Stability of 5V references compared to AD586LQ

[Andreas]

Hello,

I widely use battery powered ADCs (LTC2400 based) with AD586LQ (temperature compensated) as work horses for my measurements.
The reason why I rely on them can be seen in the attached diagram: ADC13, ADC15, ADC16 are built with AD586LQ.
ADC17 is based on LT1236AILS8-5 which is typically about a factor of 2 less stable.
And even the LM399 based 6.5 digit DMMs (the K2000 with additional temperature compensation) are not so stable when reading my most stable 7V reference LTZ#4.

Since the AD586LQ is now near unobtanium (in small quantities) I am looking for a replacement.
Of course a heated reference could do the job but I want to operate the systems battery based so power consumption of the whole cirquit should stay below 5 mA.

with best regards

Andreas

[Andreas]

Hello,

first try is with AD586MNZ (the same reference in DIP8 plastic package)


I built ADC29 + ADC30 with AD586MNZ for testing.
Note: I am measuring the LTZ#4 at the input of the LTC2400.
The 5V reference is at the reference input.
So if the reference drifts to positive values the readings of the LTZ#4 on the input of the ADC go down.

Since the reference is still not pre-aged much I measured the average ageing of the references
ADC29:  +0.517 ppm/kHr
ADC30:  +0.366 ppm/kHr

Unfortunately the plastic package of the reference suffers from humidity sensitivity
(only the GND pin is soldered to the PCB the others are connected with thin wires so the PCB has no influence)

ADC29: +0.41 ppm / %rH with a time constant of ~2.61 days
ADC30: +0.42 ppm / %rH with a time constant of ~2.13 days

The diagrams show the readings from day 0 of measuremen in red.
The "prediction values" based on ageing and %rH in green.
And the filtered %rH values which are measured all 15 minutes and PT1-filtered by a time constant of 2 days or 3 days in blue.

As next candidate I want to evaluate a REF7050 in LS8-package. (still waiting for the PCBs)

with best regards

Andreas

[deepfryed]

MAX6350 seems better on paper at least, though only available in the plastic DIP packages. Also the DIP versions are EOL, but still available from Digikey.

[The Soulman]

ADC17 is based on LT1236AILS8-5 which is typically about a factor of 2 less stable.

I would expect those to be better than that, also after the first month of use? Do you see correlation with temperature or humidity?

[zepto]

What about LT1027? Advantest used it in their AD7461A which allowed for integration periods up to 10 seconds.

[Andreas]

Hello,

MAX6350 seems better on paper at least, though only available in the plastic DIP packages. Also the DIP versions are EOL, but still available from Digikey.

MAX6350 has a large ageing drift (initially ~10 ppm/year for the first 5-7 years). After 10 years I now have ~5 ppm/year which is still much higher than AD586 (< 1 ppm/year).
Additionally seasonal changes due to humidity of ~5 ppm.
https://www.eevblog.com/forum/metrology/tales-from-the-mouth-of-a-voltage-reference-ageing-box/msg3618924/#msg3618924

ADC17 is based on LT1236AILS8-5 which is typically about a factor of 2 less stable.

I would expect those to be better than that, also after the first month of use? Do you see correlation with temperature or humidity?

The parts in my ageing box are read under constant conditions (heated up to 50.0 deg C before measurement each day). ADC17 is temperature compensated.
I fear it has to do something with the die attach (EPOXY) which creates hysteresis over temperature changes.
https://www.eevblog.com/forum/metrology/tales-from-the-mouth-of-a-voltage-reference-ageing-box/msg3618927/#msg3618927

What about LT1027? Advantest used it in their AD7461A which allowed for integration periods up to 10 seconds.

which package? (the metal can package is no longer available).
I had used LT1027CCN8 (DIP8) in ADC4 + ADC8 with about 0.5 ppm/%rH humidity drift.
This was solved when using the AD586LQ in ADC13.
The LT1027DCLS8 (LS8 ceramic) in ADC21-ADC24 showed the following.
- many of the samples show large hysteresis on temperature changes -> you have to select them to be < 1ppm.
- if soldered directly to the PCB there is again a large temperature hysteresis -> so it can be only used "dead bug" style.
but even then the ageing drift after 5 kHrs is still too high for my needs.

https://www.eevblog.com/forum/metrology/t-c-hysteresis-measurements-on-brand-new-lt1027dcls8-5-voltage-reference/msg1040415/#msg1040415
https://www.eevblog.com/forum/metrology/t-c-hysteresis-measurements-on-brand-new-lt1027dcls8-5-voltage-reference/msg1349435/#msg1349435


I have been asked wether I have already tested molding of the plastic reference in paraffine.
But I have no experience with that. And I fear that over my temperature range of 16-34 deg C I will not get it really tight.
So as with other conformal coating I might get only a increased time constant for the humidity.
Does someone else have experience with paraffine?

with best regards

Andreas

[guenthert]

Well, availability seems to be about zilch, but the ADR1001 comes (will come?) with a 5V output, doesn't it?

[iMo]

Yep, the ADR1001 comes with 7V and 5V and/or -5V and/or 10V out, and all the stuff is in ceramic and hermetic/ovenized (default=70C, you may set the temperature with a single ext resistor from aprox 40C to 120C).
Finetuning of the output voltage possible with a single ext 110ohm resistor (the zener current).
You would need buffers for some configs. The performance wise perhaps better than the 399 (let us wait on the final release). Hopefully we will get the official release and DS soon ..
Will not be cheap (but much cheaper than $$ to get 5V off the ADR1000 or ADR1399), however (perhaps at the ADR1000 level, the estimated price in fall 2022 was $60).

PS: the notorious link https://www.eevblog.com/forum/metrology/adr1001-ovenized-voltage-reference-system/

[deepfryed]

Wow that sounds nice, looking forward to getting my hands on one.

[iMo]

@Andreas: some times back I had an email conversation with the guy (in CZ) producing the nixie tubes, asking him whether it would be possible to encapsulate a small electronics module into a small evacuated glass tube like the nixies he makes. He indicated basically doable with several technology challenges, as usual.. 

https://www.daliborfarny.com/

[Conrad Hoffman]

Way back when I did my reference I found plastic packages unsuitable. AFAIK, there is nothing you can pot things in, including paraffin, that will solve the problem, just slow it down. The old GR mica capacitors were sealed in an aluminum box using resin and they included a generous tube of desiccant. Probably good for a long time, but not forever. It might be a suitable answer for a reference, since you have to re-cert it every so often. IMO, the tube is a great idea. I've been going to try and duplicate the old vacuum thermistors used for oscillator stabilization, but never seem to get around to it. They have to be evacuated to a pretty decent level and I don't have a good enough pump, but a reference would be less fussy. It could even be back-filled/flushed with dry nitrogen or argon, no pump required at all.

[Andreas]

Hello,

sorry but the ADR1001 is out of scope (5 mA limit)

Nice idea with the tubes.
Optimum would be with copper wires soldered into Kovar eyelets like in the vishay VHP resistors.
The usual Kovar leads will cause additional thermal EMF.
 
with best regards

Andreas

[magic]

@Andreas: some times back I had an email conversation with the guy (in CZ) producing the nixie tubes, asking him whether it would be possible to encapsulate a small electronics module into a small evacuated glass tube like the nixies he makes. He indicated basically doable with several technology challenges, as usual.. 

You could empty a TO-3 can, mount stuff inside, add desiccant, and solder it shut.

[iMo]

The tube with some traditional electronics components inside cannot be evacuated to the same level as with the nixies. The vacuum will be several orders of magnitude worse.
But that is not a big issue, imho. What is important it will be hermetic and partially evacuated (lowering the thermal flow inside).
The wiring - that is one of the issues as the tube's pins cannot be soldered but welded (there is the large heat during the process in that part close to the pins). Or to use long wires instead of the pins.
Also the module itself has to be placed at the very top of the tube to be away of the heat (you will be melting the glass at the very bottom). Perhaps a longer tube will be required, or to use some thermal insulation between the bottom and the module at the top (or cooling of the top side somehow).

[iMo]

@Andreas: some times back I had an email conversation with the guy (in CZ) producing the nixie tubes, asking him whether it would be possible to encapsulate a small electronics module into a small evacuated glass tube like the nixies he makes. He indicated basically doable with several technology challenges, as usual.. 

You could empty a TO-3 can, mount stuff inside, add desiccant, and solder it shut.

There are feed through pins with glass-to-metal seal available.. You may make a small metal box, drill the holes, solder the feeds in, solder it shut..

[Echo88]

https://youtu.be/N1F4duEZR_s?t=1359 Like this one. Maybe there are fitting/cheap OCXOs to gut.
A glass tube ref isnt really feasible due to the glass fusing heat, but would be really sexy. 

[iMo]

Would be interesting to compare the performance of his hybrid "ADR1002" with the cheapo ADR1001

[Gyro]

The problem with most hermetic solutions, particularly the glass tube is the materials used for the leadouts (Kovar / Dumet?) and thermocouple effects. The only thing that gets around this is the hollow sleeve type glass hermetic seal where you can solder a copper wire right through without interruption. These used to be available NOS from former Soviet countries on ebay but the listings seem to have dried up. Whether this is because the warehouses are all in Ukraine, or supplies have really been exhausted, I don't know but they are worth keeping a eye out for.

Using such feed-throughs, it's really easy to make hermetic enclosures using soldered copper or brass sheet. As Conrad says, flushing with dry Nitrogen or Argon is much easier than evacuation. Argon is the most easily available inert gas (welding supplies), it has half the specific heat capacity of air though. I'm not sure whether that's actually good or bad - less cooling vs less effect from convection.

[Conrad Hoffman]

With proper design, you can seal off a glass tube without putting too much heat into the contents. I can envision an LS reference package wired dead-bug. I think there's even a way to get copper wire through glass by flattening it and keeping it small, though I may be thinking of some other wire type.

edit- look up Houskeeper glass seal (note the spelling).

[Kleinstein]

Beside covar wire another option may be a tungsten wire, as is used with some (especially halogen) light bulbs. The thermal expansion may be good enough and also the thermal emf is not that much different from copper (~ 1 µV/K difference).

With constant, relatively low power and symmetry the thermal EMF should not be such a big problem. It can add a little to the reference but a mianly constant part.

Instead of a TO3 case, I would consider a THT crystal oscillator can - some NOS ones can be optainted quite cheap (or used one found) and the space may even be enough for a SO8 case inside.

[magic]

You could empty a TO-3 can, mount stuff inside, add desiccant, and solder it shut.

There are feed through pins with glass-to-metal seal available.. You may make a small metal box, drill the holes, solder the feeds in, solder it shut..

Same thing, but transistor (or oscillator) cans are off the shelf items available everywhere that only require simple modding to put them to use.

Regarding Kovar, AD586Q uses it too. Intuitively, it shouldn't be too big a problem if the reference is a low power one and isolated from external heat sources.

[Gyro]

Some solder-in ceramic feedthrough capacitors could possibly be considered hermetic to, I'm not sure but I've certainly seen glazed ones. Parasitic capacitance might cause issues on some nodes if not isolated but would be a positive benefit on the supply pins. I don't know what wire they use but it might be copper.

I guess, within reason, the chamber should be as large as possible so that all relevant components - scaling resistors etc can be kept in an inert atmosphere, with just the connections for a fine trim pot brought out.

[deepfryed]

Is AD587 an option ?
 https://www.digikey.com.au/en/products/detail/rochester-electronics-llc/AD587KQ/12094126




Ah 5V references, should've double checked

[The Soulman]

Andreas,

Thermal hysteresis is a real concern PITA but I have not noticed such on lt1021 in H-package (tested 10 pcs, always settle back within a single ppm after a +50 dgr. C. thermal excursion).
Sadly the H versions are discontinued (although Arrow have some in stock including the 5V version (€€..)).

Also, when looking at the adr1399 datasheet it seems that the ceramic package performs (much) worse on thermal hysteresis compared to the to-46.
If this is a inherent function of a ceramic package (more mechanical stress on the die as you already mentioned), a metal can v-ref is still the way to go in 2024.
Sadly not much on offer.

Or.. avoid thermal fluctuations and keep a steady (controlled/raised) temperature.

[David Hess]

I imagine trying to solder a lid down onto copper clad board and then evacuating it, but even if the FR4 material could hold a vacuum, it would put enormous stress on the board so the reference would need to be mounted to a separate board inside, in which case why not use a real enclosure, like a copper pipe with copper caps and evacuate that?  The high thermal conductivity of copper helps maintain an isothermal environment.

And that leaves how to get the signals out and in, but solderable hermetic feedthroughs exist, or feedthrough capacitors.

[Kleinstein]

There is no need to evacuate. The point of a hermetic seal is keeping the humidity constant inside The actual humidity level choosen should not be that critical. Starting near the normal environment may even help to make residual leaks less relevant. So an extra gas like argon also has downsides.

Even without vacuum it would make sense to have a seprate PCB and enclosure to avoid stress from variable air pressure.

[David Hess]

Besides atmospheric effects, evacuation would also remove the thermal effects from convection.

[magic]

There is no need to evacuate. The point of a hermetic seal is keeping the humidity constant inside The actual humidity level choosen should not be that critical.

Not sure if long term temperature variations could shift humidity distribution between the chip and surrounding air inside the can?
My first instinct would be to fill it with desiccant to be sure.

[iMo]

The epoxy packages are full of water (like 5% of the volume). With temperature variations the water and water vapors will mess with the package, imho. I would get rid of the water before closing an epoxy package (or the epoxy pcb) into the hermetic package. Afaik 2 days at 150C shall do the trick..

[Kleinstein]

If the temperature changes, there is some shift in where the humidity goes preferably. However in most cases the difference is not very large as the binding energy for the water seems to be not that variable. As an example the RH in equilibrium with salt solutions does not depend much on temperatur. This also applies to many solid materials (e.g. wood at a given level of humidity).
Adding some desiccant can be helpful as it increases the amount of water the system can hold - so a small leak wout not have that much effect and one may start rather far on the dry side if wanted. One may get some initial drift from getting the plastic parts in equilibrium with the closed cell.

One may still get a shift of humidity if there are temperature differences in the can with parts that can hold humidity. This could be an argument against adding desiccant if it gets a different temperature than other parts inside.

[Conrad Hoffman]

Most voltnuts don't care about size, at least I don't. I wonder about the feasibility of buying dice and mounting them to a simple alumina substrate that could be put in the hermetic housing of our choice? Alas, I have no wire bonder. Or, buy some easy-to-seal package from somebody like this- https://www.jitai-electronics.com/hermetic-package-ic/ No experience in this for many years, so I don't know if it's a crazy idea or not.

[iMo]

Such a package will cost you an arm and a leg.. Perhaps more than an LTZ1000A with all the resistors around. My guess..

[Andreas]

Hello,

very interesting ideas here.
But I fear the only one that I might test is the suggestion with the paraffine.
(hoping that the time constant will be > 1 Year so that I cannot distinguish it from normal ageing).

In the mean time the PCBs for the REF7050 have arrived and I started with population of the first sample (missing only some parts).
The next sample I will try with a dead bug mounting.

with best regards

Andreas

[F64098]

I mentioned in long time ago in another thread. The Morion MV180 contains a MAX6350ESA in the inner housing to generate the reference voltage.
So it's double hermetic contained and also precisely temperature controlled. Maybe it's not the best reference at all, but for me it looks like a "reference" for
other, unsealed references in view to their reaction to air humidity or air pressure.

Best regards

Frank

[iMo]

There are "(Ultra) High Vacuum Sealants and Adhesives" for usage with vacuum gear, afaik.
Pretty expensive, but somebody told me in past you may seal a glass tube with it hermetically..
Would be great to investigate..

[Conrad Hoffman]

The usual concern for high vacuum stuff is outgassing. The NASA list here- https://ntrs.nasa.gov/api/citations/19970027853/downloads/19970027853.pdf is useful if you look for low total mass loss (%TML) and volatiles. The problem with good epoxies is you either have to buy a rail car full or they're quite expensive, with high minimum orders. Interestingly, some of the RTV rubbers are quite low, probably seal well and are easily available. I don't have any info on the permeability of paraffin vs epoxy vs RTV vs the hot resins that so many things used to be sealed with. GR standards were sealed with hot tar-like resins (or maybe they were tar!) as were ordinary D-batteries. IMHO, it's hard to beat a soldered joint!

[Zoli]

My small contribution:
1. I've built two references, based on LT1236-10, schematics and picture attached. Details on the construction: the inside aluminum box is sealed with rubber "ring" on the top, and RTV along the wire exit; is suspended in hard foam within the fibreglass enclosure. Is powered from 2 9(8.4 Li)V batteries. They agree for the last 3 years with my 34401A's within the last(sixth) digit; I consider them  5&1/2 digit references.
2. Vacuum grease: I've worked on a few(~10) vacuum related projects; not high, down to 10^-3 Torr range. The goto vacuum grease for those projects was Dow Corning 111(or the current equivalent).

[Conrad Hoffman]

Very nice- is that a Rose enclosure?

[Zoli]

Very nice- is that a Rose enclosure?

The fibreglass is Hammond. The aluminum is unknown origin, of which I had two lying around; originally was some kind of remote sensor, the cable exit is an SMA hole.

[iMo]

The next sample I will try with a dead bug mounting.

I would highly recommend.. Use thin wires (say 0.1mm), such the package has got some freedom of movement (a trampoline  ).
Like a small 12x12mm cutout in your pcb, with pads for the wires.
Btw, why not having a temperature sensor on your pcb (wired to the mcu)?

PS: I would trampoline the LTC2400 as well..

[DavidKo]

What about using the enclosure with some desiccant inside. Before closing leave the whole setup at ~125°C to let the water evaporate and than seal the container. You can use the silica gel or zeolite.

Glass metal interfaces are always tricky. What I remember from vacuum lectures the glass is usually named after the metal to be compatible. If the tube design should be used. One can use the viton sealing around the legs (copper for example), evacuate for longer time period and fill the tube with some inert gas with small over pressure. water will have hard time to get in and the device will have "constant" atmosphere.

[Andreas]

Hello,

I would highly recommend.. Use thin wires (say 0.1mm), such the package has got some freedom of movement (a trampoline  ).
Btw, why not having a temperature sensor on your pcb (wired to the mcu)?
PS: I would trampoline the LTC2400 as well..

The sensor was one of the missing parts. (I wanted to see the golden reference on the photo).
Now populated. (orange golden blob on the reference).
Hoping to get the same temperature stability (< 1 ppm over 10-40 deg C) like on my AD586 references.

Trampolining the LTC2400 is useless.
The LTC2400 does a self calibration (offset + full scale) on each single measurement.
So it behaves like a chopper amplifier.

For the reference: I typically solder the GND-Pin of the reference directly to the NTC pin to get some stability and sense the temperature.
The other pins with thin VERO-Wire (~0.2 mm)

My small contribution:

Nice design.
Did you do any T.C. measurements (hysteresis) or rH% measurements with these samples?

[Zoli]

...

My small contribution:

Nice design.
Did you do any T.C. measurements (hysteresis) or rH% measurements with these samples?

Thanks for the appreciation; the build reflect what I've had available at the time, and try to respect common sense and best practices, both in the enclosures and electronics.
Measurements: unfortunately I didn't had/have the possibility to make any of those measurements(lack of access to environmental chamber), only pressure/vacuum test: passed ±50kPa for 1h.
As bonus, original resolutions close-up/bare PCB(I still have 3 left ).

[iMo]

@Zoli: perhaps add the diode in your next pcb revision (see below). Also the RV1 trimmer - generally it is not recommended to use those plastic trimmers with the precision references..

[GigaJoe]

did play a lot trying to seal plastic chips .

major problem of plastic it do affect the crystal from any angle , as it molded and surrounded,  comparing to the metal \ ceramic case, where crystal attached to one side.

so first and foremost, as little as possible minimize any forces \ tension during mount \ soldering \ fixing chip on board  ;  should be like a float design. 

at second as a result from above,   anything that apply any forces to the plastic surface -  not work at all.  any solid material such as paraffin , rosin , create a forces that transferred over plastic to the crystal , basically any solid filling that contrast or expand, don't work at all ,

the best case i got -   floating design, thermally compensated by set of diodes,  heater +55C all submerged in , transformer oil ,  in the thermos.
as liquid -  Silicone or fluorocarbon-based oils would be the best,  then transformer oil , then mineral oil ...

an alternative: DIP8 , glued to the owen +85C,  and very thick around few cm Styrofoam box.  some kind headache to adjust temperature compensation for this temperature.

finally - small tin can , Italian canned fish, edge tinned , and cap are dual copper FR4 , copper wire through,  soldered on both sides, can soldered around edges to PCB cap.   -  more  less

heated oil bath - i have better then other result

[Zoli]

@Zoli: perhaps add the diode in your next pcb revision (see below). Also the RV1 trimmer - generally it is not recommended to use those plastic trimmers with the precision references..

Thank you for your concerns; let's see what the datasheets has to say:
The LDO

FEATURES
Reverse-Battery Protected
No Reverse Current
No Protection Diodes Needed

ABSOLUTE MAXIMUM RATINGS
IN Pin Voltage .........................................................±20V
OUT Pin Voltage ......................................................±20V

APPLICATIONS INFORMATION
Protection Features
The LT1761 regulators incorporate several protection
features which make them ideal for use in battery-powered circuits. In addition to the normal protection features
associated with monolithic regulators, such as current
limiting and thermal limiting, the devices are protected
against reverse input voltages, reverse output voltages
and reverse voltages from output to input.

The reference

Trimming Output Voltage
A 10kΩ potentiometer is recommended, preferably a 20 turn cermet type with stable
characteristics over time and temperature.

The trimmer is a Bourns 3296W:

Features
Multiturn / Cermet / Industrial / Sealed

Additionally, if you take a closer look to the previously posted picture(Close-up.jpg), on the top you can see a diode in DO-41 case connected between the switch and the low voltage detection board. It's a 1N5817, whose sole role is to protect the LP2951 on that board, even if it supplies the reference board, too.

[andrewtaylor]

In the mean time the PCBs for the REF7050 have arrived and I started with population of the first sample (missing only some parts).
The next sample I will try with a dead bug mounting.

The REF70xx are very nice, I also recommend the dead bug style which I use in all my reference boards.
I recommend you take a few (3..6) REF7050 and compare them first for the one that best suits your needs (drift, ageing, etc.) before choosing one for your final PCB.
All REF70xx are good, but some variation exists in the batch you'll get.

[iMo]

..The trimmer is a Bourns 3296W:

Features
Multiturn / Cermet / Industrial / Sealed

..

Imho that is not an indication for a suitability with precision voltage references..
I was using those blue bricks with several references and found out with temperature sweeps (like 20-50C) it made small voltage jumps. I've been still using one in my 2xLT1021 ref with a wax drop put over the adjustment screw.

Also what I've learned - those trimmers have pretty large "dead zone" when turning the adjustment screw up/down. After I adjust the proper voltage I always turn the screw a little bit back, such it is positioned in the middle of the dead zone. Thus the screw is not engaged with the internal gear. It seems to me that helps a bit with stability. Mind the plastic gear inside (afaik) changes its shape with temperature/aging..

[Svgeesus]

1. I've built two references, based on LT1236-10, schematics and picture attached.

Could I ask about the design of your trimmer circuit?

The LT1236 datasheet suggests a trim circuit with a diode between the trim output and the wiper of the cermet trimmer, says that this is related to maintaining a low tempco, and mysteriously hints that for this reason the resistor values should not be altered (while not giving resistor or trimmer tolerances, tempco or any other details).

[Svgeesus]

Oh - I hadn't noticed that the 10V and 5V versions use quite different trim circuits. 

[Andreas]

I recommend you take a few (3..6) REF7050 and compare them first for the one that best suits your needs (drift, ageing, etc.) before choosing one for your final PCB.
All REF70xx are good, but some variation exists in the batch you'll get.

Hello,

Unfortunately I have no socket for the LS8 package (a DIP8 package eases selection).
I have planned to populate 4 PCBs (depending on the first 2 measurements).
Do you have numbers for hysteresis over temperature and initial drift (how long).

Mind the plastic gear inside (afaik) changes its shape with temperature/aging..

In my AD587LW-Design I also use these trimmers. But with a rather high resistor (> 1 Meg) at the wiper narrowing the adjustment range to some ppm.
Additionally some trim resistors as pull up / pull down for adjusting the range directly at the adjustment input.
The original BOURNS trimmers are somewhat better than some no name trimmers with the same shape.
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg1449488/#msg1449488

with best regards

Andreas

[Conrad Hoffman]

I avoid no-name trimmers but always pad trimmers such that they cover such a small range that even gross performance problems can't affect much.

[iMo]

I recommend you take a few (3..6) REF7050 and compare them first for the one that best suits your needs (drift, ageing, etc.) before choosing one for your final PCB.
All REF70xx are good, but some variation exists in the batch you'll get.

Hello,

Unfortunately I have no socket for the LS8 package (a DIP8 package eases selection).
I have planned to populate 4 PCBs (depending on the first 2 measurements).
Do you have numbers for hysteresis over temperature and initial drift (how long).

FYI - below the aging of the REF70xx from the DS rev 09/2023  ..

[3roomlab]

how about the MAX6126 ?
their app note looks impressive



https://www.allaboutcircuits.com/technical-articles/electronic-component-aging-prediction-the-long-term-drift-in-voltage-references/

https://pdfserv.maximintegrated.com/en/an/AN7009.pdf

[Andreas]

Hello,

first result from ADC#32 with REF7050 in dead bug mounting.

Temperature cycling from 10 deg C to 40 deg C with 0.12 K/minute
one temperature cycle is nearly 12 hours.

The T.C. is fairly low but the large temperature hysteresis of 7-8 ppm @25 deg and the large ageing drift of ~4ppm/day makes the REF7050 useless for precision work.
At maximum they can be used for ratiometric (RTD temperature) measurements

how about the MAX6126 ?
their app note looks impressive

7 ppm due to humidity change after a short power interruption?

"The glitch
shown at approximately 2400 hours corresponds to a rise
in humidity that occurred due to the backup power supply
not immediately turning on."

I fear I have enough bad (slow responding) humidity sensors in my ADC collection.

with best regards

Andreas

[Andreas]

Hello,

just for comparison: ADC#25 with AD586LQ under similar conditions.
Hysteresis is some uV (typically < 1 ppm) in this case and no visible ageing drift.

with best regards

Andreas

[slugrustle]

Not sure if someone else mentioned it or not, but ADR4550DEZ looks good.  Only trouble is the ceramic package ones don't seem to be available.

[iMo]

Hello,
first result from ADC#32 with REF7050 in dead bug mounting.

The DS shows pretty large initial drift. Wait 2-3w what will happen.
The hysteresis could be caused by the large aging drift, imho.

[miro123]

Hello,

first result from ADC#32 with REF7050 in dead bug mounting.

with best regards

Andreas

I don't fully understand the graph. I have few questions/remarks.
The graph looks very good at both ends between two cycles - T=10C and 40C . Do you perform some compensation of initial time drift?
Also going back and forth to 10C looks very good.

I have learned from AD AppNotes - First perform oil bath aging test, then T hysteresis tests. On such way you are looking at one parameter instead of many issues together. The thinks get worse when you look at many uncorrelated parameters together e.g. temperature humidity and aging
Best Regards,
Miro

[Andreas]

Not sure if someone else mentioned it or not, but ADR4550DEZ looks good.  Only trouble is the ceramic package ones don't seem to be available.

That is one on my todo list.
It was already mentioned in a different thread:
https://www.eevblog.com/forum/metrology/ti-new-ref54evm-evaluation-module-leaked/msg5181666/#msg5181666
From datassheet the ADR4550D looks really good.  (except for initial ageing). But hoping it will get better over time.

The DS shows pretty large initial drift. Wait 2-3w what will happen.
The hysteresis could be caused by the large aging drift, imho.

I fear no: the drift within 12 hours is much smaller than the hysteresis in one cycle.

I don't fully understand the graph. I have few questions/remarks.
Also going back and forth to 10C looks very good.

Attached a diagram with different colors for full cycles: (25-40-10-25)
Initially I started with 23 deg and a cold phase. (red)
The cold phase nearly returned with no hysteresis to 23 deg.
But after warm phase a large hysteresis is shown.
Somehow at 25 deg it goes to the old track.

Ageing then takes place during the next warm phase (green)
So the minimum voltage is shifted and also the end of the cycle.

The final measurement values (blue) show also a warm drift.

with best regards

Andreas

Edit: No there is no compensation of ageing drift in this case.
The first measurement is more or less to get an overview and to calibrate the on board NTC to the external temperature sensor.
Of course the hysteresis usually gets somewhat lower on following cycles. But not with nearly a magnitude higher to below 1 ppm.

[Andreas]

Hello,

first result from the other sample (ADC#31) with REF7050 directly soldered to slotted PCB.
As expected the hysteresis is larger than on dead bug mounted PCB. (18-21 ppm at lower temperatures).

But there is also an unexpected "jump" near 29 deg C.
And a very weird T.C. curve which is hard to be approximated.

with best regards

Andreas

[Kleinstein]

The datasheet suggests that there is some digital calibration / TC adjustment. For the 2nd sample it loops like a brakepoint in correction near 29 C.

The datasheet also gives rather high values for the hysteresistis. So at least they a honest in this aspect and the hysteresis does not come as a surprise.
There is a slight chance that the hysteresis gets significant smaller for a reduce temperature range.  The more normal use case is also with a slightly smaller temperature range. Of cause longer times at the temperature extremes (e.g. a few days in transport) may also increase the hysteresist.

[Andreas]

The datasheet suggests that there is some digital calibration / TC adjustment. For the 2nd sample it loops like a brakepoint in correction near 29 C.

Hello,
interesting theory.
it looks like there is a hysteresis on falling temperatures over ~ 7 degrees from 36 to 29 deg C before the jump.

But why does the jump exist only on falling temperature? and not on rising.
And why a 30 uV/3.333V (~9ppm Jump) within ~1 deg C for a 2 ppm/K reference?


with best regards

Andreas

[iMo]

The TC in REF70 is digitally compensated and specified with the box method from -40 to +125C. Inside the box it could be almost anything and bumpy, imho.

[TUMEMBER]

The TC in REF70 is digitally compensated and specified with the box method from -40 to +125C. Inside the box it could be almost anything and bumpy, imho.

Maybe that's why ?

[Andreas]

The TC in REF70 is digitally compensated and specified with the box method from -40 to +125C. Inside the box it could be almost anything and bumpy, imho.

Hello,

from which document do you get this information.

On the current datasheet there is only the hint of "temperature curvature correction" which is standard since 2nd and 3rd generation references.

with best regards

Andreas

[Kleinstein]

The block diagram shown a digital core. It is still a bit unclear what that means.

With quite some newer parts (e.g. e-trim OP-amps) they use some DACs instead of old style Zener zapping or laser trimming. This can still be essentially static working DACs, as a kind of digital trimmers. The idea is that trimming can be done relatively fine (zener zapping is limited), with multiple parameters and after encapulating in the case.  This is how they can get quite low offsets even with CMOS OP-amps (e.g. OPA192) in a plastic case.
The digital trim may allow for more parts to the temperature compensation, more than just absolute value, linear and curvature. I don't hink the correction would be with an ADC, math and than a correction, more like addition of correction terms and fixed DAC values to scale the amplitude for these correction terms. More terms allow for lower TC also over a large range.

The 2nd part looks like it has a rather poor voltage - temperature curve with quite some corrections applied and this includes also a rather steep section. Besides the "jumps / cusp near 29 C there also seems to be brake in the curvce at some 35 C and a bit unclear at 8 C.
The hysteresis part may effect the correction parts differently and thus the difference in the curve going up and down.

I would consider that 2nd part borderline - defective / worth going to the reject bin. It is alte least limited in the useful temperature range.

[Andreas]

Hmm,

I am somewhat confused ...   
some minutes later (omitting the first 1000 minutes for overview).

with best regards

Andreas

[iMo]

Years back chuckb made test with the ref7025
https://www.eevblog.com/forum/metrology/ref70-new-from-ti/msg3569115/#msg3569115
and he got 56ppm hysteresis while cycling 25..61C and back.

[Andreas]

Hello,

should have remembered that. I wouldnt have made a PCB especially for REF7050 in this case.
I fear I will at maximum mount the ADC31 also as dead bug style to see if the "dents" will disappear.

with best regards

Andreas

[Andreas]

Hello,

update to ADC32 (dead bug mounted reference)

https://www.eevblog.com/forum/metrology/stability-of-5v-references-compared-to-ad586lq/msg5278726/#msg5278726

after Offset Drift T.C. compensation and T.C. compensation.
The ageing drift is largely reduced.
The hysteresis is still there.

with best regards

Andreas

[Kleinstein]

It may be interesting if the hysteresis gets much better with a small temperature span. Normal use usually is quite a bit smaller. There is a slight chance that a smaller range could be much better, e.g. if there is a threshold for things like thermal stress before it really causes hysteresis.

[Andreas]

Hello,

ADC#31 after mounting the (same) REF7050 dead bug style:

see here for reference:
https://www.eevblog.com/forum/metrology/stability-of-5v-references-compared-to-ad586lq/msg5280709/#msg5280709

- There is a ~400 uV/3.333V shift 120 ppm (soldering drift?) between the PCB-mounted and dead bug mounted measurement.
- the dents do no longer show up (so probably was some interaction between PCB and reference)
- T.C. between 15 and 35 deg C (at thermally isolated reference so 10 to 30 deg C environment) is increased from ~3 ppm/K to ~6 ppm/K
- hysteresis is still too large (~14ppm) but now more near room temperature.

with best regards

Andreas

[branadic]

Looks not like a reference, but more like a t.c.-attractor.

-branadic-

[enut11]

Hi Andreas.
LQ chip available from Mainline Electronics UK, ebay item number 254576463764.
Enut11

[Andreas]

Hello,

Obviously they sell not to every country:
"This Shop has no inventory available in your location"

On the other side: I still have some AD586.
The only question that arises: what to do when my stock is depleted.

with best regards

Andreas

[Gyro]

Strange, because they list shipping to just about every other EU country, eg. £10.29 to the Netherlands. I think they simply forgot Germany.

It's a good price for an LQ though. I picked up one of them a while back but haven't got around to testing it yet. I wish I'd bought up some bare SVR PCBs when Joe Geller was shutting up shop!


Edit: Oh, "Excludes: Afghanistan, Austria, Germany, Libya, Nauru, Russian Federation, Somalia, Switzerland, Yemen". You're in interesting company!

[magic]

I see only 1pc available and are you sure that it's legit?
You can get dozens of AD586-not-really-L from AliExpress, anyone can import and resell them.

[Gyro]

From the seller's other items, my ebay-fu says yes, probably.

[Andreas]

Hello,

fortunately some ADR4550D showed up datecode #2351
Two of them mounted dead bug style on the REF7050 PCBs.
Hoping that the hysteresis will be much lower.

with best regards

Andreas

[argintviu]

Regarding hermetic enclosures, I was thinking to make one for a NOMCA network used for a 7->10V amplifier as it's the only non hermetic part in the reference. At first I thought to make a square hole in the PCB right under the SOIC and pour refined beeswax under the package then over it. This way the package would be surrounded with wax. From what i know it is supposed to be quite water resistant.

The second cheapest thing (I think) would be to cut, unroll and flatten copper pipe to make a small box and solder it. Then carefully remove the bases of 2 x HC49 THT crystals (to get 4 wires), make holes in the copper lid for the HC49 bases and solder them on. Then solder the resistor divider to the wires, put the lid on and solder the cap to the box. 3 wires are enough for the divider, the 4th could be tied to the case. For HC49 crystals, the seals are usually Kovar-glass and the case plating is nickel (according to the material declarations) so it should be easy to solder it to copper. Wax could also be poured inside. Might be fiddly though .

[iMo]

Wax itself will not be hermetic (we talk here water vapors resistant).
The package made of metal like the copper foil with hermetic feeds through would be the solution. No need to pour wax inside (wax may contain water as well). I would heat up the complete package with resistors or reference chip inside with a small hole made in it for a while (say 100-150C such the water on the inner surfaces evaporates) and then I would shut the hole up with a solder drop (while the package is still hot). Under normal conditions all surfaces are covered with a thin (called mono-molecular) layer of water (and sometimes with other stuff as well) here on our planet..

PS: another option is to put the package (with the small hole in it) into a high vacuum and solder the hole shut inside the vacuum after a while, but that would be a pretty difficult exercise for a hobbyist, imho.

[argintviu]

I see. It's probably the same with silicone oil, right? Can't they be desiccated by heating before using?

Regarding the assembly, it would probably need to be baked for a few good days at 100C to get it dry before sealing. One could also use a piece of pipe directly like David Hess suggested and put the HC49 feedthroughs on each of the 2 caps, but then it would be difficult to hold them at the same temperature. Maybe use a larger diameter pipe and have both feedthroughs at the same end and mount it vertically. The hard thing is separating the bottom part of the crystal without breaking the seals.

[argintviu]

To make things easier, one can solder a capillary tube to the case, draw the air out, mechanically crimp/flatten it and then solder the top for a final seal. I saw this being done on older fridges after filling in new refrigeration gas.

[iMo]

You may buy the ceramic hermetic feeds through on ebay or else. I DIY in past such I soldered the feeds through off the old bulky metal germanium and silicon diodes. And yes, as I wrote above the complete package has to be baked at those temperatures for some time to get rid of the water inside, then sealed up with a drop of tin. No oil or wax needed when the package is soldered properly.
The epoxy package needs to be baked at 150C for 48-72 hours to get rid of the water (5% of the volume, btw).

[argintviu]

Unfortunately the cheaper, soviet hermetic seals are long gone from the bay. The only other thing I found that seems to have the hollow type solderable kovar seals are the K40Y-9 paper-oil capacitors. But I have no idea how they're built and if they can be dismantled in a convenient way.

[Kleinstein]

There is no need to get the inside of a hermetic chamber dry. The problem with humidity is not the water by itself, but the changes in the humidity. A temperature change effects the RH. With a circuit that gets warm in use and is used on and off this will cause some hysteresis type variation from humidity dirven out from the PCB when hot and back in when clod. So sealing with a low humidity does have benefits. It would be less of an issue for a circuit that runs 24/7.

Wax and plastics are not 100% humidity tight, so they won't keep humitidy out over a very long time. The first target would be to slow down the humidity variation over a year. Once this is filtered out, humidity would no longer change very much. With large enough a container and absorbing material, one could keep humitiy out for an extendet time even with plastics seals.

[Andreas]

Hello,

first result on ADC#35 (ADR4550D + LTC2400) measuring a ADR1000A reference: (first NTC calibration of the on board sensor)
the initial drift ~4-5 ppm for the first 12 hours is now settling at ~1 ppm per 12 hour temperature cycle. (0.12 deg C / minute ramp)

Hysteresis seems to settle at ~4ppm (+/-2 ppm) which is better than ADR4550 in SO-8 plastic package (~16ppm) but still more than I hoped.

with best regards

Andreas

[miro123]

Seems promising Andreas.
The normal sequence is
1. Ageing tests  - then aged components are used for further tests
2. TC tests
3. Hysteresis tests
On such way you exclude counting of multiple errors at once.

[Andreas]

Hello,

in my world:

0. NTC calibration of on board sensor (but why not peek for ~T.C. and hysteresis)
1. after ~1 week first T.C. test (at least if I have stable buried zener references like AD586). Hysteresis is measured at the same time.
2. T.C. of the Offset of the 2:1 divider
3. INL of the ADC (with T.C. and offset calibrated out)
4. further tests like noise (popcorn), PSRR ...

so NTC calibration was the main task

with best regards

Andreas

[PCB.Wiz]

I see TI have a part code for REF54500, but no stock visible ?

Looking at Digikey, I also see Masach Tech Ltd. and Würth Elektronik have seamless RF shields, some for clip over a mounted base, some for solder.
https://mm.digikey.com/Volume0/opasdata/d220001/medias/images/987/MFG_MS102-10S.jpg
Available in trays so production suitable, they could give a simple cavity.

I guess they need open vias during soldering, which could be soldered or taped closed manually later.
 

[Andreas]

Hmm,

I am not shure what you want to tell me with that.
I am not a native english speaker.

But I think that a plastic packaged device is not suitable for my needs.
With or without external cavity.
Or do you have experiences with that device regarding humidity?

with best regards

Andreas

[Andreas]

Hello,

next initial measurement this time with ADC36 with 2nd ADR4550D.

It shows lower T.C. and lower initial ageing drift.
But slightly higher hysteresis of 6-7 ppm.

with best regards

Andreas

[miro123]

Hi Andreas,
What type of temperature curve are you using?
Ramp-up Ramp-down rate? Do you apply dwelling time for every measure temperature point?

[Andreas]

Hello,

the Setpoint ramps with 0.12K/minute with some plateau at minimum and maximum temperature. (so one complete cycle lasts ~12 hours)
Since temperature is never the same at 2 different points the actual temperature on the PCB differs somewhat from the heat spreader
especially when the refrigerator is switched on/off.

see attached picture

with best regards

Andreas

[iMo]

It seems to me your temp speed is still too high (0.12C/minute).
Would be interesting to see what happens when you do for example 3-4 temp points in each direction only and wait say 2 hours on each temp point till the temp at the specific temp point settles..

[Andreas]

Hello,

might be an option to reduce the hysteresis somewhat under controlled conditions.
But I fear that it will not help under my "real life" conditions.
I have in summer gradients up to 3 deg C/hour.
One extreme gradient in winter (forgot to close the window) was 6.5 deg C/hour.
And there are other references which show < 1 ppm hysteresis under this conditions.

On the other side I have recognized that in the ADR4550 datasheet they "cheat" with hysteresis.
They have introduced the terms "half cycle hysteresis" (~100 ppm) and "full cycle hysteresis" (~8ppm)
So the "low thermal hysteresis" is only related to the "full cycle hysteresis" which I would call "ageing drift" instead of hysteresis.
The full cycle hysteresis is practically useless.
After transport of a reference I can measure the actual temperature but not the "history" wether it was higher or lower before.

I made a comparison of the last temperature cycle from NTC adjustment (0.12 K/minute) with a preliminary T.C. adjustment (0.06 K/minute).
The problem is that the ageing drift is still much too high (now 1.5 ppm/day). But I fear the ramp speed has little influence on the hysteresis.
(should say 2*half cycle hysteresis).

with best regards

Andreas

[miro123]

Hi Andreas,
I dont want to discourages you but I want to share my little experience.
I was searching to good Vref candididate voor my sigma-selta adc acqusition board. Requirement ware small size not ovenized Vrefs for 2,5V. tested few of them two LTC6655,  two REF5025, two REF6025,  four ADR431, two ref3425. Nor of them give me sufficient results. Sometime results ware inconsistent due to influence of many parameters. aging, humidity just to name few of them.  Finally I give up.
I lowered my requirement and start testing six LM399 and four ADR1399
Long sotry short -
1. ADR1399 in TO packade is definitly suprerior against LM399.
2. Both LM399 and LM399 are another league of devices. - I get consistent results with good and bad performing devices.
3. I like testing of ADR1399 due to the tighter tolerances.
At the moment I'm processing the RAW data.
See Attached - I have measured the voltage difference between two ADR1399 - The volatge diference is roughly 12mV - Data is captured with DMM6500 - 17200sec~5Hours.
Two Notes
  - some TC difference appears from my test jig -
  - those ADR1399 are pushing my test equipment to the limit.
  - ToDo list - I want to do cross correlation of results between my - HP3456,HP34970, DMM6500 and M3500. Then I can extract more info about the source properties.


 

[Andreas]

Hello,

I know that LM399 / ADR1399 would fit my precision needs.
But I want to have low power consumption too.

Perhaps you should have tried LT1236AILS8-5 with a LTC1043 or LTC2053 precision divider to get 2.5V.
Most of them have a low hysteresis even on the very first temperature cycle. (see table of selection)

The only downside is a ~ -3ppm/K tempco on most devices.
I want to evaluate how good this can be compensated by measurement of the internal chip temperature at the trim pin.
The Trim pin is amplified and then measured by a 12 Bit ADC to get ~0.03 K resolution.

with best regards

Andreas

[Andreas]

Hello,

first results from LT1236AILS8-5 devices.

I had chosen #08 and #13 from the table.
#08 had low T.C. but higher hysteresis and is populated on ADC#33 now.
#13 had ~2ppm/K T.C. but low hysteresis and is populated on ADC#34 now.
As temperature sensor I use the TRIM-PIN of the LT1236.
The pin has ~ 4.0V with ~ 1.7 mV/K and 100K source resistance.
To evaluate that I use a amplifier to scale 10-40 deg C to 1.5V - 3.5V for the 12 Bit ADC of a PIC.

Against my NTCs with 10 Bit ADC the temperature is relatively linear.
But the 12 Bit ADC appears a bit more noisy (+/-1 count around current value) whereas the NTC was stable except for the transitions.
Now the question is if it is related to the additional high ohmic amplification or if the higher resolution ADC is the guilty.

Peeking to the T.C. of the LT1236 it seems that now both have around +/-2 ppm (4 ppm total) hysteresis.
#08 has near zero T.C.
#13 has 2.2 ppm/K
But both have some "jumps" at the lower temperature ends during each temperature cycle.

when comparing to the old measurements 10 years ago I see that obviously the measurements where made at 0.3K/minute temperature gradient. Whereas today I use the more realistic 0.12K/minute. Can this be the reason that the hysteresis was lower in the old measurements? Does the die attach creep more with lower temperature gradients?

with best regards

Andreas

[Andreas]

Hello,

I have found the root cause for the "jumps" at low temperatures.
Found it accidently when trying to add some thermal isolation.
The package is _very_ sensitive to any touching. (package stress).
Jumps of several 10-100uV can be observed.

The fault that I have made is to put the LS8-package directly on the PCB to ease soldering the wire from the other side to the GND Pin.
So the package was touching the PCB (and in between there was the plastic sticker with the number from previous T.C. measurement 10 years ago).

I solved this by rewiring and placing some spacer between PCB and the reference.
After soldering the spacer was removed.
Now I got a "soldering shift" in output voltage and a new fast initial ageing.
So it will take some time until the reference will be more stable.
But I fear there will be still a 3-4 ppm hysteresis.

Another observation was that there might be some temperature sensor drift.
I am using the chip internal temperature dependant trim voltage for the temperature sensing.
And it looks that either the absolute voltage drifts with reference voltage or there is a problem
That I have different reference voltages from LT1236+bridge evaluation
and the 12 Bit ADC (where VRef = VDD from a 5V regulator).

with best regards

Andreas

[Andreas]

Hello,

and the same for ADC#34.

Initial drift is a bit lower because I had more time between soldering and first measurement.

2nd picture: the blue line is from previous T.C. measurement showing the drift by re-mounting.
Against red: the current T.C. measurement. Hysteresis here ~4 ppm. And 1 ppm drift between cycles.

3rd picture: the drift between the temperature measurements red: new green before re-mounting.
But of course there could be also that the sensor8 distance which I use as master temperature
to the LT1236 reference is slightly different due to re-moung.

With best regards

Andreas

[Conrad Hoffman]

Mechanical and thermal stresses play havoc with all references. My plan, as yet unrealized, is to space the device upside down and dead-bug the wiring with fine constantan or manganin. The device is small so the wires should support it just fine. Using ABC (anything but copper) will reduce the thermal flow out of the part and probably increase the stability and lower the power requirements. I'm guessing the added resistance of the wiring will be a constant and not affect anything, but that remains to be seen, depending on the current flow for heating.

[Andreas]

Hello,

Manganin would be good.
Constantan has too much thermal voltage against copper.
And the chips usually have 2-3 deg C more temperature than the environment.
(even when the ground pin is soldered to the PCB).

with best regards

Andreas

[Kleinstein]

Another problem with konstantan wire is that it is hard to solder. Maganin wire is easy to solder.