Ultra Precision Reference LTZ1000

[RikV]

Tj138waterboy,
Due to my age, I may have bad eyes, but on xDev site I only find a schematic in .pdf format but no real project (ECAD-)documentation nor Gerber files. Could you point me to where these links are? Please?

[Tj138waterboy]

Unless im mistaken TIN is the only one to design and complete the FX version and probably would be the only person with the files but I reread previous threads and found this info. It appears you would be better off getting one of the many oshpark pcbs as the FX is multilayer.

Nope, due to high cost only 14pcs of these modules will exist in the world, sorry.

KX one is available on OSH Park for PCBs : https://oshpark.com/shared_projects/HfKcqjV3
(LTZ chips you can buy from ADI and resistors from Edwin Pettis here or VPG).

[RikV]

Is the quote in your reply from TiN himself? this surprises me since he advised someone in this thread to build the Fx instead of KX since it was a "better design", in his own words.
Being 4-layer is no problem and not so excessively expensive in regard to the other components on the board. Anyway, OSHpark is not the only company to produce multilayers in small quentities and by far not the cheapest. But thanks for the effort.

[r6502]

@RikV

Have a look here:
xDevs.com KX LTZ1000-based DC Voltage reference design

somewhere in the middle of the article, there you can find what you're looking for: Gerber files packed in *.rar format.

Have fun to build it.

Guido

[syau]

@RikV

Have a look here:
xDevs.com KX LTZ1000-based DC Voltage reference design

somewhere in the middle of the article, there you can find what you're looking for: Gerber files packed in *.rar format.

Have fun to build it.

Guido

Think he is looking for KF gerber file instead of the KX

[alex-sh]

anyone know where to source ltz1000ach nowadays without 100 weeks lead time?

Yes, it is a disaster.
..

Could it be they are going to discontinue the LTZ1000? They source vendors till the stock is empty, the vendors are changing their designs to the ADR1000 or ADR1001 already. It has no sense to manufacture three almost identical references, especially the ones in metal..

Did a Mouser back order on 22 Oct, received 11 Jan.

This is pure luck.
Most metal can vref and precision resistors lead time is months.
I am trying to source Vishay VHD divider 10k/27.2k - lead time is 47 weeks!

[syau]

anyone know where to source ltz1000ach nowadays without 100 weeks lead time?

Yes, it is a disaster.
..

Could it be they are going to discontinue the LTZ1000? They source vendors till the stock is empty, the vendors are changing their designs to the ADR1000 or ADR1001 already. It has no sense to manufacture three almost identical references, especially the ones in metal..

Did a Mouser back order on 22 Oct, received 11 Jan.

This is pure luck.
Most metal can vref and precision resistors lead time is months.
I am trying to source Vishay VHD divider 10k/27.2k - lead time is 47 weeks!

Did ordered some VHD and VHP in Oct 2021 using express service and received in Mar 2022

Price of VHD was around USD$40
Price of VHP was around USD$50~$80

[alex-sh]

anyone know where to source ltz1000ach nowadays without 100 weeks lead time?

Yes, it is a disaster.
..

Could it be they are going to discontinue the LTZ1000? They source vendors till the stock is empty, the vendors are changing their designs to the ADR1000 or ADR1001 already. It has no sense to manufacture three almost identical references, especially the ones in metal..

Did a Mouser back order on 22 Oct, received 11 Jan.

This is pure luck.
Most metal can vref and precision resistors lead time is months.
I am trying to source Vishay VHD divider 10k/27.2k - lead time is 47 weeks!

Did ordered some VHD and VHP in Oct 2021 using express service and received in Mar 2022

Price of VHD was around USD$40
Price of VHP was around USD$50~$80

Try to order the same resistors now lol 

[dietert1]

Aside from the rant:
When building a LTZ1000 there is no need for those unobtanium resistors. As far as i understand the most critical resistor is the 120 Ohm one that determines zener current. As the typical zener differential resistance of a LTZ1000 is about 7 Ohm, there is a suppression factor 1/17. And the voltage on that resistor is about 0.6 V, another factor 11 lower than the 7 V zener voltage.
Total suppression factor should be about 1/180. There were some tables posted with these suppression factors, also for the other resistors, but i don't remember in which thread.
If you use a 5 ppm/K resistor like UPW25, the extra TC caused by the resistor is 0.03 ppm/K max. Congratulations if your reference build achieves a TC that low. One can also characterize 5 ppm resistors to select some with a zero TC temperature in the 20 to 40 °C range. And put the board into a hermetic enclosure with desiccant..

Regards, Dieter

[Kleinstein]

The most critical resistors are the 2 resistors to set the oven temperature. These only matter as a ratio and this can take into account some of the dirft (e.g. the square part of the TC that is often relatively consistent for resistors of the same series / type). Also the use of a resistor array is possible and attractive here. In theory one could have the option to measure the actual ratio (the set voltage) and this way at least detect long term drift if it happens for these 2 resistors.

The attenuation factors are good enough that the TC is usually not that relevant. The limited temperature range also leads to usually better than nominal TC for most resistors.
The need for good resistors is because of the drift with time. Data on the long term drift are however rare and not very reliable - this makes the choice difficult.
Looking at better than 10 ppm/K resistors is mainly in the hope that the lower TC resistors also have better long term stability.

[dietert1]

If somebody wants long term stability, they should probably avoid custom made parts as those will be new. Better get parts that have been on shelf for some years.

I think a good LTZ1000 build should incorporate tuning of the LTZ1000 TC using an oven temperature scan and some means of amplifying or attenuating the amount of TC compensation by the transistor. Usually amplification is required. One method is a voltage divider made of a 10 or 20 Ohm resistor in series with the zener and that 120 Ohm resistor.  This can raise the compensation from "native" -2 mV/K / 7V = -285 ppm/K to -300 ppm/K or whatever is needed.
After this kind of fine tuning the oven stability becomes less critical. When building an array it should even be possible to use one external oven for all references.

Regards, Dieter

[dietert1]

In an LTZ1000 or ADR1000 reference one might parallel the two transistors in order to improve noise. One can still measure temperature from Ube. Anybody tried this?

Regards, Dieter

[MK]

In an LTZ1000 or ADR1000 reference one might parallel the two transistors in order to improve noise. One can still measure temperature from Ube. Anybody tried this?

Regards, Dieter

Isn't most of the noise coming from the zener itself?

[dietert1]

There is only one zener in the LTZ1000, but two transistors.

Regards, Dieter

[MK]

That is irrelevant, the transistors are going to be ~2-5nV root hz, the zener about 50-100nVroot hz, the transistor noise is almost irrelevant in any of the sensible designs in use

[dietert1]

So, you did not try and you won't try.

Regards, Dieter

[magic]

Measure noise at pin 4, compare against pin 3.
If it's less, there is no point even thinking about it.

Something to consider:
LT1007 runs its input stage transistors at 120μA each and achieves 60nVpp typical noise voltage over 0.1~10Hz.

[dietert1]

With the notorious 70K resistor the transistor in the LTZ1000 has a collector current of about 93 uA, so the base current will be less than 1 uA. Don't know what this means for the noise.
Anyway, with a reference one would not so much worry about noise but about long term stability. My understanding was the very low transistor currents were to protect the device from aging. But maybe that was fantasy.
Also i remember an argument about whether one should use a capacitor across the zener like proposed in the ADR1000 datasheet or whether it can cause accelerated aging of the zener.

Regards, Dieter

[Kleinstein]

100 µA of collector current is a reasonable current to get low noise (e.g. 2-5 nV/root Hz range) from a BJT. The base current is only adding current noise, not helping with less voltage noise.
So there is a good chance to get rather low noise from the transitor with the standard 70 K resistors. This is at least for the white noise part, but the LT1007 example shows that also 1/f noise can be pretty low with a comparable current. For the comparison the actual transistor size may also matter and it is not that easy to compare.
100 µA of collector current is actually not that small. It is a perfectly OK choice for an 2N3904 used as an amplifier.


I don't see very much chance for aging from a little more current. The transistor operates with a very low voltage (collector at base level and thus some 0.6 V only). So no chance for realy hot electrons.
The current density is still pretty low to expect anthing of electromigration in the aluminum.

[aronake]

-	Datasheet	Lymex	Janaf
R1	1	0.14	-0.14
R2	0.3	0.4	-0.4
R3	0.2	0.03	-0.07
R4/R5	1.0	0.95	+1.2

-   Datasheet   Lymex    Janaf       Aron
R1       1.0      0.14        -0.14     -0.15
R2       0.3      0.4          -0.4       -0.45
R3       0.2      0.03         -0.07    -0.065
R4/R5  1.0      0.95        +1.2       0.95

Adding my sensitivity measurements on this classic table. Done with base of reference design.

Some added observations

R1       Higher senstivity at lower ohm. -0.17 at 100 ohm
R2       Higher senstivity at lower ohm. -0.53 at 55K ohm
R3       Higher senstivity at lower ohm. -0.078 at 55K ohm
R4/R5   Lower senstivity at higher ratio. 0.66 at 15. 0.99 at 12.

[Dunckx]

Mouser currently have 1153 LTZ1000A...for how long?

[Martinn]

Hi all,

having read through a good part of this thread, I'd like to build a few (1-3, depending on cost) LTZ1000 references as well. I don't usually do precision stuff, best I have right now is a recent DMM6500 6.5 digit bench DMM.
What I'd like to do is
- build 1-3 units (depends on resistor cost)
- let them run a few 100 hours
- bring them to the local cal lab for a precise measurement (and repeat that say every year cost permitting).

I have a few questions! It seems that the peak of this activity was the massive group buy https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg1837169/#msg1837169 of branadic's board in September 2018. I hope there's still someone around...

- Are the design files of this group buy (schematic/PCB in eagle for KiCad format) somewhere available?

- How useful in real life is the 10 V output (besides aesthetics and the extra callenge)? It seems this adds complexity, extra error sources (amplifier) and potentially lots of extra drift (boost divider).
Let's say I want to measure an unknown 5 V reference. I can either hook up the LTZ1000 7 V to the DMM6500 first to see where it stands or I can connect both unknown 5 V and LTZ 7 V to the DMM in ratio mode (not sure what's better).

- How important is an output buffer to the zener voltage? I'd connect mostly DMM inputs (10M) to the output and also the cal lab probably will do that. A buffer adds error sources and cost. AN86 https://www.analog.com/media/en/technical-documentation/application-notes/an86f.pdf p. 46 lists an LTC1150/LT1010 as possible buffer, but I don't think I need the large current drive and would feel better without chopper artifacts.

- What power supply scheme are you using? The branadic circuit has an LT1763 ultra low noise LDO on board, generating +12 V. As the supply would run continously, an abandoned linear lab supply is probably excessive. Switch mode wall wart too noisy. So a small mains transformer/rectifier might be OK (12 V rectified giving some 18 V)? Or pre-regulate to maybe 14 V to reduce power in the LTZ box?

- What about ground, earth and shielding? Guarding seems not necessary as everything is low impedance. A mains supply with unshielded transformer would AC couple the mains voltage to the output, so a 1 nF...1 uF capacitor from supply GND to Earth (PE) would make sense. This also rules out 2 pin supplies, as they don't have a PE connector. What about the LTZ1000 box itself? Connect to PE (how)? Connect (via capacitor) to supply GND? Anodized aluminium is a pretty good isolator, did anyone bother trying to properly contact the anodized housing parts?

- Precision resistors: I asked Edwin and he might be able to supply a set. Apart from that, are there any realistic sources? 70 k and 12.5 k seem pretty exotic values. I think this is the main killer for the entire build.

Thanks! Martin

[Andreas]

Hello,

just some experiences / thoughts.
- you will need some kind of buffer (otherwise in case of a accidential (even very short) short cirquit the reference will age very fast and need some months to recover).
- I recommend either a ADA4522-1 or a LTC2057 as buffer (LT1010 is not needed).
- I would build at least 3 devices so you can decide which one is the worst of the group.
- few 100 hours is not always sufficient. I have 2 samples of LTZ which needed over one year to stabilize below 2 ppm/year.
- 10V output is useful for devices with artifact calibration like 3458A
- I am using battery supply (14-18V) and LT1763 (14V). The LT1763 should be far away from the rest of the cirquilt to improve PSRR.
- with the batteries completely within the metal housing the housing can be used as guard.
- keeping power losses constant helps to improve stability of the reference.
- the special resistor values are usually built to order.
- Besides Edwins I have also ordered Vishay from Powertron/Teltow or built from series connected UPW50 stock values (RS).

with best regards

Andreas

[Martinn]

- I am using battery supply (14-18V) and LT1763 (14V). The LT1763 should be far away from the rest of the cirquilt to improve PSRR.

Just found https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg874560/#msg874560 this table where you listed the battery currents as around 20 mA.
So with 2 Ah batteries this give you around 100 h of runtime - 4 days. How do you handle that? Replace batteries every three days? Float charge the eneloop NiMH? Regularly top up the charge?

[Andreas]

Hello,

I use a 24V switchmode supply with a analog cc/cv (200mA/17.4V after diode) for charging (not during measurements).
100 hrs is a bit optimistic as the current is up to 30mA with buffer and the cells are
neither completely full at float voltage nor completely empty at under voltage switchoff (13.8V)

So I calculate maximum 50 hours operating time for a fresh set of batteries.

with best regards

Andreas