Ultra Precision Reference LTZ1000

[chickenHeadKnob]

@diligentMinds:

I am still digesting the references you have given, thank-you. One question to clarify: you state that epoxy packaged parts are still useable if the entire unit is in a hermetic (ovenised) package, does that also apply to FR4 ?  For the hobbyist, not having to chase down exotic circuit board material and rare parts would be a great benefit. I am pretty sure I can construct an aluminum can with teflon gaskets and purge/fill with dry nitrogen at a lesser cost.

Edit: sorry I see you already answered this, use the best stuff you can get. I need to do some experiments in a chamber just to confirm to myself how good or bad FR4 is.

[chickenHeadKnob]

Unless you use a package that has true hermetic sealing (borosilicate glass and Kovar steel, with the cover silver-soldered on), I'm not certain that you would be able to keep the humidity out.  Water vapor has a strange way of finding a path through most materials-- including most epoxies and most silicone compounds.  Like this (for example):

http://www.sinclairmfg.com/catalog/hybrid.html

The problem with the Kovar/Pyrex hermetic packages is that the Kovar leads make *great* thermocouples when in contact with copper-- this is *not* what we want-- (is there another way that is effective)?

I'll be the first to admit that I am *not* a mechanical engineer-- Is there someone on this forum that can speak to this?  How do we make a high-performance hermetic package using materials a hobbyist can obtain?  We want something that will stop water vapor from entering the package, and (secondarily) we want to mitigate the effects of barometric pressure changes.  (The temperature we will control actively).

I am way behind and trying to play catch-up hockey here, so my ideas are likely something you have already thought of and discarded. As I only possess a simple mind my solutions tend to the brutally  simple. Some ideas I have been toying with tonight:

• if moisture and leaky seals are the problem;  pressurize, force the leak in direction you want, separate input and purge valves, gage and top up when required. This is an old problem and old solution 
• if the PCB substrate is the problem; get rid of it!, go dead bug or as I call it air wire in the most critical section, use cadmium-tin solder
• if the PCB substrate is the problem; minimize it, I have some very thin double sided, its .014 mm thick and quite cheap. or use mylar flex PCB material, either one would need a rigid backing. By rigid backing I mean have the thin PCB material held loosely against an aluminum plate which has  a thin insulation layer, say mylar, with bolts that have foam washers and go through oversized holes in the PCB material. That would allow the PCB material planar freedom of movement /li] 
  

with all of the above I am also considering an oil bath (degassed,dewatered mineral or silicone) with pressurized nitrogen headspace, connections to outside through nitrogen so that you wouldn't have leaking oil.

[Hypernova]

Regarding the issue of board thermal conductivity, how about metal core PCBs? Just wondering out loud.

[branadic]

I'll be the first to admit that I am *not* a mechanical engineer-- Is there someone on this forum that can speak to this?  How do we make a high-performance hermetic package using materials a hobbyist can obtain?  We want something that will stop water vapor from entering the package, and (secondarily) we want to mitigate the effects of barometric pressure changes.  (The temperature we will control actively).

Well, there are a few possible ways. You could use second-hand double oven oxco's in the bay and reuse the package, buy hermetic packages from companies like Schott:

http://www.schott.com/epackaging/german/overview/products/gtms/index.html

or construct your own hermetic package using rf-metal cases, copper or brass sheet metal and feedthrough capacitors for the electrical contacts from inside to outside.
Avoid sealing materials such as silicon, this is not hermetic at all, no epoxy glue nor plastic. To reach hermetical sealing prefer soldering, welding and ceramic or glas as they are diffusion resistant, exactly what you want. I again would also recommend to use LTCC, the substrate you will find if you open for example an oscillator. This could be cheaper compared to your mentioned Rogers 6035HTC.
Dry the package and fill it with nitrogen before closing it.

[muvideo]

I'll be the first to admit that I am *not* a mechanical engineer-- Is there someone on this forum that can speak to this?  How do we make a high-performance hermetic package using materials a hobbyist can obtain?  We want something that will stop water vapor from entering the package, and (secondarily) we want to mitigate the effects of barometric pressure changes.  (The temperature we will control actively).

Well, there are a few possible ways. You could use second-hand double oven oxco's in the bay and reuse the package, buy hermetic packages from companies like Schott:

http://www.schott.com/epackaging/german/overview/products/gtms/index.html

or construct your own hermetic package using rf-metal cases, copper or brass sheet metal and feedthrough capacitors for the electrical contacts from inside to outside.
Avoid sealing materials such as silicon, this is not hermetic at all, no epoxy glue nor plastic. To reach hermetical sealing prefer soldering, welding and ceramic or glas as they are diffusion resistant, exactly what you want. I again would also recommand to use LTCC, the substrate you will find if you open for example an oscillator. This could be cheaper compared to your mentioned Rogers 6035HTC.
Dry the package and fill it with nitrogen before closing it.

Another option, for prototypes at least, is copper clad pcb board used as enclosure,
35um copper should be enough as moisture barrier.

As for sealing materials,  it's true that most polymers are unable to stop water.
Silicone is uneffective as barrier, it let pass water and also higher molecular weight
vapours,  we learnt the hard way, I had some headaches with white leds blackening,
and then understood that silicone dome is essentially transparent to most vapours.

The only exception I know is parylene, AFAIK it's used as conformal coating, and
is deposited using chemical vapour deposition, not easy to deal with.

I'm still wondering if there is any cheap way of pumping out the water, to keep the
umidity constant.

[quarks]

I have worked on the schematic a little more (if I go for a experimenting board, where I can change between different OP amps and Output options), but do not yet know if I will go that way.
I have implemented the X=solder-copper-junctions and jumpers to be able to either connect to Zener + Sense directly or after the chopper amp buffer or if I want to, to connect the KVDs (like in AN86 Figure C7).
I also think about guard rings (as descr. in att.) but do not yet know how to exactly do it with DipTrance. 

For the best possible results I will probably make a design without any jumpers/sockets. But I first want to find out what is the best possible solution.

Comments and/or more hints or are welcome

[babysitter]

I consider a big guard polygon sufficient, there is not a big need to count picoamps here and use guard rings which are seemingly a problem in DT. Compared to real picoampere applications where a leaked electron is gone when it is gone, the ZENER+ supply will provide the next one. I wouldnt put too much sub units (amplifiers) on the reference board; maybe you want the amplifiers behind a KVD, maybe not, and good housing is easier with a smaller PCB.

You can easily use an external buffer/amplifier circuit and measure the voltage difference from the well-known input coming from the LTZ to your output if it is external, an internal buffer hides the reference from you if you dont go for a test point. (My possible-problem, my buffer brings up a unknown value of 35-40 µV offset but hopefully stable.

[quarks]

The most sensitive point on this circuit is the base connection to the temperature sensor transistor on pin-6 of the LTZ1000.  You could create a guard-ring (which is just like all other traces in your PC design package) that surrounds all of the connections to this node (on all layers of the PC board).  The guard would be connected to the collector of that same transistor (pin-8 of the LTZ1000).  The rest of the circuitry will work just fine without any guarding. 

Very good, I will try the guard ring around pin-6.
What do you think about the guard copper pour on both layers connected to the case? Will this help if it is also inside the ring?

If you use higher quality PC material (the material I already suggested in a previous post, or the ceramic material already mentioned by others), then this guard ring could be left out-- *however* do *NOT* use water-soluble flux for soldering-- this leaves a residue behind that is very difficult to remove-- and it is conductive enough to cause problems with sensitive circuits such as these.

Does anyone know a PCB maker (pref. in Europe) offering PCBs with this material?
What is the best way and stuff to clean the board? I have tried several (IPA, Aceton, pure bio Ethanol), but I am not yet happy, if I look at it under my microscope after cleaning. Even "my special mix" for camera cleaning looks not quite perfect.

...LT1881, it seems that this part doesn't have sufficient common-mode [Vcm] down to ground on the Inputs...bias the LTZ1000 *must* work down to 0V on the inputs.

thanks I missed that (LT1881 is now off the list, already changed it in the schematic)

The LT1051A has excellent specs for this purpose-- and you can probably find original units in the ceramic package...
I'm guessing Minco-Tech will charge you ~$10-$20 for each one in small quantities.  ...
...I would use an LT1051A, because it has the best long-term (time) drift and temperature drift of all of the possibilities...
If your not going for the highest accuracy possible, then, what's the point?

That is the plan and I will try to do the best I can achieve (within my limits).
I will now first order the easy to get LTC1051CN8 package, so I can at least have a first look at it and compare to the LT1013 I already have.

Thanks again for sharing your knowledge

[babysitter]

Hi Quarks,

The guard ring: should be on both sides of the PCB, and be the first thing the guarded node sees, so no cooper pour inside the guardring. Dont just guard pin 6, but the whole node. I suppose you will not be able to discern if you lose anything when you don't guard but as it is free...

Advanced PCB substrate: Leiton HF provides rogers RF materials, will not be cheap. You might want to get a brick of teflon and press cooper nails in it and build your circuit from nail to nail.

For cleaning:  Run the populated board through the dishwasher gets rid of a lot. Or try to get your hands on dichlormethan (toxic!). I have Tetrahydrofuran as a plastics solvent at work, so not so healthy, if you want to give it a try and report back, i could bring you a sample.

You might want to do a hermetic housing for the reference circuit, suggestions:

Use the same tuner-style sheet metal boxes from otto schubert that I used. If you solder them all around you should have your hermetic seal and the sheet metal will spread heat if you use a double oven - cooler method.
To connect the power supply and extract the stable voltage and test points, use thru-hole capacitors - otto schubert or oppermann have them. Ceramics should block water better than silicone and plastics.

You could heat it up gently for a while - below 70°C for a extended period of time, and when sealing use cold air which contains low humidity and silicagel  to keep water vapour content out of the sealed box.

Include a heating resistor and a NTC inside the box, too, in case you want to double ovenize it one day or at least want to know the inside temperature.

[quarks]

Hello Hendrik,

yes I alredy planned to have the guard ring around all involved connected parts (similar i.e. see att. Datron/Wavetek board). That makes a ring around a "quite large" space. That is why I wonder, if I also should have the copper pour inside connected. 

If I can find a manufacturer with Rogers or similar material, I will hopefully not be shocked by the price. I already invested so much time and money, therefore it will probably not stop me to go for it. But we will see. 

About your dishwasher tip, I would not do that. I do not even like to use my ultrasonic cleaner, because I do not want the whole assembly ducked in the fluid.
About Tetrahydrofuran cleaner, that sounds like asking for cancer and I will stay away from it.

So far I have pre choosen ROSE housing like
http://www.reichelt.de/Rose-Aluform-Gehaeuse/ROSE-04-08-12-08/3/index.html?;ACTION=3;LA=5;ARTICLE=102470;GROUPID=5200;artnr=ROSE+04.08+12+08
It is not exactly hermetic housing but at least IP66 sealed.
But I also would like to see your solution, when we meet.

Temp meassurement inside the box is an option I have already thought of, but not yet implemented.

bye
quarks

[branadic]

So far I have pre choosen ROSE housing like
http://www.reichelt.de/Rose-Aluform-Gehaeuse/ROSE-04-08-12-08/3/index.html?;ACTION=3;LA=5;ARTICLE=102470;GROUPID=5200;artnr=ROSE+04.08+12+08
It is not exactly hermetic housing but at least IP66 sealed.
But I also would like to see your solution, when we meet.

Dear countyman, you better use TEKO 371 / 372 / 373 / 374 / 392 / 393 / 394 / 1680 / 16120 or 16160 instead of your IP66 case.

I clean my pcbs with IPA in a bag in a heated ultrasonic cleaner, that works fine for me.

[quarks]

Dear countyman, you better use TEKO 371 / 372 / 373 / 374 / 392 / 393 / 394 / 1680 / 16120 or 16160 instead of your IP66 case.

I clean my pcbs with IPA in a bag in a heated ultrasonic cleaner, that works fine for me.

Hello branadic,

thank you for your reply.
The TEKO could be good for inner shielding. But they only seem to be max. 28mm high (that is not enough, when I want to use my Burster resistors). For the outer box I want a decent + sealed one and the ROSE is so far overall the best I could find.

About your cleaning method, please confirm if I got it right.
You put the fully assembled board in a bag and then fill this bag with IPA and then put it in the heated ultrasonic cleaner.
What is your tip/method after that with the cleaned but "wet" assembled board?

bye
quarks

[robrenz]

Please excuse my butting in.  IMO Ultrasonic cleaning only dislodges the dirt and gets in suspension/disolved in the cleaning solution. You must thouroughly rinse after the ultrasonic with virgin solvent that is at the purity level you are after. And that rinse should be a spray rinse not a dip rinse.

[BravoV]

And that rinse should be a spray rinse not a dip rinse.

Cause the dissolved contaminant even dilluted, will form an even "coating" throughout the object ?

[branadic]

There are bigger rf cases available, not at Reichelt but from several other providers:

http://www.ukw-berichte.de/ukw-docs/bau-lit/gehause.html
http://www.pollin.de/shop/dt/NjI5OTM1OTk-/Bauelemente_Bauteile/Gehaeuse/Gehaeuse_5.html

You got me right, that is exactly the way I clean my circuits. The ultrasonic cleaner is filled with DI-water, sure. Avoid using acetone as not all parts and packages can withstand it. There are a few sorts of plastic that will get etched. I'm not sure how your Burster resistors will behave.
After cleaning I dry the board with compressed air, but you can also put it in an oven for a while.

I don't know a manufactor for LTCC or the mentioned Rogers 6035HTC in germany, but maybe the use of RO4003C is an alternative choice? There is a german guy who manufactures them for a fair budget www.pcb-devboards.de also in 1.6mm. I expect this material has lower moisture absorption compared to FR4 and the Tg is much better Tg>280°C DSC (IPC-TM-650 2.4.24).
Even if it is said to be not nessecary I would nevertheless provide the slots in the substrate. Do what ever you can to get minimal stress to the leads. You will find statements that quote that this package is not sensitive to stress, but I wouldn't count on that if you want the last ppm.

[quarks]

There are bigger rf cases available...

You got me right, that is exactly the way I clean my circuits...
After cleaning I dry the board with compressed air, but you can also put it in an oven for a while.

I don't know a manufactor for LTCC or the mentioned Rogers 6035HTC in germany, but maybe the use of RO4003C is an alternative choice? There is a german guy who manufactures them for a fair budget www.pcb-devboards.de also in 1.6mm. I expect this material has lower moisture absorption compared to FR4 and the Tg is much better Tg>280°C DSC (IPC-TM-650 2.4.24).
Even if it is said to be not nessecary I would nevertheless provide the slots in the substrate. Do what ever you can to get minimal stress to the leads. You will find statements that quote that this package is not sensitive to stress, but I wouldn't count on that if you want the last ppm.

Thanks again.

the inner shielding would be nice to have, but if it not fits in size, I will have to change my board size, outer housing connector positions and so on. Therefore I think I will start with the ROSE and get a Teko 392 which seems to just fit inside. 

good, I got the cleaning part right. For a naked board this sounds good to me. But when I order my boards, I will hopefully get a perfectly clean surface. And for my soldered fully populated one, I still feel uncomfortable.

I will contact your suggested manufacturer. I also found a note, that my local PCB maker can offer a so called "Teflon" Type board (he descr. it Keramik-Polymer for HF and names Rogers, Taconic). I will try to find out details about it next week.
If I can, I will still try to get Rogers RT/Duroid 6035HTC and also ask for your suggestion RO4003C.

About the slots in the board, I still do not know, how to do it in Diptrace (other than just drill holes around the LTZ like Hendrik did, but still interested to find out). If it is easy doable and does not make problems with the rest of the layout I am considering it.

bye
quarks

[branadic]

And for my soldered fully populated one, I still feel uncomfortabe.

You can do the cleaning procedure twice, but up to know I had absolute no problems. As said I heat the ultrasonic bath too.

About the slots in the board, I still do not know, how to do it in Diptrace

Just draw the slots in the dimension layer and make them as suggested by the pcb manufactor (normally >2mm for the milling tool), that's it.



https://www.eevblog.com/forum/projects/lm399-based-10-v-reference/msg219527/#msg219527

But why not use an "off the shelf" program such as Eagle Light or KiCad?

[quarks]

You can do the cleaning procedure twice, but up to know I had absolute no problems. As said I heat the ultrasonic bath too.

with a "cheaper" design I would just give it a try, but here

Just draw the slots in the dimension layer and make them as suggested by the pcb manufactor (normally >2mm for the milling tool), that's it.



https://www.eevblog.com/forum/projects/lm399-based-10-v-reference/msg219527/#msg219527

in principal I know what to do, but in Diptrace I just have not yet found out how to really do it.

Have you build the linked board? That really looks great  !!!
What have you used to make it?
How did you make the thermal isolation block, is it Teflon or plastic?

But why not use an "off the shelf" program such as Eagle Light or KiCad?

I have used Eagle many years ago and now tried Diptrace, which I already like better.
KiCad I have not yet tried. Have you compared these programms?

bye
quarks

[branadic]

Yes I have build the linked board twice, one for a friend and one for me. It was also cleaned in IPA with ultrasonic.
After first measurements I replaced the LT1001 by the CERDIP version, had the luck to find two of them in the bay.

I made a >300h thermal burn-in at 120° without power supply, it was then running a few weeks. I then assembled a heater on top of the resistors cause I only used 5ppm 0805 SMD resistors and it is now running a few weeks in the lab at 21.4°C and RH=32% with heated resistors. The reference is stable at 10.00179V-10.00180V there, measured with a 34401A the whole time. Still have no own >6.5 digit meter yet.

The board is normal black FR4, manufactured at pcb-devboards. The layout was inspired by the different pictures of boards with the LM399 in web.
The thermal isolation is made off PA6/6, two caps screwed together and filled with non-conductive foam, the slots are still free inside. It is just to keep airflow away. It's not that I have a use for this board as the voltage is also not common these days, but it was nice to see what can be reached with small amount of invest.

I've tried KiCad but decided to use Eagle Light as we have the "professional" version at work, so I can easy transfer bigger and multilayer designs at work and finish the layout there.

[babysitter]

@Quarks: There is a PCB distribution house right in your town, chances are good that you can ask them for Diptrace hints.
I can tell for sure as my wife was working there, spending a lot of time explaining customers (like me) how to do stuff with EAGLE, as that was the CAD system she was responsible for

[branadic]

I can tell for sure as my wife was working there, spending a lot of time explaining customers (like me) how to do stuff with EAGLE, as that was the CAD system she was responsible for

There is no need to discuss about the quality of Eagle, it's far away from professional even in the "professional" version, the libs are mostly buggy and wrong in layout, but if you have updated most of the libs it is as easy to use as MS Paint
I have made a lot of boards with Eagle up to now, from rf to high resolution pcb based sensors, high precision or simple analog stuff, test equimpent like my DC-500MHz FET probe etc. It is up to the user to be able to calculate impedance matching and things like that, but it's not always a disadvantage to be able to do that. It's good to have the basic layout skills.

[quarks]

@Quarks: There is a PCB distribution house right in your town, chances are good that you can ask them for Diptrace hints.

I already contacted them. They unfortunately do not seem to know Diptrace at all.

edit: with the kind help of fmaimon here is the solution http://www.diptrace.com/forum/viewtopic.php?f=4&t=8964

[babysitter]

Lets not start a hate-war-religion-thread about layout packages

Does any of you gentlemen seeing the need for burn-ins and big magic slots and guardrings and vhp202 resistors have a method to show or measure the improvements of those thingies against a easy "cheap-ass"approach as mine? How much better do you want to be ?

[saturation]

I wouldn't be surprised if your 732B actually is far less than 2ppm/year as its a much aged reference.  The real trick is measuring that because at least either the reference or your DMM will need calibration against a known standard in the calibration cycle to check the rate of drift of both.  From the factory the 732B is 0.8ppm in 3 months.  The real hard part in making a metrology setup is tracking the data over time to insure the stability is true.

It's not totally "easy" to make a voltage transfer standard that holds 2ppm per year-- but is can be done by the average engineer using average (and well known) approaches to the problem.  But-- (short of going out and buying a million-dollar Josephson Junction Array)-- to get below 2ppm per year takes a lot of study, hard work, patience, and help from your friends.

I already own a Fluke 732B, which (according to it's published specifications) will hold 2ppm (or less) per year.  So, I want something better.  I want *less* than 1ppm per year, and my goal is (somehow) to achieve 0.1ppm/year-- a lofty, but worthwhile goal (for me at least).  For most people, the Fluke 732B is "Good Enough"-- but not for me.

[babysitter]

The principle of calibration (how does a reference or DMM agree with a better one?) and the quest to distribute the SI representation is clear to me.

My cheapo-no-magic reference aka "V_bs" ( Volt babysitter) was compared to a volt-nuts 4 other references (including a LTZ in a 3458 which was used for the 24h, 31h, 1 Week and noise experiments) that haven't been calibrated relative to a cal-lab for a while but their behaviour is closely compared to each other regularly. (If one of them wants to go for a ride voltagically speaking, you wouldn't realise if it is alone, but it will visibly differ from its siblings!)

My V_bs  behaves closely like that cluster, how it differs in the long term I will see, got it back yesterdays from the first experiments. Now the V_bs is on its own for say one or three months...

What I meant with my last question is that it is increasingly getting harder to tell who is the reason for a deviation, the noise and drift of the V_bs is no longer easy to discern from that of the 3458A.

Also, in October iirc I will get a 34401A at work back from the Böblingen Cal lab of Agilent. This has not the greatest long-term-stability, but immediately after arriving it will be reasonable close to what they measured against their traceable references. Knowing the instruments error as noted by the lab, taking some readings from the babysitter volt abusing the 34401As gratuitous extra digits coming out of GPIB will give a cloud of datapoints for you know what... the V_Agilent vs. V_bs and later V_bs against V_volt-nut and also V_bs vs V_Quarks . 

But I have to tell, the main reason for the LTZ reference is the predecessor, a REF102-based 10V, did not comply with my minimum requirement: Measure it at work, stuff it in the backpack, take it home and back to work the next day, measure again, delta only at the last digit. This one does so, after travelling much more. Now I can trust it for testing the DMM at home.

Greetings,
Hendrik