref_1 and Ref_2 references, I reasoned, what's the essence and Progress in metrology of my original design Goals, which in the volt-nuts community is commonly known as the Dr. Frank Reference.
(...)

All these assertions look really impressive to me. Too bad LTZ1000(A)CH seems to be almost deprecated right now. Digikey has zero stock, perhaps you can order quantity 40 at about 36€ each. Mouser doesn't seem to be willing to send any to the EU (?) and Farnell says "available until stock is exhausted". They seem to have 22 LTZ1000CH in stock right now.

Taking into account that the much newer ADR1001 doesn't seem to have been experimented with, as thorougly as the venerable LTZ1000, i decided to order four while they are still available. I definitely don't need them right now, but I'm pretty sure that, after stock is exhausted, it will be much more difficult to get these, and prices will surely go higher. You know, just in case.

I like Dr. Frank's design, but, seeing people here ordering purpose made vishay metal canned resistors, or other high performance, high priced resistors for their LTZ1000 references, I wonder how these FLCY resistors, so much cheaper, could be so good. Surely they have been carefully selected from a much bigger group?

I would really appreciate any comments you guys could have to make about this matter.

[Dr. Frank]

The AE FLCY are very good, as they are lacquered.
they really have a low and equal t.c. over several samples and do not drift much over time.
Highly recommended, better than the PWW type.
Frank

[tatel]

It's not easy to get some of these FLCY resistors. I emailed alpha electronics Japan and France email addresses with no answer so far. However it looks i could have found a couple of sellers, one from Germany, and another one from Hong Kong. Since I could get some other, difficult to find part from the Hong Kong seller, I contacted him, and at least received an automated reply. We will ll see how it goes next week.

I want them for R12 and R13 and perhaps R4 and R5 (but could try a NOMCT16031001AT array instead if people here thinks it would be good enough)

For the remaining, non-critical or not so critical resistor, i'm planning to use trough hole parts. As I understand it, these could be 5-10 ppm TC, but long term stability would matter the most. So I  was thinking about PTF resistors but saw some negative comment about them. I'm unable to find any previous recommendation about it.

I would appreciate any advice people here could give about these non-critical resistors

[branadic]

Have you checked Rhopoint Components?

-branadic-

[tatel]

No, this is what i found in Europe:
https://de.lankamicro.com/

But thank you very much anyway, it's helpful to have another european resource.

About non-critical resistors: PTF56-PTF65 is what i think about if looking for long-term stability, little sensitivity to humidity, etc, and are available on 5-10 ppm/k 1%. I'm not sure about the cons (well, they aren't cheap nor easily available right now, it seems). Im also looking at other series from Ohmite (MOX700) and RESI (MMFR2568) that seem to have better specifications (like 5 ppm/k, 0.1% tolerance) and lower price.

Also I have seen specs like Yageo SMDs that seem to be like 25 ppm/K and 5% tolerance? Of course something like that would be easy to find, and cheap. But after reading about 5-10 ppm/K these specs seem a little bit too rough?

I'm pretty sure there's people here with quite a bit of experience that could give advice about this non-critical resistors for LTZ1000 matter. Any hints please?

[The Soulman]

On my ltz1000 build I've used the ptf56 resistors in critical and non critical places.
Digikey still has them in stock but prices wend up considerably.

I've used series and or parallel combinations of resistors to obtain the correct value, below includes the resistors for the 7->10v divider.
Stability has been great, temp-co could have been better but likely can tweaked a little. (used a none A ltz1000 without a optimized temp-co compensation resistor)

edit: those digikey prices are from February 2024, My ltz1000 is up and running 24/7 (with a few power-downs) since May 2024.