How to age a voltage reference?

[NaxFM]

Hi everyone,
I want to experiment with the stability over time of various voltage references which don't use ovens (mainly the three versions of the REF102).
For example, the datasheet for the REF102 specifies a stability of 20 ppm over 1000 hours, which are 40 continuous days of waiting. I want to test the medium-long term stability, but waiting for 40 days just to have a baseline to start doing the real measurements is a bit too much, especially if I want to test other references. I just can't wait months before doing measurements each time we want to try a new reference. It's for a univeristy project so i really don't have all the time in the world.
Is there a way to accelerate this aging process? I know there is because manufacturers do it all the time, but how to do it at home? I have a thermal chamber, but don't know the correct procedure...
Is it sufficient to cook the boards for a few days at a set temperature while powered on or maybe it is something more involved and difficult to achieve?
What do you say? Do you have any experience with aging components?

[Kleinstein]

A higher temperature speeds up most of the aging processes. However there can be also some hysteretic effects and moving to a different equilibrium. So the stabilization can be faster, but towards a different final state. One the temperature is reduced new stabilization to the changes condition would start. So extra heating can also make things worse.
A different final state can come from thermal mismatch (e.g. plastic and metal mix) and also with humidity effects, as a higher temperature usually also means lower relative humidity.

With a reference in a plastic case, chances are aging at a higher temperature could do more harm than good. At least it can not fully eliminate the time needed at the normal temperature.
Even with a ref in metal case there can be some thermal stress between the silicon and metal and a polymer die attach that can creep. So a higher temperatur may help, but could also cause more problems. I would at least avoid a much higher temperature - so more like some 20-30 K above normal operation to get about a 4-8 fold speed up for the initial part.

[Andreas]

Hmm,

you do not write which Package.
The (now obsolete) metal can package (M-Version)
The (now nearly obsolete) DIP 8 package (P-Version)
or the SO-8 package.

Which stability do you want to achieve?

With plastic packages (Epoxy) or SMD-packages (soldered on Epoxy-PCBs) you have additionally the problem that Epoxy absorbs water from air humidity.
The Epoxy then swells and this has a influence on the chip and finally on the output voltage.
The time constant for this is rather large (in the 5-7 days range).

So the sad truth for plastic packages or SMD packages on a PCB in an uncontrolled environment (temperature + humidity) is that you need minimum 1 Year to observe the changes between summer (60-70% rH in my LAB) and winter (40% rH).

I have one REF102AP (DIP8 plastic package) in my ageing box.
Due to self heating (30-40 deg C PCB-temperature) the references are relatively dry.
During measurement each day the PCB is heated to 50 +/- 0.1 deg C further drying the references.
X-Axis is in days Y-Axis is in PPM since first measurement.
on the light blue line you can see the seasonal changes due to environment humidity changes.
On  ~day 3300 I had a 9 day power outage which shows what happens under full humidity conditions. (Peak to 20 ppm)

Edit:
So now you can imagine what happens if you do a "Burn In" with a plastic package or a SMD-package soldered to a PCB:
The package will run dry. In normal conditions afterwards you will have   ~1-2 weeks large drift until the package stabilizes to the %rH value of the environment.

with best regards

Andreas

[alligatorblues]

Maybe you could connect the ICs to sockets on like a breakout board that can put the ICs in parallel for input, and perhaps age 10-20 of them at the same time. Are they expensive? I mean, is ig feasible to pu chase that many at one time?

Then you just have to wait 1,000 hours once for all the ICs. Just a thought. 

[Andreas]

Hello,

depending on the ageing rate which you want to achieve 1000 hours may be not enough.
Often you need 5000 - 10000 hours until the reference ageing stabilizes to a relative low level.
The 1000 hour figure is only a "contest" test condition.

Each reference has also its individual ageing process. So some age faster others need much longer.
It is also dependant on the package of the device. And even the "same" packages of different manufacturers have different performance.

Here a comparison of some MAX6250A and  one AD586MNZ.
While the AD586 stabilizes after ~1000 .. 2000 hours the MAX6250A all still drift with ~10-15 ppm/year.

Note: x-axis is in days not hours.

with best regards

Andreas

[iMo]

Some time back I messed with about 60 epoxy DIL references and the result was: do not mess with epoxy references, it is waste of time.
All those references you mentioned above were created 40++ years back for at that time new or mainstream 12bit DACs and ADCs working with +/-/5/10V levels (like 2.5mV LSB resolution). In such an app they may work fine. Not suitable for anything better than that..
PS: also those old ref01, ref102, ad586 and others in metal are good for 12bit only. To see the nice pictures like Andreas shows requires a lot of luck today (not counting his tremendous effort and $ with picking up the couple of right ones and observing them for years).

[Andreas]

To see the nice pictures like Andreas shows requires a lot of luck today (not counting his tremendous effort and $ with picking up the couple of right ones and observing them for years).

Hello,

for my ageing box: the only effort is keeping them at constant temperature (50 deg C during measurement).

Of course you are right with the EPOXY packages and humidity. But there are also differences.
I have measured 0.5 ppm/% rH on 2 LT1027CCN8-5
From Lars I know that he measured between 0-0.3ppm/% rH on some AD587JN in plastic package.
But I do not know how ROHS has affected the used EPOXY in the mean time.

And since humidity influence on EPOXY has a time constant of > 1 week it is hard to measure.

with best regards

Andreas

[iMo]

It depends what REF102 packages the OP wants to age.
I would not consider epoxy packages for any serious application (Q: what is your app??).

I saw somewhere in a DS or an AN that they recommended a "burn in" of the 10V reference such you simply pull 10mA off its output (1k resistor against GND), with its input at say 30V.

The die temperature will then be something like

T_die = (30V * 0.0014A + (30V - 10V) * 0.01A) * theta_JA + T_amb = 242mW * 110 degC/Watt + 25 degC = 51.6 degC

for PDIP8 epoxy package.