Power Supply Staging for Multiple REF102 Voltage Reference

[EPAIII]

I would like to construct a multiple output Voltage reference and, after some searching, the REF102 chip which is available at the 0.025% precision point seems to be a good start. It can also be used for a precision current source, but that is another story.

While viewing videos and looking at the TI datasheet for this chip I came across an application drawing that suggests that these chips can be stacked in series to provide multiples of their 10V output (10V, 20V, 30V, etc). This is shown in Figure 11 of the TI datasheet and the glory of this method is that absolutely no other components are necessary, just the wires to interconnect the stack of chips. Wow!

https://www.ti.com/lit/ds/symlink/ref102.pdf?HQS=dis-mous-null-mousermode-dsf-pf-null-wwe&ts=1628574137141&ref_url=https%253A%252F%252Fwww.mouser.com%252F

So I want to try this in order to make a Voltage reference that has ten or twelve Voltages from 10V to 100V or 120V. The problem with this is that each REF102 must power all the other REF102s that are below it in the stack via the Voltage/current from their output pin. So the second REF102 must provide the 1.4mA operating current needed by the first one. And the third must provide that current for both the first and second one or 2.8mA. Actually all of the REF102s in a given stack would have their output current reduced by 1.4mA times one less than the number in that stack. With only a 10mA maximum current available on that output pin, you can reduce the current available to the output terminals to zero with just eight REF102s in the stack. What I need to do is have separate FLOATING power supplies for several groups of three or four of the REF102 chips. Making these power supplies is easy: transformer, bridge rectifier, capacitor, and a linear regulator chip. With the transformer input and no common ground, the supplies will float. But this brings up the question of weather I can just turn those multiple power supplies all on at the same time with a simple power switch or do I need to stage them: one, then after a small delay the second one, then after another small delay the third one, etc.

I can probably stage the power supplies by using larger filter capacitors in each successive power supply. Or perhaps some (solid state) relays to time each, in turn.

But is this necessary? Anyone have any thoughts on this?

The REF102s from a US supplier are in the range of $7 each so a test with perhaps six of them (two groups of three with two power supplies) would risk about $45 in parts. And a stack of ten would have about $75 at risk. So I ask for whatever help I can get.

[PartialDischarge]

https://www.eevblog.com/forum/metrology/stacking-ad587s-for-higher-output-voltage-(20v-30v-etc-)/

[EPAIII]

Thanks for that. I had searched but that thread did not turn up.

It appears that the AD587s do not stack well. Are you trying to say that the REF102s may have similar problems? I know that the TI datasheet only shows three of them stacked and I am thinking about going to the extreme with this idea, perhaps beyond even the wildest dreams of the TI engineer who included that drawing in the datasheet.

A lot was said about noise in that discussion. My initial use for this would be for analog, not digital meters. So I think the noise would not be that important of a consideration and with the slow response of an analog meter, I also wonder if the noise would not just mechanically average out.

I am probably jumping in over my head, but the waters do not seem to be too hot or too cold. And I am a good swimmer.

https://www.eevblog.com/forum/metrology/stacking-ad587s-for-higher-output-voltage-(20v-30v-etc-)/

[dmendesf]

The current is not "consumed". All current that enters the "+" must go somewhere. If you don't load the output then all of it will go to the ground of the ic. Just don't load the outputs and you will be ok. Use a buffer amplifier to sample the outputs.

[magic]

The upper ones source the quiescent current of the lower ones and the lower ones sink the quiescent current of the upper ones. Perhaps those in the middle have the easiest life.

Anyway, why not take one such reference and amplify it? How much accuracy is really needed for testing those analog meters?
For the price of a dozen REF102 you could likely buy quite decent resistors.

[EPAIII]

If you had looked at the suggested circuit diagram in the TI spec. sheet, you would see that the highest REF102 in the stack DOES SUPPLY the V+ for the one under it to operate. This current comes with a plain, wire connection from the output pin of the upper one to the power in pin (V+) of the one below it. Hence, the output current from the upper one is being "consumed" by the lower one. This repeats for the second one supplying the current for the third.


(Apologies to TI over what may be a copyright protected drawing. I have seen it in other places so it has also been copied by others; at least one other on this BB.)

I am well aware of Kirchhoff's current rule as well as possible exceptions. The present issue is certainly not an exception. Use whatever word you wish, but part of the 10 mA output current of that upper IC is no longer available to any external load. It is used (consumed) in the stack of chips. TI even has a note to the effect that this current use must be accounted for in your design(s). And since the second REF102 is also supplying operating current (V+) to the third one, there is another quiescent current being used/consumed. The total quiescent current for a stack of three will be 1.4 mA x 3 = 4.2 mA. And each of the REF102s will have only 10 mA - (2 x 1.4 mA) = 7.2 mA.

As for not loading the outputs, an output with no external load is an output that is NOT IN USE. Any place/circuit which receives the output will (here comes that word again) consume some current. And in a piece of equipment of this nature, where the output can be sent to a large variety of devices, the actual load will be hard to control. All I can do is to LABEL the front panel as to how much current can be drawn before the reference is no longer operation within it's specifications.

Note: The TI spec sheet shows the value of 1.4 mA quiescent current as the guaranteed, maximum value. So I am using that as a design parameter. I am also well aware that actual devices may and probably will consume less quiescent current. But a design can not count on that.

The current is not "consumed". All current that enters the "+" must go somewhere. If you don't load the output then all of it will go to the ground of the ic. Just don't load the outputs and you will be ok. Use a buffer amplifier to sample the outputs.

[EPAIII]

I am not asking for an analysis of my idea. That being said, I have and still am considering other options. This post is about one detail of one of those options, the stacking of REF102s and I still would wish to have that question addressed if anyone can. If not, then so be it. I will probably just try it and then report on my results.

As for how much accuracy is needed, consider two things:

1. A general rule is that the reference should be 10 times more accurate than the device being calibrated/checked. Yes, I know that this rule is often bypassed, but I am shooting for that anyway.

2. I have a number of analog meters and there are many, many more out there. Some of them, some of mine included, are in the 3% to 5% range of accuracy. However, there are analog meters that are a lot more accurate. For example, I have a laboratory style, Weston analog meter with a mirrored scale that is specified to well under 1%. If I were able to check the calibration of this meter with the accuracy that the REF102 provides, then I could use it for checking/adjusting/certifying additional scales on other analog meters and also for rough checks on digital meters.

3. I think it would be a cool circuit and a cool device to have on my bench. Yea, I said two things: so what?

There are other good reasons for having a standard that is as accurate as possible.

Amplify it? Then you need some precision resistors and they also are not cheap. And if you need specific values, they can have long lead times: I am presently waiting on delivery of some 100Ks at 0.01% that I ordered over a month ago. Another problem with amplification is that there is a limit to the Voltages that are possible with most OP amps. I suspect that even if a 120 or 150 Volt capable OP amp exists, it too would be expensive. We aren't talking about a 741 here. REF102s are in stock at several suppliers and do not require any precision resistors. They are truly "plug and play".

Perhaps you can suggest an OP amp and a circuit, including the needed resistors AND a practical scheme for resolving offset errors, that would cost less. In doing this please keep in mind that I do not have a 5, 6, or 7 digit, digital meter to use for setting this up so it must be self calibrating (which the REF102s are).

The upper ones source the quiescent current of the lower ones and the lower ones sink the quiescent current of the upper ones. Perhaps those in the middle have the easiest life.

Anyway, why not take one such reference and amplify it? How much accuracy is really needed for testing those analog meters?
For the price of a dozen REF102 you could likely buy quite decent resistors.

[BradC]

https://www.ti.com/lit/an/sbaa203/sbaa203.pdf

[EPAIII]

Thanks Brad.

At first glance that looks like it may be even better than the REF102. The ability to stack them appears to be nothing short of phenomenal. And only one power supply seems to be needed. And they come in other Voltages like 2.5 and 5. Interesting possibilities.

And if I use them with one power supply, my original question becomes moot.

So, four REF5025s gives me four steps of 2.5V and another 21 REF5010s get me up to 220 V. I can use a step up transformer to get up to a 230 or 240 V supply and to provide isolation from the line.

The only down side is they are only available to 0.05% while the REF102s go down to 0.025%. But I can probably live with that. And that TI Application Report talks about the error level getting better when multiples are stacked due to the random nature of the actual errors. So the precision increases when you go higher on the stack. Perhaps I could buy more of the 2.5V version in order to provide a better random distribution at the bottom end of the stock, perhaps between 0V and 50V. But then the price increases too.

Food for thought and more research.

Thanks!

https://www.ti.com/lit/an/sbaa203/sbaa203.pdf

[magic]

https://www.ti.com/lit/an/sbaa203/sbaa203.pdf

That's a fun project

I am not asking for an analysis of my idea. That being said, I have and still am considering other options. This post is about one detail of one of those options, the stacking of REF102s and I still would wish to have that question addressed if anyone can. If not, then so be it. I will probably just try it and then report on my results.

Well, okay. Speaking of stacking tens of REF102, I think it may work out in terms of quiescent currents and power dissipation. With three references, you can see that the third one runs on the same quiescent current as the top one and the second one is essentially doing nothing. It seems that the fifth one would also run on a current recycled from the third one, and the fourth one would run on current from the second one. So only the first and the last would need to source/sink any current to others and you could help them with emitter followers or balancing active loads. Draw a full schematic and figure out what happens, including when some output drives a load; that may be cheaper than building a prototype

Note that somebody in the old thread reported trying something of that sort with AD581 and seeing oscillation. So...

Speaking of amplification and resistors, I admit I don't know exactly how much precision you can buy for the price of REF102. Something to possibly consider is to make the chain a power of two (like 16 resistors for 160V) and verify INL by breaking it in halves and comparing them with a Wheatstone bridge. Total overall gain could be verified with a Hamon divider. But that's work Opamps can be "boosted" for higher output range, lot's of schematics out there, but that's admittedly also work.

[EPAIII]

I have been reading the materials on the REF50xx series and it is becoming more clear that it is a better choice. The TI article that Brad posted talks about not three or a few of them in series but numbers like 1000 and even 10,000. Hard to imagine, but they have photos of the actual devices. And yes, that means a 100 kV reference that is accurate to 0.05%. No misprint, 100 kV. How's them apples? I was bit by 50 kV once, but 100 kV is certainly out of my league in a home shop.

And only one power supply would be needed which for me would be about 150 to 250 VDC. The trick they use in stacking such large numbers of devices is to turn them effectively into two terminal devices and jacking the quiescent current up to 3mA instead of the spec sheet value of 0.8mA. Apparently that happens as a result of connecting the output terminal of the device to the V+ terminal, which contrary to the spec sheet's recommendation, places those two terminals at the same Voltage. They claim this turns it into an ideal Zener diode. A single current limiting element is used in series with the stack to limit the current to that value. Perhaps it would run away without that; they don't say. But that is what a Zener would do if connected directly to a Voltage that is higher than it's Voltage rating: it would draw current until something went pop.

I also did some pricing and it appears that I can get an assortment of them for around $3.00 each. That is a lot better than the prices I got on the REF102s: about half. I will need two 1uF capacitors with each one, but I have a bunch.

Thanks for the help so far and keep it coming. Perhaps there is a better way still. I am getting excited about this. Wild ideas do work.

[Kleinstein]

Even with divices that need a higher supply voltage than the reference output one can stack them in a way that the supply current don't add up. The chain would only see 2 times the supply current. For the supply there are 2 chains and the chips get 2 times the reference as there supply.

Stability against oscillation can be an issue with such a construction. It should be OK with just zener like shunt references.
With the sereis connection with quite some capacitance, I would consider added protection diodes to avoid possible reverse voltages when turning off with a load present.

[magic]

I will need two 1uF capacitors with each one, but I have a bunch.

Note that noise reduction capacitors need to be low leakage types - do the math
And that 1µF is unlikely to make a difference within the bandwidth of most analog meters.

[dmendesf]

I'm sorry, I didn't look at the schematic. Instead I had the REF50xx stacking in my mind. I've seen the 100kV ref before... It's only possible because of the "series of zeners" architecture.

If you had looked at the suggested circuit diagram in the TI spec. sheet, you would see that the highest REF102 in the stack DOES SUPPLY the V+ for the one under it to operate. This current comes with a plain, wire connection from the output pin of the upper one to the power in pin (V+) of the one below it. Hence, the output current from the upper one is being "consumed" by the lower one. This repeats for the second one supplying the current for the third.

[ Attachment Invalid Or Does Not Exist ]
(Apologies to TI over what may be a copyright protected drawing. I have seen it in other places so it has also been copied by others; at least one other on this BB.)

I am well aware of Kirchhoff's current rule as well as possible exceptions. The present issue is certainly not an exception. Use whatever word you wish, but part of the 10 mA output current of that upper IC is no longer available to any external load. It is used (consumed) in the stack of chips. TI even has a note to the effect that this current use must be accounted for in your design(s). And since the second REF102 is also supplying operating current (V+) to the third one, there is another quiescent current being used/consumed. The total quiescent current for a stack of three will be 1.4 mA x 3 = 4.2 mA. And each of the REF102s will have only 10 mA - (2 x 1.4 mA) = 7.2 mA.

As for not loading the outputs, an output with no external load is an output that is NOT IN USE. Any place/circuit which receives the output will (here comes that word again) consume some current. And in a piece of equipment of this nature, where the output can be sent to a large variety of devices, the actual load will be hard to control. All I can do is to LABEL the front panel as to how much current can be drawn before the reference is no longer operation within it's specifications.

Note: The TI spec sheet shows the value of 1.4 mA quiescent current as the guaranteed, maximum value. So I am using that as a design parameter. I am also well aware that actual devices may and probably will consume less quiescent current. But a design can not count on that.

The current is not "consumed". All current that enters the "+" must go somewhere. If you don't load the output then all of it will go to the ground of the ic. Just don't load the outputs and you will be ok. Use a buffer amplifier to sample the outputs.

[EPAIII]

Yes, they have different stacking arrangements.

But I am seriously looking at the REF50xx chips now. They do stack with no current loss. Apparently all of the chips in that series have the same stacking ability so I can toss in four of the REF5025s which are 2.5 Volts. I need to verify that, but it seems to be the case. So along with a stack of about 20 or 25 REF5010s to get to about 200V or 250V, I can have a base of four REF2025s which will provide 2.5V increments. I am envisioning a number of terminals connected to the 2.5V and 10V points in the stack. By choosing the terminals to use, that will provide steps of 2.5V from zero up to 200V or 250V.

And they are about half the price of the REF102s.

Apparently the only down side is the REF50xxs only come in 0.1% and 0.05% versions while the REF102s can be had at 0.025% precision. I think I can live with that.

Now my power supply problems seem to center around finding a transformer that has a secondary Voltage around 200 Volts. And do I need to have a regulator on it or can I just rely on the REF50xx's supply rejection specs. I am leaning toward using a regulator. Perhaps I can implement it on the low side of the PS.

I'm sorry, I didn't look at the schematic. Instead I had the REF50xx stacking in my mind. I've seen the 100kV ref before... It's only possible because of the "series of zeners" architecture.

[Kleinstein]

The Ref5025 will likely not stack well in the simple configuration. The REF5020 sopposedly needs a minimum of 2.7 V for the supply and this should also apply to the 2.5 V version. So 4x2.5 V are better done with the Ref102 and a divider. REF5025 could still be stacked with a different supply.

Die Supply for the whole chain should be regulated, to get a reasonable stable current. One may get away with a reasonable good constant current source, ideally in a way to also withstand a short.

As the refference chain only needs a low current (e.g. 2 mA) one can us a diodes / capacitor voltage muliplier. So one could get away with a more normal 2x24 V or similar transformer.

The main idea with the stacked reference is that with a good DMM one can measure the individual steps and than add the voltages. So the initial accuracy is not that important. Just for the 0.1 % range it is much easier to use resistors.

[magic]

The difference between REF102 and REF50 is that the former is a buried zener reference while the latter is bandgap-based. The former in principle has better long term stability. While both should meet their declared accuracy specification, the 10.02V you get out of REF102 may stay that way for a long time instead of becoming 10.019V (made up numbers, consult the datasheet or search the metrology section for real-world performance data).

Kleinstein, is it known if bandgap references happen to have anything resembling a normal distribution of output error? Or at least symmetric around zero?

I wouldn't expect averaging improvements when it comes to buried zeners. I have a lot of five LT1021B-5 and all of them are on the low side and three are almost identical. Granted, they are the B grade, trimmed for minimum tempco rather than minimum offset. But I think there are good reasons for that: if a process variation in given batch causes the zeners to be off then all of them will be off by similar margin. Trimming will also need to be done in the same way to most of them and will yield the same outcome, particularly if it's "discrete" trimming with fuses as in LT1021.

Bandgap reference OTOH rely on transistor matching and their largest error source is probably variations between individual transistors made on one wafer (which are amplified some 10x IIRC). Such errors could be more "properly" random, if you are lucky, and result in improvement when devices are stacked or paralleled.

[Kleinstein]

Averaging can help with both bandgap and burried zener ones. However that can be the effect that 20 from the same batch can be quite similar. So most may show a rather similar TC and a rather similar errror in the absolute voltage and aging. Chances are they came from the same batch and same machine to do the the final trim.  For the absolute error there can be be just discrete steps in trimming - here averaging can help than. The other possiblity would be an error cause by residual drift (trim before fully settled) - this error can be biased and similar for a batch. The higher grades may be sected after trim and the lower grade may be missing the good ones. The alternative can be a more accurate / slower trim process. The distribution can differ for different parts and still variations over date (e.g. some batches  may be treated / selcted differently, e.g. no good ones selected, need to through out more bad ones because of an more off center start).

I don't think there is a principle difference between zener and bandgap. There is a tendency that bandgap refs are more modern designs and therefore preference other trim methods (e.g. digital PROM based instead of old style fuses). The discrete steps may still be small enough to vary inside a batch, at least for tight tirmmed parts.

For the cheap versions the plastic case is likely the main weakness. It reacts to humidity and this causes stress to the chip. How good the chip handles stress depends on the design (may be still just luck / try and error). I don't see a real difference between Zener and bandgap in this respect. With the bandgap there are a few more ways to vary and maybe get a lucky low sensitivity version. For the noise I would expect more RTN/popcorn type noise with a zener and maybe more random walk like LF noise with a Bandgap. It still depends on the process.

[EPAIII]

Re: the 5025 not stacking well.

I was and still am somewhat concerned about this point. The TI Application Report, Stacking the REF50xx for High-Voltage References which BradC made me aware of says that something that is unexpected happens when the 5010s are configured as two terminal devices AND DRIVEN WITH A HIGHER THAN NORMAL CURRENT LEVEL (3-8 mA as opposed to the spec sheet's value of 800 uA, typical). This configuration as a two terminal device includes connecting the Output pin to the V+ pin, which insures that those two pins are at the same Voltage. The whole point of that application report is that this configuration and current level insures that the device effectively becomes a two terminal, ideal Zener which operates at the rated Voltage of 10 Volts and also at the rated accuracy (0.1% or 0.05%).

The authors of that application report further state that they verified that other members of the REF50xx series also exhibit that behavior. I am not 100% certain that this statement applies to the REF5020 because the TI spec sheet states it requires a higher Voltage margin, but they seem to be saying even that as well as including all the other members of the REF50xx family.

The TI spec sheet states that all members of the REF50xx series need to operate with a supply Voltage (V+) which is at least 200 mV higher than their rated output Voltage except for the REF5020 which requires the same 2.7 V for V+ as the REF5025.

I therefore suspect to be able to confirm that the REF5025s will also exhibit that behavior when configured in that manner and when operated with 3 to 8 mA of current at their ground terminal. I presently intend to confirm this as a first step when I have the needed parts at hand. Then I will add one or two 5010s to that stack and again check the results. I do not expect that a different supply will be needed, only a bench supply that can deliver 12 to 32 Volts at that stage. One step at a time and I will post the results.

If the above checks out, then I will construct a supply that can deliver about 265 to 290 Volts and try a final stack size which will allow Voltages of up to 260 to be produced.

If I am attempting to have a reference chain that goes up to 260 V, I do not see how any reasonable amount of Voltage doubling with a 24V or 2 x 24V = 48V transformer would be of any help. And the application report plainly calls for 3 to 8 mA, not just 2 mA. But I have not used Voltage doubling in any prior power supply designs so I am going to look at it. Wouldn't multiplying the Voltage by six also mean that the transformer(s) also need to supply at least six times that needed 3 mA or 18 mA. And that probably means 20 or 25 mA to be on the safe side.

As for your last paragraph, "The main idea with the stacked reference is that with a good DMM one can measure the individual steps and than add the voltages. So the initial accuracy is not that important. Just for the 0.1 % range it is much easier to use resistors." I am not sure what you are saying.

I am hoping that, by using the 0.05% REF50xx parts, I can achieve individual steps at that, 0.05% accuracy level without any need for verifying with a more accurate DMM or other device. Hence, for me the initial accuracy IS that important. I am hoping to construct a Voltage reference that is more accurate than any meter that I presently own.

And I would like to be able to tap off individual Voltages from that stack at 2.5 V intervals over the entire range from 2.5 V to 260 V. Of course, only one such Voltage would be in use at any given time. Perhaps this is overly ambitious, but I will see.

I am in the process of determining how to economically construct a DC supply between 265 and 300 Volts at present. This seems to be mainly a question of the transformer(s) used. I do not think that a switching supply is a good idea as it would probably introduce the switching noise as an undesired factor. I AM going to look into your Voltage doubling or multiplying suggestion as that could possibly solve the transformer problem. I may even have one that would work.

In any case, thanks for your answer.

The Ref5025 will likely not stack well in the simple configuration. The REF5020 sopposedly needs a minimum of 2.7 V for the supply and this should also apply to the 2.5 V version. So 4x2.5 V are better done with the Ref102 and a divider. REF5025 could still be stacked with a different supply.

Die Supply for the whole chain should be regulated, to get a reasonable stable current. One may get away with a reasonable good constant current source, ideally in a way to also withstand a short.

As the refference chain only needs a low current (e.g. 2 mA) one can us a diodes / capacitor voltage muliplier. So one could get away with a more normal 2x24 V or similar transformer.

The main idea with the stacked reference is that with a good DMM one can measure the individual steps and than add the voltages. So the initial accuracy is not that important. Just for the 0.1 % range it is much easier to use resistors.

[EPAIII]

The authors of the TI Application Report in describing the behavior of 100 REF5010s used to construct 1000 Volt References seem to confirm the random behavior of the errors in the individual devices. They report that the trimming adjustment they included in these 1000 V references was not really needed as the average Voltage of ten such 1000 V sources was 1000.0022 V after 3.5 months. That is over three orders of magnitude better than the 0.05% spec. of the individual 5010 devices.

That seems to be a very strong confirmation of the individual errors being very much random.

I do not know of any such numbers for the REF102s.

The difference between REF102 and REF50 is that the former is a buried zener reference while the latter is bandgap-based. The former in principle has better long term stability. While both should meet their declared accuracy specification, the 10.02V you get out of REF102 may stay that way for a long time instead of becoming 10.019V (made up numbers, consult the datasheet or search the metrology section for real-world performance data).

Kleinstein, is it known if bandgap references happen to have anything resembling a normal distribution of output error? Or at least symmetric around zero?

I wouldn't expect averaging improvements when it comes to buried zeners. I have a lot of five LT1021B-5 and all of them are on the low side and three are almost identical. Granted, they are the B grade, trimmed for minimum tempco rather than minimum offset. But I think there are good reasons for that: if a process variation in given batch causes the zeners to be off then all of them will be off by similar margin. Trimming will also need to be done in the same way to most of them and will yield the same outcome, particularly if it's "discrete" trimming with fuses as in LT1021.

Bandgap reference OTOH rely on transistor matching and their largest error source is probably variations between individual transistors made on one wafer (which are amplified some 10x IIRC). Such errors could be more "properly" random, if you are lucky, and result in improvement when devices are stacked or paralleled.

[Kleinstein]

The  the ref5020 likely needs the higher supply to operate the internal circuit (e.g. OP) for the reference itself.
The 200 mV headroom suggested by TI for the higher voltage seems to be needed only when sourcing a current. When sinking a current the higher voltage one (5 V and 10 V) seem to be OK without a higher supply. The 2.5 V version may be boarderline - I would not bet they would all work in shunt mode - some may but some may also fail and there is a chance the accuracy could suffer.

For getting fine steps over a large range, one could have three  2.5 V steps at the bottom and than follow on with 10 V steps. The choice is than from the lower and upper terminal. Alternatively one could have a precision divider from the lower 10 V.


Voltage muliplying rectifiers of cause need more current from the transfromer. It is just often easier to get a transformer for 24 V than a 230/230 V transformer. The extra effort for the voltage multiplier at low power is relatively moderate (diodes and capacitors). The alternative is to use 2 transformers in series: e.g. as 230V -> 12 V and 12 V -> 230 V.

Some 3 mA at 250 V are only 750 mW, so it does not need really much power. I would still avoid transformers smaller than 3 VA if space is not at a premium. The very small transformers behave a bit odd and have increddibly poor efficiency. The no load loss is usually lowest for a 3-5 VA power rating.

[EPAIII]

Kleinstein,

Yes, I am a bit nervous on the 2.5 V ones working 100%. That is why I want to test it. I will probably buy 8 or 10 of them and breadboard it. I don't have the best meter in the world, but I think I may be able to detect any small differences. One method that may work would be to build two strings of them and check for differences with the most sensitive meter scales.

As for three 2.5 V steps at the bottom, I thought it was obvious that was what I was thinking of. But actually I was thinking of four instead of the minimum of three. And then 25 steps of 10 V. Sorry if I was not clear.

On the transformer, a 230 V is not very useful. My standard, linear power supply calculation goes something like this:

RMS (230 V) x 1.414 = P-P (325 V)

P-P - 1.5 V (rectifier loss) = 323.5 V (324 V if Shockley)

323.5 - allowable ripple (1.5 V?) = 322 V

And that is a lot higher than 265 V or 270 V which would be a good operating Voltage for a 260 V reference stack. I would have to drop 52 Volts in manner. Even if the 3 mA current regulation would drop 5 or 10 V, that still leaves a lot.

I was looking for a 195 to 210 Volt RMS transformer. And, compared to that, a 230 V is dead easy to find. One thing I am looking at is four, 48 V transformers in series for 192 V or thereabouts. That's almost marginal, but with some luck it can work. Many transformers have higher Voltages when the current is low compared with it's rated value. Since this is a low power device to begin with and it will not be used often, I am not concerned with efficiency. If you are worried about being "green", you must realize that you can easily waste 10 or 100 times as much energy or resources trying to build something with the best operational efficiency while expending so much more in the cost of the materials. I have a practical slant on how to judge what is "greener". If it costs less in currency, over all it's life, then it is "greener".

Anyway, this is why I am going to look into your Voltage multiplier idea. I have seen multipliers that had over 10 steps, perhaps as many as 20.

The current in a stack of references like this seems to me to be divided into two components. First there is the operating or quiescent current which will be 3 mA in this case. That same 3 mA will flow through all of the individual steps. Then there will be the load current. That current will flow through the load so it will be + on one connection and - on the other. Both of these points will be at one step or another of the stack, including the lowest and the highest. This load current will flow through all of the REF50xx devices between those two connection points. It will not effect the amount of current drawn from the power supply.

I think that the quiescent current or at least 2.8 mA of it will need to be subtracted from the 10 mA maximum current that the devices can handle, leaving only 7 mA available for the external load. But this should be enough for most uses. I am not certain of this and it is another item that I need to validate after constructing the device.

The  the ref5020 likely needs the higher supply to operate the internal circuit (e.g. OP) for the reference itself.
The 200 mV headroom suggested by TI for the higher voltage seems to be needed only when sourcing a current. When sinking a current the higher voltage one (5 V and 10 V) seem to be OK without a higher supply. The 2.5 V version may be boarderline - I would not bet they would all work in shunt mode - some may but some may also fail and there is a chance the accuracy could suffer.

For getting fine steps over a large range, one could have three  2.5 V steps at the bottom and than follow on with 10 V steps. The choice is than from the lower and upper terminal. Alternatively one could have a precision divider from the lower 10 V.


Voltage muliplying rectifiers of cause need more current from the transfromer. It is just often easier to get a transformer for 24 V than a 230/230 V transformer. The extra effort for the voltage multiplier at low power is relatively moderate (diodes and capacitors). The alternative is to use 2 transformers in series: e.g. as 230V -> 12 V and 12 V -> 230 V.

Some 3 mA at 250 V are only 750 mW, so it does not need really much power. I would still avoid transformers smaller than 3 VA if space is not at a premium. The very small transformers behave a bit odd and have increddibly poor efficiency. The no load loss is usually lowest for a 3-5 VA power rating.

[magic]

I think it would be smarter to compare four 2.5V in series against one 10V because two identical stacks of 2.5V may produce the same error and hence no difference between them.

The 10V may even run in normal mode from the recommended supply voltage (12V, 15V, whatever it is) for test purposes.

You will want to test dynamic resistance too - whether the voltage drop changes significantly with bias current.

[Terry Bites]

https://e2e.ti.com/blogs_/archives/b/precisionhub/posts/understanding-voltage-references-part-3-how-to-achieve-shunt-reference-flexibility-with-series-reference-precision

Its desirable to have all referrences operating at the same current to ensure even dissipation. It may be more ecconomical .
Daqq posted this previously.... http://www.ti.com/lit/an/sbaa203/sbaa203.pdf

[Kleinstein]

A slightly higher voltage is not that bad.  50 V drop at 3 mA would be only 150 mW. That is easy with a resistor (preferred before the filter cap to get extra filtering and improved power factor for free) . The ripply would normally be more than 1.5 V (I would plan with some 5% so maybe 10 to 15 V and avoid an overly large capacitor). So a 230 V transformer would be not so bad.

Keeping the power loss low helps with precision. It is not so much about the energy costs or consumption, but extra heat one usually does not want.

Using more current from the chain could cause quite some possible trouble. Still plan for it in case of a fault, the normal current would still be limited to the set string current. So a constant current source at the supply side would also provide some protection. For this the current source should be able to also withstand the full voltage (e.g. short of the chain). The normal current drawn by a meter is more like some 50 µA or 100µA for an analog meter.
With a constant current source to withstant some 300 V it does not matter to have some 20 V more that are dropping under normal use.
The constant current source would replace a regulated supply - the references regulated there supply if they get the constant current.

The 2.5 V at the bottom could still be a slight problem. I would consider just a divider (chip with 4 x 10 K) and than 3 buffer amplifiers for the outputs. Alternatively the lower 2.5 V regulators could get a regular supply ( e.g. from a capacitive droper at the transformer).

Testing the 2.5 V regulator at 2.5 V supply would be quite some effort. It may only fail with higher temperature or worst case only after aging.