Low cost voltage references long term stability

[splin]

Does anyone have any experience, data, or any links/'references to long term stability measurements of cheap references? The datasheets for more costly devices tend to include typical values for long term drift but the low cost devices, perhaps not too surprisingly, don't - unless anyone knows otherwise?.

By low cost I mean < $0.5; the most common seem to be TL431 variants but any type of device which includes a reference such as voltage regulators, ADCs, DACs, microcontrollers would also be of interest. Even zener diodes data would be interesting - I've looked but not found any studies of low cost types (there is some discussion on volt-nuts but only of temperature compensated types such as 1N829 which are quite expensive now).

TIA, Splin

[rossmoffett]

We use Fluke process calibrators at work for a variety of different field calibration applications.  They're recommended to be calibrated yearly to be NIST traceable.  I sent one in after five years expiration of calibration, was within 0.01% voltage reference of 5VDC.  I don't know what drives the reference internally, but that's at least one point reference for what probably contains an ultimately cheap component reference inside compensated to temperature drift.  The device itself retails for north of $3k though, of course.

[Andreas]

Hello,

the LM385 (or derivates e.g. LT1004) are used in some portable DMMs.
It is specced with 20 ppm/kHr typical.

But usually this parameter is only valid for the nowadays obsolete metal can housing.
With plastic housings you will have additional humidity drifts.

On the other side: with a T.C. of typ 80 to max 150 ppm/K a 20ppm parameter is difficult to measure.

With best regards

Andreas

[Solder_Junkie]

I made a low cost multi-meter checker using an MAX6126AASA25+, it's a surface mount reference chip of 0.02% accuracy when manufactured.

In common with some other low cost references, the long term stability is quoted as 20ppm per 1000 hours.

It's fine for my purposes, which is just to check basic meters.

The lower cost LM285Z series are also shown as 20ppm per 1000 hours, but with a much lower tolerance to start with.

The MAX I/c costs around 16 GBP and the LM285Z around 1.5 GBP

[splin]

Hello,

the LM385 (or derivates e.g. LT1004) are used in some portable DMMs.
It is specced with 20 ppm/kHr typical.

But usually this parameter is only valid for the nowadays obsolete metal can housing.

Coincidently I found this yesterday, reading Linear Technology's AN82-2 which states:

Most of the long-term stability figures shown in LTC reference data sheets are for devices in metal can packages, where assembly and package stresses are minimized. You can expect somewhat less performance for the same reference in a plastic package.

Looking at various datasheets though LT mostly seem to specify which package the long term stability figures apply to. The LT1027 datasheet clearly shows the metal can package as obsolete but only quotes the stability figure for that package - presumably they don't think it's worth the cost of characterising the plastic packaged versions.

The LT1004 and LM385 datasheets though don't state the package so as you say its almost certainly the obsolete metal can part. I expect its a specification that's just been copied across from the old LM385 datasheet and the newer plastic packaged parts have probably not been characterised - hardly worth doing for $0.2 parts.

As Bob Pease stated in his musings "What’s All This Long-Term Stability Stuff, Anyhow?"

http://electronicdesign.com/archive/what-s-all-long-term-stability-stuff-anyhow

"This usually makes the customer happy and/or shuts him up because (a) it is usually true, or close to true; and (b) if he wanted to get data, it’s a lot of work for him! And he doesn’t usually come back quickly. If at all."

Deliberately quoted slightly out of context (he was talking about long term extrapolation of the 1000hr figures) - but I think it sums it up nicely. My guess is that anyone who really cares about long term stability will do their own testing - those that don't aren't going to know if the datasheet figures are remotely accurate. In any case they are almost invariably 'typical' figures so they aren't guaranteed anyway.

An example where the figures can't be trusted is TI's REF5050-Q1 (the automotive qualified part) datasheet. The headlined features include:

EXCELLENT LONG-TERM STABILITY: – 5ppm/1000 hr (typ) after 1000 hours[/b]

Wow. Remarkable for a $1.6 part. But further down it quotes:

LONG-TERM STABILITY (SO-8 parts):
0 to 1000 hours         90 ppm/1000 hr
1000 to 2000 hours   10 ppm/1000 hr

Hmm so not 5ppm/1k hr but still excellent. But most interesting of all there are some graphs (figs 22 to 24) showing the results of testing 96 samples over 2000hrs. Really useful - if only all datasheets published this information.

But It seems it was a mistake. The earlier revisions of the datasheet for the standard version of the (not automotive) REF5050 was exactly as above. The latest revision however states the long term stability as:
0 to 1000 hours       100 ppm/1000 hr
1000 to 2000 hours   50 ppm/1000 hr

Ok. A lot more credible. Also the stability graphs (figs 22 to 27) are now totally different. So almost certainly a mistake and the automotive part datasheet didn't get updated when the standard part datasheet was changed.

With plastic housings you will have additional humidity drifts.

On the other side: with a T.C. of typ 80 to max 150 ppm/K a 20ppm parameter is difficult to measure.

With best regards

Andreas

True. But the datasheets for many references show that the temperature coefficient is typically 0, or near 0 at around room temperature, increasing either side so perhaps not so difficult.

[Andreas]

True. But the datasheets for many references show that the temperature coefficient is typically 0, or near 0 at around room temperature, increasing either side so perhaps not so difficult.

Hello,

unfortunately the references cannot read datasheets.
So they also cannot obey the typical curves.

In theory (with perfect design and adjustment and no tolerances in process)
the positive PTAT curve and the negative VBE voltage may compensate
with a zero at room temperature.

Practically the voltage references are only tested at 25 deg and min + max temperature limit.
The min and max voltage of these 3 points divided by (max - min) temperature gives the so called T.C.
The oscillations and gradients in between the measurement points are undefined.

With best regards

Andreas

Edit: my 4.5 digit DMM with a LM385 reference has a 84 ppm/K tempco around room temperature.

[dannyf]

Edit: my 4.5 digit DMM with a LM385 reference has a 84 ppm/K tempco around room temperature.

That's very interesting: the reference has different voltage output at the same temperature.

It means that it has memory - its output is not only a function of its current environment but also its previous environments.

[Andreas]

Hello,

that only means that the temperature sensor (outside the DMM in this measurement)
has not exactly the same temperature than the reference.

With best regards

Andreas

[splin]

True. But the datasheets for many references show that the temperature coefficient is typically 0, or near 0 at around room temperature, increasing either side so perhaps not so difficult.

Hello,

unfortunately the references cannot read datasheets.
So they also cannot obey the typical curves.

In theory (with perfect design and adjustment and no tolerances in process)
the positive PTAT curve and the negative VBE voltage may compensate
with a zero at room temperature.

Practically the voltage references are only tested at 25 deg and min + max temperature limit.
The min and max voltage of these 3 points divided by (max - min) temperature gives the so called T.C.
The oscillations and gradients in between the measurement points are undefined.

With best regards

Andreas

Edit: my 4.5 digit DMM with a LM385 reference has a 84 ppm/K tempco around room temperature.

LOL. Yes you're right for any specific device but the temperature curves shown in the datasheets are *supposed* to be typical. Of course its entirely possible to get a whole batch of parts which don't remotely correspond to the published typical specs - particularly if the manufacturing process has been changed without the datasheet being updated. Or the manufacturer isn't too scrupulous about publishing accurate datasheets.

I would expect though that usually it should be possible to select some parts for testing which have reasonably low TCs at room temperature (where the data sheets show the temperature compensation should provide low TC around room temp) - I was asking about low cost, <$.5, after all.

Thanks,

Splin