Stacking Voltage References To High Voltage Extremes

As children, we all probably had our ideal career paths. As an adult do you still harbor a secret desire to be an astronaut, or to drive a railroad train? Or have holders of other jobs become the people you envy?

As a Hackaday writer it’s probably not too controversial to admit a sneaking envy for the writers of semiconductor application notes. True, often their work consists of dry demonstrations of conventional uses for the products in question, but every once in a while they produce something off the wall and outside the device’s intended use, so out of the ordinary that you envy them their access for experimentation to the resources of a large semiconductor company.

Take Texas Instruments’ Application Report SBAA203, from May 2013. “Stacking the REF50xx for High-Voltage References” (PDF). A laboratory specialising in accurate measurement of high voltages had the problem that the stacks of Zener or avalanche diodes they were using as voltage references lacked both precision and stability, so investigated the properties of the REF5010 10V precision voltage reference.

You'll never be satisfied with a mere Zener diode again.
You’ll never be satisfied with a mere Zener diode again.

They found that by ignoring the device’s data sheet and directly connecting its output pin to its power pin, the REF5010 became equivalent to an ideal Zener diode. In this mode multiple references could be stacked in the same way as a real Zener diode, and very stable and high-precision voltage references could be created with very high voltages. They made a PCB with ten stacked REF5010s for a 100V reference, and then stacked ten of them for a 1000V reference. Leaving it for 24 hours to settle, they achieved a precision of +/- 2.5ppm, and after 3.5 months their average reading for the ten 1000V references they built was 1000.022V.

The 1000V reference would be impressive enough, but they weren’t finished. They built a series of boards holding 500 REF5010s for a 5KV reference, and stacked 20 of them to make a 100KV reference. These boards were mounted in a tower looking not unlike the Tesla coils we sometimes feature here. They note that it probably hits the record of simultaneous use of TI parts in a single device.

This may well be the first extremely high voltage precision reference to feature here at Hackaday, but we’ve certainly had our share of HV articles. Earlier this year we had a trio from [Steven Dufresne]: A conucopia of high voltage sources looking at ways to make your EHT, High voltage please, but don’t forget the current looking at selecting the right HV power supply for an application, and Wrangling high voltage looking at construction techniques.

Thanks [Nathan] for the tip.

23 thoughts on “Stacking Voltage References To High Voltage Extremes

  1. I first thought the requirement for a 100kV precision voltage reference seemed a little odd, but then you think about if a spec for hairyplaney stuff specifies 100kV resistance to be sure of surviving lightning, then it makes sense.

    1. It still doesn’t make much sense. Pretty much every conceivable scenario at those kinds of voltages either don’t need that kind of precision, or specify it as a minimum (like the surviving lightning thing).

  2. Wouldn’t it be simpler to just vacuum tubes to build a high voltage regulator. I recall way back in the day there were some TV sets that had tubes like the 6BK4 or 6EA4 which they used in HV shunt regulators to regulate the HV delivered from the flyback to the second anode of the CRT. Seems like you could use one of those to achieve something like this and probably come out a lot cheaper and more practical.

    Also those tubes have been used by some to build x-ray generators as some varieties will emit some x-rays. usually the sets that used these had a metal cage around them to contain any radiation.

  3. I think voltage stability and drift of a tube regulator is nowhere near as good. The 0A2 regulator tube in my old shortwave receiver is only good withing to within about 1% of 150V.

  4. What bothers me is the “they found that by ignoring the device’s data sheet” part. If you ignore the stuff they specify in the data sheet, don’t be surprised when the precision part of the precision reference goes out the window. You can’t quote the specs and ignore the data sheet.

    1. Data sheets are more of a “Here, let’s get you started with some of the basic functionality of this thing” rather than a law of the universe.

      Plus in general, if you’re going to have a problem with the gratuitous abuse of specs, you’re probably in the wrong place.

  5. For a precision refference I prefer the LM199 (299, 399). Don’t recall if they are stackable though. I see TI dumped them from their line though. LT still makes them but ouch priced. For less demanding specs I like LM10, which contains not only refference diode but a buffer amp as well. As I recall the HP3458A meter they cite as a refference uses 10 LM199s (Hp numbered) as its primary refference. Good for .4 uV per YEAR.

    1. LMx99 are ovenized zeners with an output of 6,95Vnorm, you will need 30% more parts in series. Not to mention the problem of supplying power to each of the heaters. Maybe you can ignore the datasheet and use only the “zener” , put 1000s of them in series, and then put the whole stack in an oven to heat them externally.
      Also, the 3458A uses the LTZ1000ACH, which is a different beast entirely.

Leave a Reply to RWCancel reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.