Resistors for LM399 references

[RoadDog]

I’m preparing to build a few references based on the 399 as it’s easily obtainable at the moment, they’re affordable to build and they’ll be good enough for my needs for the time being.

I plan on building three so I can compare them over time and know if one takes off on an excursion.

My question is about resistors. Would I be better off buying higher priced low TC resistors such as 5-10ppms or would I be better buying a good quantity of cheaper resistors and TC matching them? I understand it’s the ratio between the resistors that matters. I don’t know that I’ll save money doing so but if I can match them better for ratio and TC by going through a bunch similar to how Conrad Hoffman describes building a KVD in the mini metrology lab article then I might go that route. I have a null detector so I could rig up some batteries and a heat bath of some sort and go through the process.

I also have some Fluke resistor networks I scavenged from 8846 boards that TiN used in some of his standards he built. I’m thinking I want to hold onto those for now and use them at a later time. These will be my first references. I’d like to save them for future projects when I have more experience and knowledge for something a little higher precision.

[Kleinstein]

The resistors directly use with the LM399 are not that critical. I would consider 25 ppm/K resistor good enough and even 100 ppm/K may be OK, though possibly boarderline.  For the 2 resistors used for the voltage gain only the ratio matters, for the 3rd resistor to set the reference current it is the full value. The TC of the resistors is also one 1 aspect. Chances are the more relevant parts are the long term drift and maybe the effect of stress - the problem here is however that there are few useful specs in the area. So the TC is often used as a more general quality parameter, assuming that low TC resistors would also be good with the others.

It can make sense to get the gain from multiple equal resistors  (some, like Susumu RR and similar are relatively cheap when bought in 100). There are also resistor arrays that can provide rather stable ratios.
It depends on the application if the raw 7 V or an amplified signal (e.g. 10 V or 14 or 20 V) is more relevant.

There is the point where a 2nd LM399 would make more sense than using very expensive resistors.

[Gyro]

The old NatSemi->TI AN184 (References for A/D converters) gives a good breakdown of the error contribution of resistors in various reference circuits...

https://www.ti.com/lit/an/snva510b/snva510b.pdf

[RoadDog]

Thanks for the input. I’ve been reading over the threads here, some years long with a ton of info, and reading about builds other folks have done that I can find online. Was just an idea that popped in my head for the gain section to 10V after reading Conrad’s articles. He achieved some amazing performance with cheap resistors. He mentioned that long term stability was unknown.


I’m working on these and a 335a I picked up as projects over the winter. I think I can get the references  close with my current equipment and then can have a friend measure them with an 8 1/2 digit meter he uses at work and have him do so maybe once or twice per year to see how they act over time.

[MrYakimovYA]

There is the point where a 2nd LM399 would make more sense than using very expensive resistors.

Excuse me, could you clarify it in more details, please. I thought that combining more than one voltage reference cause to reducing the noise not the output stable...

[KT88]

The 3 dB noise reduction from paralleling 2x LM399 would give you more bang for the buck than investing in expensive resistors…

[iMo]

The resistors in the 7->10V stage are critical from TC point of view, indeed.
In case you want, say, < +/-1ppm/C TC of the 10V ref, the "ratio TC" of the resistors has to be less than +/-0.5ppm/C (the LM399 itself is typically -0.5ppm/C). Their "ratio TC" applies because the 2 resistors create a "divider", thus the "absolute TC of the resistors" is not that critical, but their "ratio TC" or "matching TC" matters.
With "25ppm/C" resistors in that stage you may get anything between + or - 50ppm/C at the 10V output easily (+/-500uV/C).
I was using the LT5400 4x10k resistors (DS: "matching TC" 0.2ppm/C, absolute TC 8ppm/C) getting aprox 10.54V out, with 25ppm/C resistor in the 399 "cathode", and the overall TC was less than -0.6ppm/C with many 399's I tried.

The "long term stability" and the "voltage reference noise" are different topics, sure..

[branadic]

Most people tend to miss how to compensate the overall t.c. with a bit of copper in the 7 V --> 10 V stage. Just use a proper resistor network (TOMC, NOMC, TDP), set the coarse gain with the given resistors in the network, fine-adjust the gain with parallel resistors to individual resistors of the network and finally t.c. compensate the 10 V output with copper in one of the resistor branches, depending on the sign of the t.c.. This is also the reason why a bootstrapped zener (powered by the 10 V) is to be preferred.

-branadic-

[iMo]

With copper TC compensation you would need to calculate the "compensation resistor value" == length and diameter of the required copper wire, while you know that TC of copper is aprox +3930ppm/C. An excel tool would help (even though the practical implementation would require pretty long pieces of copper wire, imho)..

[branadic]

It's not that such excel tools don't exist, see attachment, but this file was posted elsewhere before here on eevblog.

-branadic-

[RoadDog]

I missed a bunch of these responses. Some good info. I ordered some Vishay 10ppm thruhole resistors(20k & 9k) along with the 399's and LT1001's to build the basic datasheet circuit. I'm going to build some burn in boards, watch them for a thousand hours or so while I work on designing my boards and fine tune the outputs when I build them. I'll have to settle with as good as I can get with my HP 34401. Then maybe send them to someone after a while and get better measurements of their output.

Thanks for the info folks. I know there is a ton of info in the lm399 megathread but it gets hard to find what you want sometimes. So many pages spanning years.