MX Reference

[hwj-d]

Also make-up pads may be treated with who knows what.

They are used, but i washed them. 

No, joking aside, thanks for the hint. You are right. Something others must go there. A little cap of foam not touching other parts will help much better.

[MisterDiodes]

I explicitly recommend Andreas design, especially the additional capacitors in parallel to each base-emitter diodes of the LTZ1000s transistors, that's C11 and C12 in his schematic.
It's definitely not true, that these capacitors affect the stability of the circuit, the opposite is correct.

I have added these two capacitors to my original prototype design from 2009 (it's in principle the datasheet circuit).
That greatly improved the EMI suppression, and the circuit shows no longer such spikes, even when a switch mode P.S.U. is present directly near the box.
Before that, the RF shifted the ovens set point and in turn the reference voltage (reversible change)

I think I might have been misunderstood and didn't explain very well - apologies:  When I mentioned that caps can disturb the stability of the circuit, I should have said "degrades the stability of the LTZ die".

So:  If you're going to run the LTZ circuit exposed to noise: on a bare exposed PCB without of a proper shielded enclosure - in a way that it was never intended  - I can see that the extra caps on the LTZ pins would be a band-aid type of fix to hide some of those noise glitches.  The problem with that is that it can increase the magnitude of current noise -across the LTZ die.  The concern is the extra caps have provided a low-impedance reservoir of charge to allow higher LTZ peak noise current flow in response to external noise energy. Excess LTZ noise current is a direct contributor to degrading long term (> 5yr) stability of the LTZ die substrate, or cause the die to never quite achieve it's most relaxed, lowest energy crystalline state.  The LTZ zener makes it's own noise for sure, but you -really- don't want to do anything to add to that.

If you're running a switcher power supply next to the LTZ circuit and those caps made the glitches disappear - you've certainly hidden the voltage glitches, but the LTZ die still is exposed to the current noise, which is not recommended. This is the same reason LT explicitly does not recommend a '2057 current driver with it's very high input current spikes injected onto the LTZ die if you're not careful.

Before Andreas complains, I know: The circuit will still work with AZ amp and extra caps of course - but if you want best long term stability for decades adding caps and/or AZ amps is not the recommended best practice.  If you've tried every other way and the caps are the only solution for your application, then that's what you have to do...But there is probably something else wrong, like inadequate shielded enclosure or too much noise nearby in the lab.  We would never, ever run a switcher power supply next a sensitive circuit during a measure, we always go to battery power for serious measures - Period.  But that's how we have to do it for many more reasons than operating an LTZ.

Again - I've built several hundred compact boards, no slots or crop circles, and they go into well shielded enclosures, LTZ-A version soldered directly down onto the board, air-draft covered,  and no real stability issues for decades.  A working LTZ circuit will not generate random large output spikes on it's own.  The circuit -itself- should NOT need those extra caps if you build it correctly, as per LT recommendations.  They are the manufacturer and have a good source of long-term feedback data working with every major test equipment manufacturer - so I rely on their experience along with what I've witnessed for decades.  That's all I can offer.

If you find the extra caps are required for your application, then you do what is required.  For me I'd find the root cause of the noise first and deal with it at that level - remove the noise source or increase LTZ circuit shielding as required, quit making the board too big with long antenna traces, etc, etc, etc.  That way you know you're protecting the LTZ die as much as possible from injected noise energy.  You -never- want to throw caps at something like an LTZ to try to hide a voltage glitch that shouldn't be there if a better method is available.  The "let's throw capacitors at the circuit until the output noise goes way"  approach should be last line of defense for LTZ circuits, and usually indicates a bigger problem with the board design or enclosure - or lack of enclosure.

For sure, do what you need to that best works for your application - everyone has different needs.

[Kleinstein]

At least in the simulation an extra capacitor at the transistor base to emitter does reduce the phase margin quite it bit. The circuit seems to be still stable, but less tolerant to capacitive loading at the output.

The more appropriate way is to keep RF noise away from the circuit, with something like shielding and filtering (e.g. ferrites) of the supply and output. For magnetic shielding the material for transformer cores is also rather effective - not as soft as MuMetall, but much cheaper and higher saturation. There should be no need for a super low field, mainly a voiding large variations. For annealing MuMetal one does not need pure hydrogen - often an argon hydrogen mixture is used, that is hardly flammable. As an other option there is amorphous / noncrystalline material with similar magnetic properties, but less sensitive to bending (and much more spring like).

I don't think the rather small current noise and even the current peaks at the input of an AZ OP would be a problem for the LTZ1000 chip. These are just minute changes and thus should not be enough to really disturb the silicon. I won't be too much concerned about the silicon itself. A point much more susceptible to aging is the glue to hold the chip. Also the bond-wires and thin metal layers are more susceptible.

The more problematic cases are large transients, when turning on a circuit that is only marginally stable. Also connecting a capacitive load can be problem. Ringing of the output voltage could easily stress the transistor from too much base current. Also excursions of the temperature could be a problem - the quadratic heater curve could cause rather high power spikes.

[MisterDiodes]

I don't think the rather small current noise and even the current peaks at the input of an AZ OP would be a problem for the LTZ1000 chip. These are just minute changes and thus should not be enough to really disturb the silicon. I won't be too much concerned about the silicon itself.

You'd be surprised at how little current noise it takes over long term to alter the lattice strain level in the substrate to affect long term ppm-level aging effects on analog circuits.  That's what we're concerned about here. Even some added pA pulsing on a die the size of a LTZ will have an effect long term and you can even measure the strain changes even after a year acoustically and with atomic force 'scopes - that's why proper circuit shielding is -very- important for the best chance at max performance long term.  Of course temperature cycling plays a big role.  The wire bonds and epoxy have some ageing effect but the crystal lattice strain changes completely overwhelm those.

Some of my clients are building devices with 25yr / 35yr and 50yr guaranteed specs, so yes there's a LOT of very expensive research going into this problem.

What I can promise you:  Best practice==> Don't just hide the injected current noise, prevent noise from getting to LTZ in the first place.  As much as practical at least.

[Dr. Frank]

I'd like to capture this thread, to publish in a quick 'n dirty manner my latest design.. it's a compressed version of my ~ 10 years old prototype, which works very fine.

The schematic is nothing spectacular, it's the original schematic @ 45°C, plus Andreas blocking capacitors. The buffer amplifier is an ICL 7650, or a LT1052, as displayed in the board drawing (these have a different connection for the chopper caps). It provides a buffered 10V output, and the direct LTZ voltage output, which can accessed better at +Ref_buf, to protect the reference circuit from unlatching the oven control.


The smart things are:
- the size, i.e. 50x80mm, so that the whole PCB fits into a tuner box
- single sided, so that all thermal junctions are on one side, and virtually at the same temperature.
- The layout is optimized for thermal symmetry, and all reference resistors and supplies are strung rigorously towards the two star points, -Ref and  +Ref.
The whole PCB will be thermally isolated by styrofoam inside the tuner box, the LTZ itself gets an additional foam cap
- all components are  through hole, only film capacitors, and one wire  bridge only required
- PCB design accepts both PWW and bulk metal film resistors (from AE, 0.2" / 5.08mm). VHP resistors will fit with a bit of bending of their leads.
- Pomona jacks will be assembled directly into the tuner box, which serves as Guard. Later, an outer case, connected to power ground, encloses the tuner box, and the jacks will be mounted directly through both case plates.
- All PWW resistors are T.C. characterized, and my 5 sets are matched to have a calculational over all T.C. of about -0.045ppm/K. That allows to trim the overall T.C. to zero by R10 ~ 820k. (200k ..1M)
- The first reference made with Andreas design has been trimmed to estimated ~ 0.02ppm/K and will serve as a bootstrap reference for further T.C. trimming, as the 3458A is not stable enough.
- as the LTZ reference voltage is fixed, 10V can be calibrated to a few ppm uncertainty by using any 6 1/2 digits  bench  DMM with ratio function, as the linearity should be sufficient.
- The T.C of the 10V output will be minimized by T.C. pairing of the 10K / 4k or 15k / 5k6 resistors.
Then, the residual T.C. will be characterized by means of the precision NTC, so that a correction can later be determined, for any internal temperature.
- 4 boards fit tightly on a standard Europe Card, so I etched and drilled the boards myself.

Frank

Edit 21.4.18: Updated schematic for LTC1052, C7, C8 tied to GND

[MisterDiodes]

I know it's not quite done - but don't forget a little access hole for your 10V pot!

[ManateeMafia]

Dr. Frank,

Thanks for sharing. I originally tried for that form factor but I wanted it to fit in an extruded enclosure which resulted in much less space along the two edges that fit in the slots. I agree with the dual resistor option for whatever parts are available.
I look forward to your build and testing of the multiple references.

You are probably correct on the high initial drift of my reference. There could have been excessive heat applied at soldering. I hope this one isn't a wash but it certainly won't stop me from building more. I sense my Ultrohm Plus resistors are not far from being finished and I would like to be ready with the next version. All options will be considered and I will add places for the extra film caps but I won't initially install them. They will at least be optional and won't need to be bodged into the board.

[Dr. Frank]

I think I might have been misunderstood and didn't explain very well - apologies:  When I mentioned that caps can disturb the stability of the circuit, I should have said "degrades the stability of the LTZ die".

So:  If you're going to run the LTZ circuit exposed to noise: on a bare exposed PCB without of a proper shielded enclosure - in a way that it was never intended  - I can see that the extra caps on the LTZ pins would be a band-aid type of fix to hide some of those noise glitches.  The problem with that is that it can increase the magnitude of current noise -across the LTZ die.  The concern is the extra caps have provided a low-impedance reservoir of charge to allow higher LTZ peak noise current flow in response to external noise energy. Excess LTZ noise current is a direct contributor to degrading long term (> 5yr) stability of the LTZ die substrate, or cause the die to never quite achieve it's most relaxed, lowest energy crystalline state.  The LTZ zener makes it's own noise for sure, but you -really- don't want to do anything to add to that.

If you're running a switcher power supply next to the LTZ circuit and those caps made the glitches disappear - you've certainly hidden the voltage glitches, but the LTZ die still is exposed to the current noise, which is not recommended. This is the same reason LT explicitly does not recommend a '2057 current driver with it's very high input current spikes injected onto the LTZ die if you're not careful.

Before Andreas complains, I know: The circuit will still work with AZ amp and extra caps of course - but if you want best long term stability for decades adding caps and/or AZ amps is not the recommended best practice.  If you've tried every other way and the caps are the only solution for your application, then that's what you have to do...But there is probably something else wrong, like inadequate shielded enclosure or too much noise nearby in the lab.  We would never, ever run a switcher power supply next a sensitive circuit during a measure, we always go to battery power for serious measures - Period.  But that's how we have to do it for many more reasons than operating an LTZ.

Again - I've built several hundred compact boards, no slots or crop circles, and they go into well shielded enclosures, LTZ-A version soldered directly down onto the board, air-draft covered,  and no real stability issues for decades.  A working LTZ circuit will not generate random large output spikes on it's own.  The circuit -itself- should NOT need those extra caps if you build it correctly, as per LT recommendations.  They are the manufacturer and have a good source of long-term feedback data working with every major test equipment manufacturer - so I rely on their experience along with what I've witnessed for decades.  That's all I can offer.

If you find the extra caps are required for your application, then you do what is required.  For me I'd find the root cause of the noise first and deal with it at that level - remove the noise source or increase LTZ circuit shielding as required, quit making the board too big with long antenna traces, etc, etc, etc.  That way you know you're protecting the LTZ die as much as possible from injected noise energy.  You -never- want to throw caps at something like an LTZ to try to hide a voltage glitch that shouldn't be there if a better method is available.  The "let's throw capacitors at the circuit until the output noise goes way"  approach should be last line of defense for LTZ circuits, and usually indicates a bigger problem with the board design or enclosure - or lack of enclosure.

For sure, do what you need to that best works for your application - everyone has different needs.

MisterDiodes,
that noise / die stress problem you have explained several times, I find it being a quite an interesting solid state physics aspect.

You're fully right, about avoiding or shielding external noise as much as possible.. I already tried that by moving the whole metrology stuff in our genuine German basement .. solid concrete walls and ceiling, with a lot of steel mesh inside, banned any switch mode P.S.U. in the lab, and the temperature change is very small, w/o any air con, which might produce mains glitches.
The prototype double reference has a single shielding box only, and it's running on mains by an external DC supply.

But anyhow, I still observed some glitches in the LTZ output, during 24h monitoring, which will very probably be caused by spikes in the mains supply, which can't be avoided in an ordinary building.
The LTZ circuit is also used in other mains powered applications, like in the 3458A, the Kei 2002, or in the Datron instruments 1281, 4910, and these are also exposed to these kinds of disturbances, w/o obvious penalty on their performance.

So it's too easy that you call for pure battery operation, or for avoidance of any external disturbance, all that is not practical nor realistic.

I assume that the double shielding / guarding, like in Flukes calibrators, or in higher grade bench DMMs, will improve the E.M.I. behavior. (But I'm still missing these shielded / guarded transformers.)


Anyhow, I still can not fully comprehend the magnitude of this degradation effect on the LTZ die by external spikes, in relation to the expected / typical -0.8ppm/year drift @ 45°C oven temperature. I have the gut feeling, that this might be a minor effect only, especially over long observation times of 5 years or more.

Please, would you mind to give some more details about the magnitude of this effect, and how you have experimentally determined / extracted this effect and its magnitude apart from the natural ageing by temperature.
Maybe you also have more practical recipes, how to avoid / shield external spikes.


Thanks a lot - Frank

[Dr. Frank]

Dr. Frank,

Thanks for sharing. I originally tried for that form factor but I wanted it to fit in an extruded enclosure which resulted in much less space along the two edges that fit in the slots. I agree with the dual resistor option for whatever parts are available.
I look forward to your build and testing of the multiple references.

You are probably correct on the high initial drift of my reference. There could have been excessive heat applied at soldering. I hope this one isn't a wash but it certainly won't stop me from building more. I sense my Ultrohm Plus resistors are not far from being finished and I would like to be ready with the next version. All options will be considered and I will add places for the extra film caps but I won't initially install them. They will at least be optional and won't need to be bodged into the board.

At last, I recommend most urgently, that you also implement a buffer amplifier behind the LTZ output.
I encountered oven unlatching (to >100°C) on my prototype, during some experiments and in regular use, which caused severe sifts of the  output voltage.
I could bring it back within a few ppm of the original value by temperature cycling, but afterwards it took over a year to reach the same low annual drift again.
Therefore, this is the most important protection measure for this circuit.

Frank

[SvanGool]

@Dr. Frank:

Finally a "volt-nut" application for my Altoids boxes (Maxim article)
Your PCB would fit it exactly

[Theboel]

@ dr Frank,
I really interested about double shielding / guarding, like in Flukes calibrators but I have doubt about what exactly You talking about, can you tell / show a little bit about it.

[lukier]

@ dr Frank,
I really interested about double shielding / guarding, like in Flukes calibrators but I have doubt about what exactly You talking about, can you tell / show a little bit about it.

Check Dr Frank's article on Fluke 5442A:
http://www.amplifier.cd/Test_Equipment/other/Fluke5442A.html

The whole case is mains earth shielded, but the inner cage (with the analog bits), mounted on a plastic insulated spacers, is grounded to the analog ground of the inguard floating power supply AFAIR.


The guarded transformed also has shields in between the primary and secondary.

[d-smes]

@Dr. Frank-
Does your current regulator have any transients or ringing at turn-on?  I ask because of the C9 100nF loading on the output of the current regulator op-amp.  If the LT1013 open-loop output impedance is ~10 Ohms (unspecified, but not unreasonable), this plus the dynamic impedance of the 1N4148 form a pole with C9 that gives rise to considerable phase shift near the loop crossover point.
Edit-
Opps... I overlooked C3 being at 22nF (some LT schematics show 2nF) which brings your unity-gain crossover down to 20 kHz or so; well below where excess phase delay could lead to instability.

[3roomlab]

The smart things are:
- single sided, so that all thermal junctions are on one side, and virtually at the same temperature.
- The layout is optimized for thermal symmetry, and all reference resistors and supplies are strung rigorously towards the two star points, -Ref and  +Ref.

wouldnt guard traces be important to the uA level traces ? and since traces have no mask? or it doesnt matter as long it is cleaned and sealed?

[MisterDiodes]

Dr Frank:

To avoid getting in trouble with a customer -  and how bypass caps can affect crystal lattice aging I will have to leave it at:  The solid state physics theory that academia teaches is the first layer, and the reality layer of actually what happens in a production situation is something on a completely different plane.  This is an area where theory ends and and reality begins.  I am not trying to avoid your question but I am not at liberty to go into extreme detail in public.

I want to say overall you did a good job on your board - you have the LT1013 in there on a compact board, star + - Ref and avoided SMT caps !!  Those are the very best techniques for low ppm!!

Lets just say:  Do everything you can for shielding that LTZ FIRST and only resort to adding extra bypass caps if there is no other way for your lab.  Maybe you have hit the "No Other Way" point but here are some thoughts I -can- freely share with you.

1)  A working LTZ circuit (on a compact board) on it's own does not generate those spikes spontaneously, I promise.  I think you have figured out this is from an external event, and that probably means your circuit is not adequately shielded.

2) RE: Double shield boxes - Yes you are learning why all that "old boat anchor" equipment is designed with an inner steel box + outer steel box.  Your basement and your RF tuner box, Altoids boxes etc. aren't really the best for shielding against low to mid freq noise, but multiple steel layers and perhaps MuMetal is what we use as required.  You are still prone to H-field problems and junk on the power lines in your basement.  The big problem (for low ppm)  with any switcher bench power supply is somewhere in there I guarantee there is a "Y" capacitor that does a great job of delivering power line spikes directly to it's output - and I'm highly suspicious that you might be seeing that.

Not all sheet steel is the same, you can experiment with different alloys and thickness for best effect.  You need something of much more substantial thickness than a candy tin.

3) Outer steel box is ground, then teflon spacers, then inner steel box as Guard.   If required you add also MuMetal depending on what freq you're shielding against.  We have steel tables with a MuMetal sheet (to help prevent UN-intentional coupling transformers with wires laying on the steel) and and anti-stat on top of that just as a direct metal insulator, but use what works for your situation.

4) We use an voltage internal regulator just ahead of the LTZ circuit.  Learn from '732 and buffer the output.  Keep your input power leads quiet and choked / bypassed where they enter the in-guard box.

5) Keep all your test leads short, twisted and shielded.  Even in your initial test box you want to keep your output leads twisted.  You've soldered your leads onto the binding posts, but I know that's your fist test unit.  Normally you use a crimped copper + gold spade or ring lug on the inside connection of the binding post for better thermal performance and to keep the system balance with the wire or spades connected on the front side.

In short:  Look at any piece of old but good precision gear and learn from that.  You'll do the same.  No ventilation slots, keep the inner and outer boxes well isolated except maybe at one point, star grounds, keep any and all leads going into and out of the box choked, protected  and / or buffered. All of that -really- helps when you're at low ppm.

6) Battery power not practical?  We have whole rooms that are battery powered, or have a 24VDC / 48VDC and battery powered 24VAC distribution bus.  If you HAVE to have mains power, we use a 1:1 toroid with high-isolation windings.  Transformer winding shield is part of the guard.  This will help if you place it ahead of your 3458a.  Build a smaller version for powering your smaller circuits.

7) Everywhere keep an eye open for current loops that encompass any surface area.  Every one of those loops becomes an H-field antenna and EVERY P-N junction in your circuit becomes a demodulator.  That includes the leads on your desk - keep them short and twisted.

Once you try all of those things, then you might add the extra LTZ caps - but for maximum performance and the best odds at a successful aging die you want to try to leave those out. 

[Kleinstein]

....

wouldn't guard traces be important to the uA level traces ? and since traces have no mask? or it doesnt matter as long it is cleaned and sealed?

The currents in the LTZ1000 circuit are not that small. Even the collector current of the transistors is at around 100 µA, and these nodes are not that sensitive to small voltage variations. So there should be no need for guard traces. The heating and thus usually slightly elevated temperature keeps humidity and thus leakage low too.

[Dr. Frank]

The smart things are:
- single sided, so that all thermal junctions are on one side, and virtually at the same temperature.
- The layout is optimized for thermal symmetry, and all reference resistors and supplies are strung rigorously towards the two star points, -Ref and  +Ref.

wouldnt guard traces be important to the uA level traces ? and since traces have no mask? or it doesnt matter as long it is cleaned and sealed?

guard traces are required at about pA of signal currents, the LTZ circuit has 100 uA to 4mA, so that's not necessary at all.

In complement, you'll need proper star points for the reference voltages, to mitigate or to avoid any voltage drops caused by unwanted consumers, like the oven current (20..40mA), and the OpAmp supply.
Therefore the negative supply for the LT1013 is so strangely bend around the neg reference star point.

Equivalently, the temperature distribution should be symmetric. As the LTZ1000 also generates some heat on the board, the two reference star points are equally designed, regarding shape and distance from the LTZ solder junctions. Due to equal heat transport, both star points should be on the same temperature, and therefore their solder joints generate the same, but opposite thermo-couple voltage - hopefully.

Also worth to mention, that the signal paths with highest currents are closest to the LTZ, and are the shortest, so first comes the 120 Ohm resistor with 4mA, then the 12k/1k divider with 500uA is next, then the 70k / 100uA.

Hope, that explains the details of my layout.

Frank

[Andreas]

For sure, do what you need to that best works for your application - everyone has different needs.

Hello,

these are true words.

But it is an illusion that you can keep out all EMI-sources even with the best metal housings
if you have a wide open door i.e. the reference negative line.
(it makes no difference if the positive reference line is buffered or not).

The best chokes have around 10 dB EMI reduction. (perhaps up to 20 dB if you have luck and the right frequency).
A capacitor easyly dampens > 20 dB if the wires are kept short.

Of course it is good to do some measures to keep out the EMI-noise (thats why I have a double shielded design + battery supply).
But it gets perverse if I have to switch off all other devices in my house and the neighbour is sitting in 2 m distance with his mobile phone separated only by a normal stone wall. Or he is using his "green" switchmode power supplies because linear wall warts are de facto prohibited.

So If you can switch off all EMI-sources during a measurement, can use shielded transformers which are hard to get in single quantities then its the best solution for you. But I fear most hobby volt-nuts do not have a shielded EMI-cabinet with multiple stage filters for the power supply.

So my needs are clear:
- I do not want to have different output values in my "lab" independent of the gear that I or my neighbour is using.
- I also do not want to have different readings when I take my references into a unknown environment e.g. for calibration.

This is an area where theory ends and and reality begins. 

So what do we have:
- no official documents from LT (no application note, no design note)
- no official published paper
- no measurement values nor a test setup to verify the story

On the other side:
I have recorded the ageing from my first 2 references over more than 6 years
there is only unusual ageing against other published papers of LTZ1000
if I short the output of the unbuffered reference.
-> Shi(f)t happens if you are working.
So what you describe (if it really exists) can only be a 2nd order effect.

with best regards

Andreas

[Kleinstein]

AFAIK chokes (ferrite beads) can have quite a good effect on EMI. Though it usually take both a choke and a capacitor of some kind together. EMI filtering can be tricky and sometimes works not as good as one hopes for - especially at a few bad frequencies.

At least the simulations show that adding capacitance at the transistors (base to emitter) reduces the phase margin. Together with capacitive loading this could go all the way to oscillation. So at least for frequencies in the 100 kHz-1MHz range the extra caps would do more harm than good. There already is a capacitance from collector to emitter that does slow down the transistor. A way to dampen RF signals would be more like adding a RC series combination in parallel to the existing 22 nF (or 2 nF) capacitor - within reasonable values this is more like an improvement on loop stability. If the circuit is not to sensitive to capacitive loading this also allows for better filtering at the output.

It should be possible to limit excursions of the heater for the case of a short at the output. It would not fully prevent anything bad, but could limit the temperature excursion to maybe 10 or 20 K. So the possible damage from failure in the heater loop would be limited.

[Andreas]

At least the simulations show that adding capacitance at the transistors (base to emitter) reduces the phase margin. Together with capacitive loading this could go all the way to oscillation.

no.

you have overlooked that loop stability is maintained by a additional 10K (in series to negative OP-Amp input)  + 100nF across the Op-Amp
in Franks and my cirquit.

Opposite to your assumption the simulation shows more overshoot with the original AN86 cirquit than with Franks design.
Of course the original AN86 cirquit will oscillate with a load capacitance of more than some nF.

with best regards

Andreas

[3roomlab]

You need something of much more substantial thickness than a candy tin.

would 3mm be considered thick in your lab?

[MisterDiodes]

Yes, 1mm ~ 3mm would be a good place to start.  We use .05 or .0625" as a starting place if you're on imperial measurements.  The inner box can be slightly thinner, and maybe a single box is all you need.  We normally go for a 1/4" spacing between inner and outer but 1/8" would probably be OK.  It depends on what you're shielding from.  Aluminum will help some against E-fields but not much for mag field issues.

Also note:  If you're using PWW resistors it's important to at least have some minimal shielding around your circuit if you're near stronger H-fields - say an industrial area where you've got power mains in conduits, motors, ovens, etc.  Those resistors are wound with balanced windings but they will see some magnetic interference effects in a strong field unless you give them at least some protection - otherwise they will give you low noise performance.

[Andreas]

would 3mm be considered thick in your lab?

Hello,

it all depends on frequency.

Shielding seems to be another dark chapter in knowledge.
You have to distinguish between electrical field and magnetic field.
For a low frequency (in near field) electrical field a thin foil (capacitive shield) is all you need for shielding.

For a magnetic field the skin effect is the mayor mechanism for shielding.
(the induced current gives a counter field which eliminates the original field).
So a magnetic shield is much more demanding than a electrical (low frequency field).

the formulas are given here for example for some materials:
http://www.chemandy.com/calculators/skin-effect-calculator.htm

Additional you have to know that all cirquits which are low ohmic
(below 377 Ohms which is the impedance of "free air" in far field)
are more sensitive to magnetic fields. (Our references are usually below 10 Ohms.)
At higher frequencies also a electrical field generates a magnetic component -> again the magnetic part is the more critical.


now you can calculate:
eg at 1 MHz and copper the skin effect depth is 65 um (that is where the field has gone down by 1/e or to around 36%)
So for a effective shielding you need a multiple (3-5) of this value to get a shielding by around 20-40 dB.
More than that is usually not doable with a practical housing because you cannot completely live without any openings for connectors etc.

If the frequency goes up by a factor of 100 the shielding can be a factor 10 thinner.
At 10kHz you will need 3-5 * 650 um or 2-3 mm with copper.

If you take a magnetic material a my of 50 (around that of tinned can) will give you together with a worse conductivity a useful thicknes of around 0.5 mm at 10kHz.
(you can get sheets of tinned steel e.g. at RS-Components but shurely there are cheaper sources).
http://de.rs-online.com/web/p/stahlblech/0682472/

If you have to go lower in frequency (eg 50-60 Hz) the thickness of the shield has to be increased if you do not want to use Mu-Metal.

The mayority of EMI emissions is in the 50-200 MHz range.
Simply because the connected antennas (power line cords) are in the 1-2 m length. (and have a resonance there as lambda/4 antenna).
Lightning has a typical current pulse of 8/20 us which corresponds to a main frequency at about 25kHz.

I am planning on adding a 3D printed cover for the board. Is there any benefit with using a conductive coating on it similar to how Keithley did their 2001/2002 design? MG Super Shield comes to mind but I wouldn't use it unless it is worth the investment.

you can calculate by yourself considering that any coating has a very bad conductivity against pure metal and the effective metal thickness is very thin.
So for a electrical / capacitive shield you can get good results if you have no other metal shield.

But for frequencies which are annoying its a waste of money.

with best regards

Andreas

[MisterDiodes]

For sure, do what you need to that best works for your application - everyone has different needs.

Hello,

these are true words.

But it is an illusion that you can keep out all EMI-sources even with the best metal housings
if you have a wide open door i.e. the reference negative line.
(it makes no difference if the positive reference line is buffered or not).

The best chokes have around 10 dB EMI reduction. (perhaps up to 20 dB if you have luck and the right frequency).
A capacitor easyly dampens > 20 dB if the wires are kept short.

So If you can switch off all EMI-sources during a measurement, can use shielded transformers which are hard to get in single quantities then its the best solution for you. But I fear most hobby volt-nuts do not have a shielded EMI-cabinet with multiple stage filters for the power supply.

So my needs are clear:
- I do not want to have different output values in my "lab" independent of the gear that I or my neighbour is using.
- I also do not want to have different readings when I take my references into a unknown environment e.g. for calibration.

This is an area where theory ends and and reality begins. 

So what do we have:
- no official documents from LT (no application note, no design note)
- no official published paper
- no measurement values nor a test setup to verify the story

On the other side:
I have recorded the ageing from my first 2 references over more than 6 years
there is only unusual ageing against other published papers of LTZ1000
if I short the output of the unbuffered reference.
-> Shi(f)t happens if you are working.
So what you describe (if it really exists) can only be a 2nd order effect.

with best regards

Andreas

Andreas:

Have you ever heard of an NDA?  Non Disclosure Agreement.  You have to sign those when you work in the industry.  That's why I can't go into extreme detail here.  I am not a hobbyist, I make my living making sure high performance semiconductor devices perform to spec long after we're all long gone and forgotten.  From manufacturing to packaging to final application.

At this end we've been helping our customers build longevity into devices for over 30 (35?) years, across hundreds of LTZ's for our own custom test gear - and that know how also goes into billions of devices you use every day  - and if you build the LTZ circuit correctly and shield it correctly it will work without glitches, because that is how they have always worked for us - at least that's how they should work in spirit and in application.

Extra caps or AZ amps are not required nor desired, but you do what you need to do. I apologize that I can't share detailed data publicly regarding advanced die aging processes because that R&D is at the -great- expense to our customers, and I've already taken some heat trying to offer aid here on a public forum.  Obviously I should not have done that in your eyes.  Obviously the manufacturer doesn't know what they are doing either after millions of devices and centuries of man-hours testing - so why should anyone take LT's advice over yours?

Most of what I have learned comes from working with LT and National Semiconductor and HP/Agilent/Keysight and Avago and Broadcom and TI and General Instrument and Siliconix and Fairchild and xxx, xxx, xx directly over the years, person to person...not a datasheet, not a college textbook written by some inexperienced PhD who never built a PN junction from sand in the industry.  It comes from working hand's on directly with the people in the lab: learning from the current device processes + customer test feedback to make the next process even better.  If that sounds like some story...OK, I understand.

Even if you don't believe me, have some faith there is a great deal going on inside the LTZ die (and similar structures) that you're not aware of .  There are several interesting proprietary methods to measure internal current-induced strain / stress waves traveling across the die... and how that affects the die crystal physics down the road.    All I can offer you in a very condensed, non-datasheet rule of thumb for buried zeners:  If you want the die to be as stable as possible 5 / 10 / 20 yrs from now, then keep the current flow as quiet as possible.  To that end good engineering practice doesn't call for adding low-impedance caps or AZ amps across a buried zener structure IF there are other methods available.  As I pointed out before:  If there is no other way to add appropriate sheilding then add the caps! Add the AZ amps!! Whatever you like!!

Of course you can't control your neighbors, so that might limit what you can do for low ppm measures.

My intent is only to pass along some tid-bit information on how it's done in the real world.  EEVblog is a great forum for that.  I try to give you what little I can share here, and believe me or not - there is a ton of information here but that is about 1/10 of 1% of the iceberg tip of some of the strange effects we see when you start working directly with the die.

I understand a hobbyist has budget restraints, but that means you get creative and learn.  Take an old box with non-working gear, strip out the guts and make that a Vref housing for testing.  Learn how to track down EMI and defend against it.   Wind your own transformer for instance if you can't buy one - or remove one from some other piece of gear.  Learn how balanced chokes work and you can build those also - if you learn what/where your EMI is coming from you can build a very good defense for it, and usually at not a lot of cost.  You can wind wire around a form, right?   OF COURSE you can filter/ choke the positive AND negative power leads!!  OF COURSE buffering the output helps keep out noise.  Even better: for noisy environments deliver a differential output by driving both the + and - output lines, , use FORCE and SENSE techniques, etc.   At least that's been our long experience and all of those things work for us when required. 

In other words, if you're going to build a Vref like a 732a/ b, then build it like you mean business for low ppm - within the means of your budget of course.  Do what you have to, but I wouldn't give advice for circuit changes that might not be best approach until more robust shielding methods are tried.

Over n' Out.  Have fun folks!  Sorry if I've offended anyone here with advice &  tips that has kept our equipment running to spec for some 30+ years, including LTZ's.

[dr.diesel]

Over n' Out.  Have fun folks!  Sorry if I've offended anyone here with advice &  tips that has kept our equipment running to spec for some 30+ years, including LTZ's.

Keep bringing it, we value your input, many thanks for all the time spent!