[volt-nuts] MAX6350 ?

[Poul-Henning Kamp]

In message <84314.1283455235 at critter.freebsd.dk>, Poul-Henning Kamp writes:

>Maxim has a buried zener voltage reference called MAX6350, but it is
>specified for "instant precision", not long term drift (30ppm/1000h).

Replying to myself here:

First a bit of background:  I use voltage references in electronics
from time to time, and have kept a lazy eye on new products.

The MAX6350 ended up in a 12bit ADC application we could never
expect to calibrate over its decadal lifetime.   Recently one of
these gadgets came back to my lab and I was pleased to see the Vref
hold op really well.

There are other new-ish Vref products and, more interesting, new
technologies, such as Analog Devices "XFET" etc[1].  None of these
Vref's are really spec'ed for metrology grade apps, their long-term
specs are really just "burn-in" indications.

I could not help wondering if any of these new chips had metrology
potential. If properly ovenized and protected, they might even give
the 20 year old LZ1000ACH design a run for the money.

I put a MAX6350 from the parts-drawer in a breadboard, added some
1µF tantallum capacitors, used a LM358 opamp as pre-regulator to
feed it 10V supply, referenced to its own 5V output (the MAX6350
has pretty bad supply sensitivity), put a piece of bubble-wrap over
it, and left my HP3458A chug along for a week.  I did nothing to
protect the 3458A's input electrodes from airflow and nothing
to deal with seebek etc, just random lab-wires used.


I have filtered out 91 of 172k measurements, because they were more
than 5 stddevs out, I have not tried to find out what caused them,
probably people turning flouresent lights on and off.

Here is the result as 3 hour averages+standard deviations, relative to
the global average of 4.999705193 V

	http://phk.freebsd.dk/misc/20100911_max6350_1.png

And here also with the original raw data:

	http://phk.freebsd.dk/misc/20100911_max6350_2.png

Since this is a 5V reference, the Y-scale corresponds to +/- 1 PPM.

The 3-hour standard deviation is quite constant at 0.12 .. 0.18 PPM
and the average is +/- 1PPM over 8 days.  Comparing to other data
collection in the lab, I am confident that residual is mostly tempCo.

All in all: I'm pretty impressed.

Poul-Henning

[1] Analog got a bit smart&silly with the long term spec, and got
soundly and deservedly beaten up by linear.com (DN229) for that,
but that doesn't mean that the XFET's are not interesting, it only
means that they are not revolutionary.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.