What kind of voltage references 'burn in'?

[ivan747]

I was making a quick experiment with a simple voltage reference design using a TL431 and an LM317 as a simple current source for it. I left it powered on for 72 hours and the thing drifted quite a lot (from 4.096V to 4.112V, about 4 LSB in a 10 bit ADC)  in the first 24 hours. Now it seems like it drifts a lot less, 1 LSB at most. I used a military grade resistor (0.5%) and a simple 1% resistor to set the voltage and used a trimmer as described by The Art of Electronics. The current source comes straight from the LM317 datasheet and doesn't use a fancy resistor at all. Oxidization and contact resistance might have to do with the drift but I can't tell. I would have to solder this and test the circuit for something like 1000 hours to really get into conclusions.

I liked this experiment. The performance is quite good, but the experiment left me wondering:

What kind of components and voltage references get long-term stability as they are left powered on (i.e. burn in)?

[SeanB]

All of them drift as they age. This is reset when you solder it again, which is why you find references specced for drift, so much in the first year, and a lot less generally in the following years. All components have this, just in some cases it is important, and in others negligible. That is why the best references are assembled, checked, then powered on for a period of months then finally calibrated or certified. That way most of the initial drift is going to occur, and any that will have short term drift are found, along with those that are going to fail early.

[babysitter]

You better don't expect that to happen.

Some parts might exhibit a hysteresis during on-off-cycling which will not appear if they are left powered on, and while it is not a safe way, a part that exhibits just little drift when left on for a while *can* possibly perform better in the future than a part that exhibits more drift in the initial phase, but don't take that for granted. Keeping turned on means temperature rise means aging and that means additional drift.

Also, some guys figured out ways to heat and cool voltage references, e.g. LTZ1000 superduper reference in a damped oscillation scheme like a ringing amplifier, which helps with that. But not with the additional aging due to higher temperature. You could try to do similar things to your reference but sucess is not guaranteed.
What works with one part doesn't necessarily work with the other, even from the same kind.

As being of more value than a burn-in i expect to come from a clean power supply, well filtered and already well stabilized, air current blocking and maybe temperature control, as done with LM399 or LTZ1000 references or in a similar fashion, quarz oscillators.

You might plot a voltage-over-time of your reference, but remember, your multimeter does that too.

Hendrik

[Dr. Frank]

... TL431 and an LM317 ..
.. on for 72 hours and the thing drifted quite a lot ..
... military grade resistor (0.5%) and a simple 1% resistor ..
.. Oxidization and contact resistance might have to do with the drift but I can't tell...
I would have to solder this and test the circuit for something like 1000 hours to really get into conclusions.

I liked this experiment. The performance is quite good, but the experiment left me wondering:

What kind of components and voltage references get long-term stability as they are left powered on (i.e. burn in)?

Hi,

your experiment for sure is quite instructing, to get a "feeling" for drift phenomena.
But, simply read the specifications for the different components, how they drift individually.
If you also analyse, which influence the exterior components have on the stability on the reference, you can estimate the total drift parameters.

I wonder, what kind of "military resistors" you've got.. there's obviously no classification / standard for such a term.
Mainly, the reliability or the stability is specified, but not the tolerance.

There exist 'Military' or "Automotive' requirements for components, presently I don't remember the military standard, but for automotive, there's the AEC Q 100, 200, etc.

First of all, the reference itself has to be capable of delivering stability, over temperature and over time.

A simple stabilizer contains a simple Bandgap, perhaps, but is neither designed for good temp.-, nor time stability.
You'll need either a well designed bandgap reference, as in the REFxxx types, or an ovenized, buried Zener reference, as LM199, LTZ1000.

The external components, mainly the resistors, have to have low T.C. (around <= 5 ppm/K) and low long term drift.
Therefore, thin film resistors are not sufficient (although perhaps being military grade reliable), but wire wound and metal foil precision  resistors are required.

A burn-in of the assembled PCB may be helpful for reduction of stress in the solder junctions.
But this won't help for those simple stabilizers you've mentioned. They are simply too unstable from the principle of their design, that a burn-in would help.

To my experience, a burn-in or a monitoring is necessary only for the highest level of guaranteed stability, as Fluke or HP/agilent delivers, but  if you cleverly choose your components, the additional effect of pre-aging is neglectable for our non-profit purposes.

For the case of the LTZ1000, it's totally sufficient to design the standard circuitry from the LT datasheet, without further tricks, and you'll get an ultra stable reference, rarely anything will beat.

Well, and it's also necessary, that you have measurement equipment which is more, or at least equally stable, to rate the stability of your D.U.T.
That means, in this case, you need a high grade DMM,  or alternatively you build 2, better 3 identical D.U.T.s to compare them against  each other. If they show a big drift, they will drift apart. If they are stable, they will stay in line.

Regards Frank

[ivan747]

The resistor I am using is a metal film resistor with a TC of 50ppm/ºC. They seem to follow MIL-R-10509. I called this miitary grade because that's more or less what they have to say about it:

http://www.irctt.com/pdf_files/rn.pdf

The actual resistor is 80K, RN55C.
It has a lot more data about these resistors than your average 0.2 cent 0805 Panasonic resistor "datasheet".

Great responses so far 

[Dr. Frank]

The resistor I am using is a metal film resistor with a TC of 50ppm/ºC. They seem to follow MIL-R-10509. I called this miitary grade because that's more or less what they have to say about it:

http://www.irctt.com/pdf_files/rn.pdf

The actual resistor is 80K, RN55C.
It has a lot more data about these resistors than your average 0.2 cent 0805 Panasonic resistor "datasheet".

Great responses so far 

yep, it's all about the stress test requirements (heat, humidity, temp shock,..), which makes those resistors "military" grade.. the 50ppm/°C T.C,. simply is a characteristics of Thin Film technology. Anyhow, those TF resistors will give average stability onlyfor volt references, but will for sure determine its performance mainly (not the volt ref element itself, as intended).

To dig deeper, use an LM399 instead, with better resistors, e.g. 3ppm/K wirewound.

LTZ1000 circuitry and housing requires some clever ideas (and money), but is not that complicated.

Frank
 

[mikes]

You can get a MAX6241 which puts out 4.096 V directly, 0.02% initial accuracy, and good tempco for less than $10.

[ivan747]

I find the resolution a bit low for reading the mains voltage, only 3 decimal digits can be obtained.

Oh wait, I can use a voltage divider that takes advantage of the full range of the ADC 

You can get a MAX6241 which puts out 4.096 V directly, 0.02% initial accuracy, and good tempco for less than $10.

About the 4.096V reference, I am sure there are many competing products out there, but I want to keep the cost of this to the minimum, even if that means trimming the reference. One disadvantage, though is that not anyone has access to a high accuracy multimeter with enough resolution to show 4.096V. You need a 6,000 count multimeter with the accuracy you want to trim the reference to, at least.

I would like to design this as a OSHW project. I originally wanted something you could publish in a magazine (single sided, generic though hole parts, nothing more complicated than a PIC or LCD) but the performance and functionality will be drastically better if I use more specific parts and double sided boards.

The original reason for the experiment was to see if a cheap generic voltage reference was good enough for our purposes.


-----

Now back to voltage references, they seem to be very closely related to statistics and math in general, at least the selection process. As far as I know, these LM399s and LTZ1000s are used for basically all multimeters from 5.5 digit onwards. Even the best grade parts have limitations in long term stability, which really is the only influencing factor because initial tolerance is trimmed and the reference is heated to a constant temperature,

These limitations in long term stability mean the manufacturers must filter their reference and only choose the most accurate ones after burn in.

From what I have understood, 'burning' a reference doesn't mean it is going to stabilize, it just means you can select the stable ones and ship your product with those. For me, it's not worth it. I will save more buying a $4 part rather than paying the electric bill of having 100 references powered up for 1000 hours plus a multimeter and a lot of my time.

[AlfBaz]

What kind of components and voltage references get long-term stability as they are left powered on (i.e. burn in)?

Fearing I'm just adding something stupid here, I was looking at the specs for a LM399 and long term stability was specified as ppm/SQRT(kHrs)
Being a noob, this didn't make a lot of sense so I entered this into a graph and you can see that whilst the drift never flattens out, initially the drift is quite large.

In the 1st 1000hrs drift is just over 30ppm, 2nd 15ppm, 3rd 10ppm, 4th 8ppm and so on

[ecat]

I'm quite happy with my spin of a poor man's 10.000V Voltage Reference,

https://www.eevblog.com/forum/projects/voltage-reference/msg198159/#msg198159

I get within  +/- 30uV wobble between readings which could easily be due to my aged Solartron 7150 Plus (6.5 digits).

The values you see in the linked post were just the chip and a noise reduction cap on a breadboard powered at +/-15V from 7x15 regulators. I've since made a half decent pcb and the recorded voltages dropped about 400uV. Stability since March, with rising spring temperatures, is easily within 100uV and the only additional "temperature compensation" device I use is a small padded box to cover the pcb. Of course the stability of my meter could be complimenting the chip 

I bought all three AD588BQs from this guy if you're interested

http://stores.ebay.co.uk/sisitronic?_trksid=p2047675.l2563
http://www.ebay.co.uk/itm/AD588BQ-High-Precision-Voltage-Reference-IC-/120930768496?pt=LH_DefaultDomain_0&hash=item1c28091670

Are they the real deal? I have no idea. I bought three so I could get some idea of the range distribution and as you can see my long time since calibration meter shows worst case 10.0007 and 10.0021V in the link.  2.1 - 0.7 = 1.4mV, so within the BQ specification. Worst case I'd say these could be AQ grade, still good enough for me.

Advantages of the AD588:
You get a buffered Vref, current source, trimmed resistors and 2 additional, good enough for the job, opamps suitable for Kelvin connection in a single package.