LM399 effect of moving the cap

[Kleinstein]

The was a discussion about the effect of the position of the plastic cap of the LM399.
https://www.eevblog.com/forum/metrology/adr1000-datasheet/msg3655423/#msg3655423

Not to disturb that thread, that is about something different, I open this up here separate.

The effect of moving the cap looks simple and can be tested. As I currently have a board with a LM399 running open, I took that chance an measured what shifting the cap to one side or the other has.

Attacheche is a measurement of the voltage shift (actually invers because the reference under test is used for the ADC and the other reference (also a LM399 is fixed).  Moving the cap towards the marker tab gives the slightly lower voltage, moving it away from the marker gives a slightly higher voltage.
The large jumps are from moving the cap carefully. The small jumps are likely popcorn noise. The grid lines are 5 µV and this some 0.7 ppm.
Tryping to move the cap again to the same side gave no significant effect.

[iMo]

Now - how to fix the cap? Is there a chance to open the cap without damaging it?

[chuckb]

After the LM399 is installed on the pcb I usually apply two small drops of Electrical Grade RTV to fix the cover to the pcb.

[Kleinstein]

I just moved a cap from a LM399: it work with a needle / small srewdriver to remove the top lid. Just take care where the lid jumps too, so you don't have to search much for it.

Getting the lid off is the easy part, getting it back on is a bit more tricky, if the ref is already soldered. It needs quite some force ( e.g pliers) and a bit care not to bend the leads.

I tried some glass fibers to fill the gap a little, so if will not more very much. The first try was not really enough, but maybe good enough for the anyway a bit jumpe ref.

I checked the other directions: the 2 tested at the start are about the extremes I found.

[iMo]

What about a wax with >95degC melting point..

[Kleinstein]

The cap inside likely used the air space as part of the insulation. The cap inside has an extra ring with some air gap the the sealed can and another ring of air all around. So this very much looks like insulation. Just filling the gap with something more conductive may not be that good.

[iMo]

Wax has low heat conductivity and large heat capacity..
For example:
https://www.candlemakingsupplies.net/shop/wax-and-additives/paraflint-h-1/

[RoGeorge]

It looks like with time it will stabilize at about the same 0.91 level anyway.   

[floobydust]

From thermodynamics, heat is transferred in any of three ways - conduction, convection, radiation.
I'm not sure which method is being changed by moving the Valox resin cap. I don't have experience with the LM399.
The cap dimensions were corrected by LTC in 2014, per their datasheet.

[Kleinstein]

It looks like with time it will stabilize at about the same 0.91 level anyway.   

It settles back somewhat, but I don't think it would really settle all the way, maybe 2/3. There is still a bit of drift in the test, as the reference was not warmed up very long and the reference under test was quite new (about 30 minutes of active heater).

[chuckb]

From thermodynamics, heat is transferred in any of three ways - conduction, convection, radiation.
I'm not sure which method is being changed by moving the Valox resin cap. I don't have experience with the LM399.
The cap dimensions were corrected by LTC in 2014, per their datasheet.

The insulating cap has large holes on the bottom for the leads to pass through. I believe the conductive heat loss of the leads change when they transition from being surrounded by air to conducting heat through the plastic. So, to prevent the conductive heat loss change I use RTV to stop the movement.

I believe the temperature change of the chip is not the issue I believe it's the temperature change of the Kovar/Copper (35uV / K) junction on the pcb. A 0.3K temperature change of one Zener lead would cause the 10uV shift.

For prototypes I also use Electrical Grade RTV to cover the PCB / solder / Kovar connections on the back side of the pcb, in place of a cover.

[ramon]

The cap inside likely used the air space as part of the insulation. The cap inside has an extra ring with some air gap the the sealed can and another ring of air all around. So this very much looks like insulation. Just filling the gap with something more conductive may not be that good.

(As chuckb has already said) Don't forget that the Kovar leads need to be at the same temperature, and that the Valox material (PBT) has 10 times higher thermal conductivity than air (and almost the same specific heat).

The issue with the Valox case is that it seems to be designed to float ABOVE the case and AROUND the kovar leads. But the datasheet does not explain how to properly fix or float the case. And this is critical information. The exact moment that your valox case touchs one lead more than another you have a thermal contact.

               Specific Heat    Thermal conductivity
               [J/g-°C]         [W/mK]  
               =============    ====================
Air            1.012              26.24
Valox          1.260             200
EPS Foam       1.300              35                   (flamable)
Aluminium      0.890             237
Copper         0.385             400
Constantan     0.390              22
Manganin       0.410              22
Evanohm        0.460              14
Steel, carbon  0.420              50
Steel, SS      0.400              15
MgO            0.874              42
PTFE           1.000             300
Wood           2.300             200
Paraffin wax   2.144             189
Water          4.184             598

****           1.900              36 

   [****] Special TOP Secret material 
          (Solid, hydrophobic, moisture resistant, non-flamable, electrically non-conductive, ...)

Mickle T. already provided a working solution with the Dewar flask:

"Some words about tilting.
One of my simple statistical voltage references with the Dewar flask isolation has no orientation effects at all, i.e. 0.0 ppm in any direction. LM399 heater current < 5 mA."

  https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg430734/#msg430734

[iMo]

Aerogel_X7           1.900              36

Filling in the air gaps in the original cap with the wax (ie. the H1 type in my above link) will a) fix the cap, b) increase the heat capacity of the cap (thus less sensitive to surrounding air flow temp changes), imho.

[bsdphk]

I believe the temperature change of the chip is not the issue I believe it's the temperature change of the Kovar/Copper (35uV / K) junction on the pcb. A 0.3K temperature change of one Zener lead would cause the 10uV shift.

I agree that contact between the cap and the leads is the major effect, and I see no practical way to eliminate contact, apart from loosing the cap entirely.

The cap absolutely makes sense for mass-production applications, but for one-off nerdery, it should be removed and a thermal solution which touches neither the LM399 nor the solder joints to it at any point should be implemented instead.

(Insert obligatory rant about thermal impedance vs. thermal insulation here)

[ramon]

It's not Aerogel. Aerogel have similar specific heat, but usually has half TC value (around 10-20 mW/mK).
Not only it is expensive, but difficult to buy.

The TOP secret material is just two times that TC number, but 1/100 or 1/1000 the price of aerogel.
It is still relatively easy to buy (but not to easy to produce).

[chekhov]

Looking at Mickle T. photo I noticed (trivial observation however) that we may talk about LM399 that has two cap types - just locked (latched) one and the one melted on the top (not sure whether it is genuine or not).
Quickly looking at three LM399 with melted cap I found one where LM399 sits tight inside and don't move at all.
Also single LM399 that I have with latched cap (desoldered from Keithley stuff) has much smaller holes for a legs, and remember how legs are usually bent to fit 2.54mm footprint on PCB - it also helps to stop cap movement.

As a side idea regarding how to prevent side cap movements, maybe put a little 'ring' on each leg made from PTFE wire insulation for example (so that it eliminates space between each leg and case). Yep it's still a thermal conductor, but at least decreases the movement.

[Kleinstein]

The cap is already not that bad, though they could have made it better. If wanted one would have avoided the cap toucing the wires, if the cap position is defined by the actual case and not the leads.
The other point is that the LM399 is sometimes mounted with short leads and sometimes mounted with longer leads to reduce the heat flow through the leads.  With longer leads it would be more convenient if the cap would extend further down, covering the leads to some extend.

Filling the gap has however another problem: it can cause mechanical stress to the case and this way effect the reference. So the somewhat loose cap that can move is a kind of a compromise.

A separate cap directly fixed to the PCB may be an option.  It should just keep the air space small (e.g. <3 mm)  to avoid turbulent convection. 
I don't know the exact distance limit. With dual glass windows things change a bit at some 10-15 mm, but here the temperature is quite a bit higher.

The jumps with the moving cap are visible, but still not very large. The more permanent part is more like comparable to the usualy jumps from popcorn noise (e.g. 3-5 µV).

The table of termal conductivity numbers have the low conductivity materials  wrong by a factor of 1000. So EPS at about 35  mW/m/K.

[iMo]

If I were equipped with a precision instrument (like many experts here) I would open a 399 cap, fill it with the Carnauba wax (melted with soldering iron). It would hopefully fill the gaps around the legs as well.
The only question here is how are the electrical params of the wax.
Interesting experiment, indeed..

..
Filling the gap has however another problem: it can cause mechanical stress to the case and this way effect the reference. So the somewhat loose cap that can move is a kind of a compromise.
..

The wax is soft, especially at 90degC. No problem with mechanical stress..

[ramon]

The table of termal conductivity numbers have the low conductivity materials  wrong by a factor of 1000. So EPS at about 35  mW/m/K.

Yes, sorry about that. I have copied and pasted from many different sources and with different units.

I have not found yet a single source (document) that includes metals, non-metals, alloys, polymers, different kinds of timber, waxes and common chemical powders.

The NIST documents elaborated in the 60s by Purdue university (Ho, Powell, and Liley) have an incredible amount of research about thermal properties of materials from 0K to 10^4K, but just cover the elements (no single alloy, modern thermoplastic, exotic wood, waxes or common household cleaning materials).

Does anyone knows of some source of data (with the same level of detail that those NIST documents) about thermal properties of widely available modern materials? Thanks.

[ramon]

(About the Wax) There are two critical temperature points: the melting point and the glass-transition point (Tg). I did some research, a few years ago, and bought this bottle in a local store. The nominal melting point of Carnauba wax is just 82-84C, I doubt that any wax have its temperature resistance above 95C. I bought it because it has another a very special property: moisture barrier

... but carnauba wax composition has acids, so you cannot apply it directly to any wire or circuit!

Kleinstein talked about the lead length. Pictures of some commercial reference with paralleled 8x LM399 devices (I cannot find now the post) shows that they soldered the device with the longest leads. I don't remember for sure, but did National or TI recommended this setup on some older application note or datasheet due to the possibility of Kovar pin to case hermetic sealing cracking issues?


Edit: That paraflint H-1 is wax + additives. And they don't say what is that additive.
        I am not brave enough to use a material (without knowing its exact composition) into direct contact with the Kovar leads.
        MgO powder is a known electrical insulator and has moderate-to-high thermal conductivity.
        What is the filling material used in the NTC/PTC steel probes? and on those high power resistors with aluminium cases?

[Kleinstein]

The datasheet suggest the cap can sit directly on the PCB. In this case one could just glue the cap to the PCB. The cap gives some minimal length for the wires.  The point with longer leads is to reduce the heat loss and also give it more room to bend , as there are 4 leads and thus more than the 3 to define a plain.
AFAIK some of the 6 digit DMMs also mount the LM399 with longer leads.

Anyway the LM399 good thing about the LM399 is that it is easy to use and still gets reasonable performance. The improvments possible is also limited - I don't think the popcorn noise has anything to do with the cap.

[iMo]

... but carnauba wax composition has acids, so you cannot apply it directly to any wire or circuit!

Those acids are usually activated at higher temperatures (like in the soldering rosin flux). Needs some further investigation, though.
How you would use it as a moisture barrier when acidic?

Anyhow - an alternative - the paraffin H1 (see my link above, 98degC melting point).

[ramon]

How you would use it as a moisture barrier when acidic?

With many different layers, each one for specific purposes:

  - Innermost layer material: MgO powder (electrical insulator, thermal conductor/stabilizer)
  - Middle layer material:  Steel & Carnauba Wax and/or PCTFE (magnetic field & moisture insulation)
  - Outermost layer material: TOP Secret (thermal insulation and buffering)

[iMo]

..
Edit: That paraflint H-1 is wax + additives. And they don't say what is that additive.
        I am not brave enough to use a material (without knowing its exact composition) into direct contact with the Kovar leads.
        MgO powder is a known electrical insulator and has moderate-to-high thermal conductivity.
        What is the filling material used in the NTC/PTC steel probes? and on those high power resistors with aluminium cases?

The Metrology section is for brave people only.. 
I doubt the additives in the paraflint H1 are acids or something electrically conductive. The acids in the natural waxes are there because everything living contains acids..

PS: if you look at Mickle's ref design he potted the divider resistors into a large paraffin block.. Not because the moisture but because the large temperature capacity of the paraffin..

[dietert1]

Yesterday i received a 32 year old HP 3457A and lucky enough the mains filter exploded after 10 minutes while cover was open. So i started with maintainance and mods. I fixed the cover of the reference. When checking today i found there were more loose parts, and here they are, between the cap and the board, covering two of the LM399 pins. Originally they may have served as spacers and meanwhile shrunk with heat. Don't want to solder the reference, but there are doubts about these short, straight pins. Think of the four leg table that will never stand still. But that's another story.

Regards, Dieter

[iMo]

..Think of the four leg table that will never stand still. ..

Cut off a leg and with three legs it will be rock stable.. 

[dietert1]

No cutting legs here! The legs should be more like those of our dog, with bends in them to adapt with ease.
Anyway those reference modules are small and hopefully there won't be a lot of mechanical stress on the LM399.

Regards, Dieter

[dietert1]

For completeness here the LM399 reference module from a 40 year old HP 3456A (model with fan). No spacers, but wider landing pattern. LM399 cap is held tight by the bent wires.

Regards, Dieter