zener serial wiring and tolerance

[001]

Another stupid question
sorry

I have 3 zeners 0.02% longtime stability and 0.0005% per Celsius
What I can get with all three zeners wired serial?

Question #2
Can I achive "zero" temperature coefficient  If I wire TWO zeners but one of whem backward?

[David Hess]

A backwards zener diode is just a diode.

Wiring all three in series may or may not yield a gain in precision but it will increase the signal to noise ratio because the voltage will be tripled but the noise will only increase by Sqrt(3) times.

[Conrad Hoffman]

Those sound like reference zeners that already have a diode in series. Usually one fine tunes the current to get zero TC over a narrow range.

[Brak]

As Conrad says, the 1N821-1N829 and 1N4565-1N4584 series diodes are sold as "Temperature Compensated Reference Zener Diodes".  They already have a forward biased silicon diode in the package.  Or maybe more than one, not sure about that.

He posted a little bit about fine tuning the current to get zero temperature current over a narrow range at:

https://www.eevblog.com/forum/metrology/old-fashioned-zener-10v-reference/

Brak

[Brak]

According to Current Sources & Voltage References by Linden T. Harrison these two images show how a temperature compensated voltage reference diode is manufactured and it's proper schematic symbol.

                   

Brak

[001]

awesome!

[Kleinstein]

A transistor can be used as a zener (usually around 6-8 V), but this is usually a rather noisy zener. It is common to use this if one want's a zener as a noise generator.

With those compensated zeners the linear TC depends on the current and could thus be adjusted, but there is an square law contribution that is usually rather similar among samples. So this would not get better with references in series. The linear TC at a nominal current might get a little better by chance. If the individual zero TC currents are not to different, one could get away with a single current adjusting the overall linear TC.

Long time drift also tends to be similar between units. So there is not much improvement expected.

[001]

Let us create a 10V reference by putting four 2.5V 0.1% references in series. This gives us an accuracy of 0.05% (0.1% divided by the square root of 4)

WHY?

[Echo88]

If you do a worst error calculation you add the percentage errors. But in reality a batch of components will have a statistical uncertainty which will not be the worst error (Bell Curve). Therefore a group of components, from the same batch, which are connected together, will show less error as a whole block, because the slight tolerances will cancel each other in theory. Calculation is done via root sum square of the errors. See here: https://www.dummies.com/education/science/biology/simple-error-propagation-formulas-for-simple-expressions/

In reality mass-connecting cheap metal-film-resistors to produce a really good metrology resistor is just a fools errand, because you need to spend a great amount of time selecting the resistors to cancel each others error. And in the end those resistors will just drift happily in every direction and your work was useless.

"But cant we make this metrology-stuff super-cheap by..." NOPE