Stacking voltage references

[Tj138waterboy]

The JJA reference is essentially stacking a lot of low voltage references (Josephson elements) to get something like 1 V or 10 V.

Some of the fluke calibrators start with two  6.5 V references in series.  In the old days of 1N82x zeners it was also common to use effectively 2 in series to get some 13 V.

For a normal instrument there is a limited use to have a stack of many references. This is more like a volt nut thing if the resistors for gain are not good enough.  Ti has an app. note for such a large stack actually used for a high voltage source - not as a commercial instrument, but more like a one off solution at a metrological lab.

And this is simply a one off project just to experiment and learn, cheap and simple as possible. The endgame which may go against ultra precision is to make everything semi-modular with header pins or slots. Can always add in a divider/gain loop/current source etc. This is just my attempt to see what I can do without lm399/ltz1000 but get very good stability and drift but mostly reduntacy with self testing against each other.  Am I wrong for thinking this is a better homegamer way of getting as good as or better than advertised vref specs by stacking and paralleling as opposed to spending a small fortune for a hv dc power supply or calibrator?

[Kleinstein]

  In the old days of 1N82x zeners it was also common to use effectively 2 in series to get some 13 V.

Hello,

The main idea was not getting 13V but to get a wider "zero TC" range by selecting the diodes accordingly.

with best regards

Andreas

The usual old circuit used a bridge like circuit, using one diode and a resistor to set the operating current of the other diode. This way there is no need for an extra amplified signal for the current, like in the usual circuit used for a single 7 V reference.
Using 2 references does not really help with the near zero TC. The linear part depends on the currents and thus the accuracy of the currents. So it is only the normal square root factor for random errors that applies on averaging.  The second order part is very similar (especially same sign) for all the diodes - so averaging does not help here.
It is also a little easier and less sensitive to the resistors to go from 13 V to 10 V than to go from 6.5 V to 10 V (it's about the same starting from 7 or 14). 
However I agree that this is different from the stacking idea here, using many reference.

Stacking a set of reference is a valid alternative to a special higher voltage calibrator. Using many independent lower voltage supplies can replace the higher voltage supply - though they are often the more expensive way.

For a high voltage reference the amplifier way, one does not necessary need an LM399 or LTZ reference. The difficulty, limitation could be more with the resistors to set the gain factor. However with only 1 reference needed one could afford an LM399 or even an LTZ1000 in stead of many ref102 or similar. While the noise improves in the series configuration (so 4 x ref. 102 would have good noise compared to 1 LM399), many reference do not help much with the TC and not significant with long time drift of the reference itself. A drift advantage comes from avoiding the divider / gain stage.

[Tj138waterboy]

Just noticed another trap after building my 10 test modules where as some have higher base voltage that requires more trim which in turn is causing more drift. Found quote in ti thread stating "first, you need to know that trimming the output voltage of REF102 will change the voltage drift by at least 0.008ppm per mV of trimmed voltage."  The worst case I have found in my 2 batches needed trimmed over 500 mV so I will probably have to bin a few refs or buy more to samples closer to optimal value untrimmed.