Ultra Precision Reference LTZ1000

[splin]

I've just measured my 3458A reference board (the only bit of a 3458A I own sadly); the voltage across the 111ohm zener current setting resistor and Q1 Vbe is 399mV which surprised me as I was expecting it to be nearer 600mV. Is this typical?

399mV means the zener current is 3.58mA; Q1's collector current is 89uA given the 74.2k resistor. 399mV Vbe @ 89uA Ic implies a reverse saturation current of approx 12E-12A which seems rather large - does it imply that Q1 may be damaged?

The reference seems to be stable enough with an ouput of 7.06V but I haven't tested it to any degree as yet. The board was powered at 18V and was drawing approx 33mA.

The reference is marked 1826-1860 with a date code of 9051. The voltage across the heater was 8V giving 264mW of heater power; room temperature was 20C so I'm not sure if it is an LTZ1000 or an LTZ1000A given that 264mW seems to be half way between the data sheet graphs @ 75C above ambient - so which is it likely to be?

[EDIT] Forgot to subtract the zener and LT1013 supply currents in the heater power calculation, so heater power actually nearer 230mW. That isn't far off the 200mW shown in the datasheet for an LTZ1000A @75C above ambient

[MisterDiodes]

Lars:
Try running an '5400 resistor pak for a voltage booster for the newer high-res 32 bit ADC's and you'll see what I mean.  Compare that to PWW.  You'll be doing more work on noise low-pass bandwidth reduction especially if you use these on an ADC input - which generally means you're pushing the noise down to a lower freq., (not really eliminates it over longer time frames)  That resistor noise spec by LT (and for Vishay films in general) is generally very optimistic, and when you press them on the issue - well it turns out those noise specs are more theoretical in nature, and probably measured on an un-mounted device (just sitting in a test jig or wafer prober, which is not -your- PCB and thermal flow).  That complete package of device mounted to a PCB changes the story.

Unless you are careful with the mount of the '5400 pak, you can easily build a board that picks up every air draft and thermal stress / vibration change in the neighborhood.  You'll have to test in your application - don't assume every spec is as-advertised.

We use PWW resistors from Pettis, GR etc. and have predicted results - if you're worried about humidity, you use a decent vacuum-deposited coating, hermetic enclosure, silica bag,  etc.   For us and what we need, in an industrial environment the PWW is a good, profitable & robust solution when dealing with vibration, sudden air drafts etc.  We send modules from low-humidity North America to the jungles of Thailand - plastic packages and all - and if you plan for it it's not a problem.  It does take planning though.

You'll find that humidity is more of a concern if you leave your box switched off for a long time - and it's certainly on a longer time scale to see effects (weeks).  Keep your precision devices in an always-on, slightly warm enclosure and humidity effects will be mitigated.   At least that's what we tend to see.

If a humidity test is called for we'll use a saturated salt solution to keep the device in a known, saturated atmosphere at known humidity over 3 ~ 6 months, etc. with power on and power off - and you'll learn more about your circuit that way.

If your application needed a smaller size, wasn't prone to vibration effects and you could live with more noise - then a '5400 pak could work for you if used very carefully.  Every situation needs an appropriate solution to meet the customer's needs.

[mimmus78]

I was thinking of do a 4 ltz1000 on one module.

PROS:
- compact PCB (more easier to hovenize)
- use only one 13K/1K divider
- feed the 70K and 12K/1K resistors with an averaged 7V (less drifty and less noisier) that maybe should even improve stability
- lower noise reference of 4 combined ltz1000
- slightly cheaper than 4 LTZ1000 modules

CONS:
- all your eggs are in one 12K/1K divider
- will be more hard to trim for tempco
- loose possibility to select the best LTZ1000 module
- single point of failure

Does anyone ever tried or want to discuss?

Inviato dal mio Nexus 6P utilizzando Tapatalk

[Andreas]

Hello,

From the last 4 built references I have
1 with large popcorn noise
1 with detectable popcorn noise
and only 2 which are without detectable popcorn noise.

From ageing drift there are also 2 good and 2 not so good devices.

So if you put all in 1 cirquit all will be bad.
With 4 different devices you will have the chance for 1 excellent device.

You need more samples so that you can select.

With best regards

Andreas

[Dr. Frank]

I've just measured my 3458A reference board (the only bit of a 3458A I own sadly); the voltage across the 111ohm zener current setting resistor and Q1 Vbe is 399mV which surprised me as I was expecting it to be nearer 600mV. Is this typical?

...

[EDIT] Forgot to subtract the zener and LT1013 supply currents in the heater power calculation, so heater power actually nearer 230mW. That isn't far off the 200mW shown in the datasheet for an LTZ1000A @75C above ambient

Surely is that typical for a diode, due to its -2mV/°C coefficient.

75°C above RT gives about -150mV lower forward voltage.

The rest of the currents also seems reasonable, so your reference is fine, very probably.

Frank

[Kleinstein]

With reference chips it is a good idea to still have access to the individual voltages. This way one can see if some parts are drifting more than others and also check if some have higher noise. Depending on the noise levels one might decide to give the better units a higher weight.

The transistor currents are not that critical, especially not for the noise. Also the average voltage would be available only after a buffer - so possible some extra drift / noise there. So having a lower noise voltage source for this gives essentially no improvement on the overall noise. The advantage of using the individual reference is that there will be no drift coupling, and even if one reference totally fails, the remaining ones would be still as good as before.  A similar thing applies to the 13K/1K divider - even if the drift of different units would compensate to a small degree this would still also work for the overall voltages.

A shared divider saves a little, but it also introduces a single point of failure and failure of one unit (for whatever reason) would cause a shift in all the temperatures. So I would definitely prefer separate dividers. Having 4 separate dividers also gives some averaging over there drift, so there is some (though not really much) return in precision for the extra costs.

There is still something to having the references all close together, as this can help to bring the lower side of the output voltages together.  For the positive circuit the ground (reference -) connections can be kind of tricky. It might need averaging resistors for this side too, even if they are small.

Compared to the standard circuit a few points might be worth a thought:
1) A way to measure the voltage of the temperature setting resistors might be useful. This might only take something like a jumper and output connector for optional access. Modern DMMs might be more stable than the divider.
2) As a short at the output can upset the heater, it might be a good idea to have a kind of supervision, to prevent excessive heating. This would be less of a problem with an averages reference as this essentially needs a buffer, but the separate outputs would be still sensitive.
3) The temperature regulation loop in nonlinear: heater power is proportional to the square of the current. Thus the quality of control depends on the power level and is getting lower at the low end. So it might be a good idea to make the loop at least a little more linear to allow good operation to lower levels of heater power. It would also reduce the upsets on transients, as power spikes would be limited.

[ap]

Lars:
Try running an '5400 resistor pak for a voltage booster for the newer high-res 32 bit ADC's and you'll see what I mean.  Compare that to PWW.  You'll be doing more work on noise low-pass bandwidth reduction especially if you use these on an ADC input - which generally means you're pushing the noise down to a lower freq., (not really eliminates it over longer time frames)  That resistor noise spec by LT (and for Vishay films in general) is generally very optimistic, and when you press them on the issue - well it turns out those noise specs are more theoretical in nature, and probably measured on an un-mounted device (just sitting in a test jig or wafer prober, which is not -your- PCB and thermal flow).  That complete package of device mounted to a PCB changes the story.

At the position where the LT5400 would be used (7 to 10V gain stage), noise can easily be filtered, also given the relatively high source impedance of that noise source. So that should not be a concern. For an A/D application, that is different, as this is usually broader in frequency band. Not so here, it is DC essentially.

[Kleinstein]

The excess noise of noisy resistors tend to have a significant 1/f contribution. So filtering is not really practical. I do not know the standard (and thus the bandwidth) by which it is measured, but -55 dB does not sound very low noise this would be something like a 12 µV of noise at 7 V. This sounds like a lot compared to something like 1.2 µV peak to peak or 0.2 µV RMS for the LTZ1000 and 0.1-10 Hz frequency band.

Filtering in the very low frequency band might be more in the semi digital domain: so using a low noise wire wound divider as a first approximation and if a second, divider might give a lower drift at the expense of higher noise, this could be measured in comparison to do correction an a very lang time scale (e.g. hours or longer),  maybe with something like a digital trimming.

[ap]

First of, the standard referenced (MIL 202, M308) in the data sheet is not covering the tradiational 1/f noise frequency range (which is from DC to some ten Hz). The frequency range covered is geometrically centered arround 1KHz and is 1kHz wide. Also, the -55dB figure, as per spec, is defined as the ratio of the DC voltage applied accross the resistor, measured in V, to a RMS noise voltage, measured in uV. So we are talking about appr. 13nV RMS excess noise here.

[splin]

I've just measured my 3458A reference board (the only bit of a 3458A I own sadly); the voltage across the 111ohm zener current setting resistor and Q1 Vbe is 399mV which surprised me as I was expecting it to be nearer 600mV. Is this typical?

...

[EDIT] Forgot to subtract the zener and LT1013 supply currents in the heater power calculation, so heater power actually nearer 230mW. That isn't far off the 200mW shown in the datasheet for an LTZ1000A @75C above ambient

Surely is that typical for a diode, due to its -2mV/°C coefficient.

75°C above RT gives about -150mV lower forward voltage.

The rest of the currents also seems reasonable, so your reference is fine, very probably.

Frank

Thank you - I don't know how I managed to overlook that.

Do you agree that the reference will be an LTZ1000A? Only I'd got the idea that originally HP used the LTZ1000 and by the time they switched to the LTZ1000A they'd stopped marking them with their own part number. I've very likely got this completely wrong though.

[Dr. Frank]

Thank you - I don't know how I managed to overlook that.

Do you agree that the reference will be an LTZ1000A? Only I'd got the idea that originally HP used the LTZ1000 and by the time they switched to the LTZ1000A they'd stopped marking them with their own part number. I've very likely got this completely wrong though.

It is an LTZ1000A, as far as I know, and nobody else found the non A version in any HP3458A references.

Frank

[MisterDiodes]

First of, the standard referenced (MIL 202, M308) in the data sheet is not covering the tradiational 1/f noise frequency range (which is from DC to some ten Hz). The frequency range covered is geometrically centered arround 1KHz and is 1kHz wide. Also, the -55dB figure, as per spec, is defined as the ratio of the DC voltage applied accross the resistor, measured in V, to a RMS noise voltage, measured in uV. So we are talking about appr. 13nV RMS excess noise here.

Therein lies the problem with Datasheets and sketchy noise specs - and trying to filter out the noise introduced by LT5400 into an otherwise precision circuit.

The 1/F noise is the problem - If you've ever tried this: You build a 7-10V voltage booster with a clever application of LT5400.  You realize it's a little too noisy for your ADC system. You low-pass filter your 10V output say to 5Hz, thinking you'll get rid of noise...Then you test everything and realize you got some noise energy at 1Hz and below.  So you dink around with more filtering, say 0.5Hz.  Now you realize you've got more noise down below .1Hz, and maybe even a bigger problem....and so on.  Then you realize what a rat-hole that 1/f noise is doing to you.

It is fairly difficult to get rid of that noise and have your Vref stable for measurements over longer and longer time frames.  Just speaking from experience here.

At some point you realize it was easier to just go with some good PWW's, or if your customer demands it a custom metal film ratio set for your boost amp divider.  It is always better to -avoid noise when you can- rather than try to remove it from a signal later. At least that's what we try for.

The next thing you realize is: Do you really need 10V for your application?    But that is a story for another day.

Couple of other "Gotcha's" on LT5400 paks - just a general head's up if you haven't used these before:

1)  The only thing going for them is pretty good TCR ratio matching.    The resistance absolute value (20%) and TCR is absolute garbage compared to real precision resistors - these are designed to only be used as ratio groups.  Be careful of that if your amplifier feedback circuit also depends on a total -absolute- value (including individual TCR of around 8ppm) of the divider as well as the divider -ratio- itself.

Note that the TCR ratio of LT5400 isn't anything that can't be done with good wire-matched PWW that are thermally tied together.  Or a good film resistor ratio set.  The advantage of using real resistors here is you also get good control over the absolute resistance value AND much better control over individual TCR AND much better mechanical stress forgiveness.  Your application may or may not need those characteristics.

2) The withstanding voltage of LT5400 is surprisingly low.  Remember these are IC style, small diffused resistors.  Be warned if you are using these as an input-signal divider; you might need surge protection -ahead- of the resistor pak.  These are not very forgiving of an over-voltage stress.

3)  These resistors - being small- have a surprising crosstalk especially on adjacent units.  Be aware of that if you have any signal with higher dv/dt or sensitive AC signals - like for instance if you're using these to build a differential / instrument amp, etc.  It's best to test on your own board before you believe anything on that datasheet.

4) Mounting a LT5400 is not a hand-soldering job.  For minimal package stress these want to be properly reflowed and watch out for solder paste thickness - if you can get them to "float" slightly off the PCB that can help.  The thermal pad is critical on the back side but remember this is a small capacitor to each resistor element also.  Lead-Free solder can introduce different stress characteristics over regular solder - make sure to test your board out mechanically so you know what the effects are.

Lots of things to watch out for with those 5400's.  They might be a good fit for some applications but for us they've always been somewhat disappointing and never helped the profit margin over other quality resistors.  Your mileage may vary.

They are smaller though. <Grin>

[ap]

I appreciate that you brougth up the issue with 1/f noise with the 5400, especially as it is nowhere mentioned in the data sheet, the noise spec in there is certainly not of help. Using the 5400, I will check on this in my application.

[branadic]

I wonder if there are spice noise models for the different resistor technologies available, so that you can make a theoretical investigation on how the resistors behave in a real circuit application with different opamps.

@ Edwin or any other: Do you have real noise measurements on your resistors? I don't like the universal statement "Wirewound resistors don't exibit 1/f noise". This might be true for the wire, but a resistor is more than just the wire. Thanks.
How do Edwins resistor compare to UPF, UPW and all the other (ultra-) precision resistors?

[Kleinstein]

There is an articlew with some measurements on resistor excess noise for some resistor types:
https://www.google.de/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwiXrOrfupfTAhXCvxQKHXGmBdgQFggoMAA&url=https%3A%2F%2Fdcc.ligo.org%2Fpublic%2F0002%2FT0900200%2F001%2Fcurrent_noise.pdf&usg=AFQjCNELohORkodPhFRGcCVkUfrvgjFiCw

[branadic]

Thanks Kleinstein. Sounds like a tradeoff between TC, noise and price of the resistor. RC55Y are cheap and tend to have low noise, as published by the linked article, but as Andreas found they have large aging drift.



Vishay S102K are more expensive, have low noise as indicated by the article, but we don't have measurements on real TC available.
For UPW50, Econister, UP805 (Edwin Pettis) and Z201 we have TC measurements made by Andreas, but no noise investigation. 

So I still hope that someone with 1k or 10k of the mentioned resistor types can measure noise to compare them to the publication to get rid of this lag.

[Kleinstein]

The excess noise of wire wound resistors is considered very low, so usually not a problem at all. The same is true for the foil resistors, even though the noise can be slightly higher. With thin film resistors it seems to depend on the type and also size. Some are not that bad, but some are. Even some carbon resistors seem to be better than there reputation.

A doubt there is much need to measure noise of WW resistors - this would be a rather difficult task. It might be interesting with thin film resistors like the LT5400.

Measuring the excess noise would need 4 approximately equal resistors and the suitable amplifier circuit and data collection part, somewhat similar to that used for low frequency. The LT5400 with 4 equal resistors might be a good candidate.

[branadic]

The noise data of RN73 (nichrome on ceramic) are looking also promissing, they are cheap and available in different packages (0405, 0603 and 0805). Would be interesting to see, how they compare in TC measurement and if there is a way of proper mounting, to get rid of CTE mismatch when mounting to pcb.

[martinr33]

Mounting is an interesting question. If you use a larger package - 0805 - you have more mass to overcome forces from the PCB. Also, slots at each end of the device, and under the device, will further reduce the ability of the PCB to strain the resistor.  An interesting question for the next version of reference voltage sources. 

[Andreas]

RC55Y are cheap and tend to have low noise, as published by the linked article, but as Andreas found they have large aging drift.

Hello,

I have only examined 1 sample of RC55Y. It could also be a "outlier" or "production fault".
(similar to USR2#1 sample on the table)
Even from the best resistors you get from time to time a "drifter".

But since T.C. was also >3 ppm/K (as expected from the data sheet with 15ppm/K)
 I didn´t want to spend my time on further samples for RCY55.

with best regards

Andreas

[mimmus78]

@Kleinstein @Andreas

Yes having 4 LTZ in one board is deftly a bad idea because you cannot separate the good
apples from bad ones and temperature compensation adjustments will be more difficult to do.

So my definitive project will be to design a voltage reference made of 4 LTZ1000 reference
modules + one mother board.

Each LTZ1000 module will have the classical LTZ1000 circuit (plus Datron heater caps) implemented
with two LT1006 and buffered output (LTC2057). I'm not planing to do any EMI mitigation on the
LTZ1000 modules.

I'll keep the idea of having the "single instance" of 12K/1K divider on the mother board.
This will save few cents and will permit also to experiment with pwm controlled temperature set
point (that is where I'd like to end with).

Motherboard other than temperature set point will do the averaging of the 4 LTZ modules.
Averaged output will be buffered again. Mother board will also implement the power supply stage,
that will lover voltage to 13V (coming form a battery bank or from another power supply).
A few components will be added on the last buffer and on the power supply stage to do a little bit of
EMI filtering.

That should keep me busy for a fair amount of time.

[CalMachine]

@Kleinstein @Andreas

Yes having 4 LTZ in one board is deftly a bad idea because you cannot separate the good
apples from bad ones and temperature compensation adjustments will be more difficult to do.

So my definitive project will be to design a voltage reference made of 4 LTZ1000 reference
modules + one mother board.

Each LTZ1000 module will have the classical LTZ1000 circuit (plus Datron heater caps) implemented
with two LT1006 and buffered output (LTC2057). I'm not planing to do any EMI mitigation on the
LTZ1000 modules.

I'll keep the idea of having the "single instance" of 12K/1K divider on the mother board.
This will save few cents and will permit also to experiment with pwm controlled temperature set
point (that is where I'd like to end with).

Motherboard other than temperature set point will do the averaging of the 4 LTZ modules.
Averaged output will be buffered again. Mother board will also implement the power supply stage,
that will lover voltage to 13V (coming form a battery bank or from another power supply).
A few components will be added on the last buffer and on the power supply stage to do a little bit of
EMI filtering.

That should keep me busy for a fair amount of time.

This is very similar to what I plan on doing soon.   

I also want to try to do something using the 2DW23X zeners.

[lukier]

So my definitive project will be to design a voltage reference made of 4 LTZ1000 reference
modules + one mother board.

I'm planning something similar, a poor-man's DIY mini Fluke 734A, four pluggable modules, each with 10V and 10K.

I've collected most of the critical parts in the last months:

• 4 x LTZ1000 from Digikey, critical resistors from hifi-szjxic eBay, Vishay S102K/RNC90, 13K, 1K, 69.8K and 120 Ohm
• other components of the classical circuit from DigiKey, Vishay Dale CMF resistors, LT1013ACN8, etc.
• Pomona 3770 low EMF jacks for the 10V DC output, buying these for 10K force/sence for all four was too expensive for the moment so I'll settle for cheapos, I guess one can always do lead reversal and offset compensation on the resistance
• 4 x VHP202Z from hifi-szjxic eBay as 10K standard

The overall idea is to design small 3458A like ref board with just LTZ1000 basic circuit, all through hole, just one SMD - unbelievably tiny LMT70YFQT temp sensor just under the LTZ1000 and the whole thing using two S5637-ND connectors to connect to module's base PCB and enclosed in some ABS 3D printed case to prevent air flow (and insulate a bit).

But my analog-fu is not good enough yet to go from there. Each module base board would carry this ref board + 7V->10V step up that I need to think about (DigilentMind's MCU with multiple PWM idea looks nice on paper, need to test it), the 10K reference resistor maybe with a small heater to keep it stable and then I was thinking about etching flat flex out of Pyralux to connect the signals from the board to the binding posts, like in 732B.

But there are even more things where I need to learn more and test various concepts, maybe the volt-nut elders here will be able to help:

• I wonder if I could use small signal latching relay, such as TQ2-L2-5VDC that is used on some of the Keithley scanner cards to switch the output between 7V and 10V to see the drift of the 7V->10V step up. With careful thermal planning and routing both lines similarly (placement, junctions) it shouldn't be an issue?
• need to find proven output protection - overcurrent mostly (plus maybe a spark gap like in 732B)
• the biggest issue of all - power. Ideally, instead of one common linear power supply in the chassis for all the modules I thought it would be neat if each module integrated a small battery, think Nikon DSLR batteries or similar, ready made and safe. The chassis would be only charging the modules (+ reading temp sensors etc). But to use a 2S LiPo I would need a DC-DC converter which isn't great. What about using some kind of Royer followed by a low noise LDO + loads of filtering (I already have BNX002-01) - need to prototype that and measure with my AM502 or maybe I'll build zlymex low noise amplifier
• how much insulation - I bought some Spacetherm offcuts so I might use that, but I've read here that overinsulating LTZ1000 is detrimental. In 732B they insulate the whole lot, but I guess it is because LTFLU doesn't have a heater and they heat the whole lot. Spacetherm still might come handy to insulate VHP202Z coupled to small heating resistor.

[TiN]

Sounds like enough work for next 3-5 years, that's for sure. LTZ1000-based circuit deceivingly simple at unprepared eye, but quickly morph into great time-vampire project, as you get your feet wet. And you can't escape it afterwards! I fell into this trap 4 years ago, and still no end. Memo for "don't-become-voltnut" members - do not build LTZ circuits 

Maybe you can use just one step-up 7V-10V module, instead of per module. That way you can save some BOM and time, as this circuit is equally difficult, if not more difficult than LTZ REF itself, if goal also not to compromise LTZ's own stability performance.

As for me, I'm tweaking PID for new bigger TEC box, which can fit VK5RC's KX reference. So far no good, need more work. Big chamber slows things a LOT.

[Andreas]

I'll keep the idea of having the "single instance" of 12K/1K divider on the mother board.
This will save few cents and will permit also to experiment with pwm controlled temperature set
point (that is where I'd like to end with).

Motherboard other than temperature set point will do the averaging of the 4 LTZ modules.

Hello,

I would make the option of placing the 12/1K divider also locally on the sub boards.
(in case there is some influence between the individual references over the set point).

I still do not get what is the idea behind a common heater set point.
Do you want to trim the output voltage by setting the temperature (around 50 ppm/K)

Putting all outputs to a averaged one is a good idea.
But how do you do the star ground concept + kelvin sensing so that the heater currents do not influence the output voltage?
Without individual (galvanic isolated) power supplies this will be a demanding task.

with best regards

Andreas