Low Cost Voltage Reference Experiment

[paulie]

By "low cost" I mean pennies not dollars. "Voltage reference" refers to any device that might function as a transfer standard for calibrating 5 or 6 digit voltmeters or serve as reference for DIY meter or PSU with similar resolution. This was inspired by hours spent browsing volt-nut megathreads and excitement over my new 5 digit $5 Ebay meters. So the plan is to evaluate likely candidates including inexpensive bandgap references like LM4040a or TL431, DIY zener type built from back-to-back penny diodes, and maybe even properly biased LEDs and transistor base/emitter junctions.

First step was to check shunt voltages to see what might be within useful range. The first piece of test equipment used here was my home grown combo super low noise PSU/5-digit voltmeter with current controlled test points (see 1st photo, DUT=NPN). Immediately devices divided into two major categories: 2.5v for the bandgap or LED and 7v for the transistors and zeners.

Next step was to get a rough idea of TC which required a mid size temperature control oven like shown in next 2 photos (my Worlds Smallest was way too small and the big one takes too long). Commercial bandgap parts were very well behaved out of the box and after some optimization the DIY zeners were pretty stable too. Haven't checked TC for the LEDs or transistors yet.

Final step is to characterize long term drift so dragged out one of the big ovens which were used couple years ago for a customer to test batteries. I had over 150 of these built for them and ended up with four units for myself. With a slight modification perfect to keep voltage references at a constant 25C (see next few pics). Also re-purposed was the 16 channel ADC multiplexer/datalogger shown below. Only mod there was to replace the ICL7135 ADC chip with one of the 5 digit $5 Ebay meters. An on-board 10x low noise amp provides for effective resolution of 6 digits (0.01mv). More than enough to track short and long term drift.

In fact my first set of off-the-shelf bandgap references finished yesterday after cooking for 5 weeks. Conclusion is most of them will work fine as transfer calibrators for some friends 4 and 5 digit meters. And that was with junky generic 0805 SMD resistors which take some time to stabilize due to self heating. Next batch will use 10ppm Vishay resistors (see photo in  the bandggap post below) which I also inherited from the day job. They actually test out at least 2-3ppm over 10deg range so performance should improve significantly.

More to come and as usual comments and questions welcome.

update: Someone asked for more info on the expander/amp seen in "meter_in_oven" photo. Unfortunately I don't have a schematic that has been updated since modifying for this project but attached is a pictorial diagram. I think the functions and labels haven't changed much so mostly still valid. I have a block diagram somewhere too but can't seem to locate it ATM. If anybody is really interested in building one it might be possible to dig up the channel switching and datalog code. In any case this diagram should serve for idle curiosity.

[paulie]

Low cost bandgap devices. Most 20 cents or less each, 2 cents for the TL431.

[paulie]

Back to back penny zeners. Less favorable voltages than the 2.5v bandgaps but potentially more stable. Not as good as "buried" type but we will see.

[paulie]

LEDs and bipolar transistors. Not as promising but certainly inexpensive.

[JohnnyBerg]

Why not try LM385? Very low cost, and much better then zener or TL431(A)

[paulie]

Yes, somewhat better than TL431 but not sure about zener. At about 20-30 cents they do qualify so I bought some from Polida2008. Lower cost than 4040a or LM336 but still 10x more than TL431. I had large qty of the others on hand so used them in the initial test. Now I'm planning to use zeners for the next run and just waiting for some buried type to arrive to start that. Because my 10x resolution expander has restricted range won't be able to include any of the 2.5v parts this time around but definitely next run after.

[TimFox]

If you are making your own temperature-compensated  Zeners with the series connection of a reverse-biased Zener (usually 6.8 V) and a forward-biased PN diode (which could be a second Zener), you can find a current for each set at which the tempcos cancel for a zero net tempco.  With your ovens, you can put several such sets in the oven, stabilize the units at one of a discrete set of temperatures (the wider spaced the better), and measure the voltage-vs-current using your 5-digit meters.  If your current values are repeatable, you can compare the voltages directly at different temperatures, or you will need to do some simple curve fitting. 
Details on compensated Zeners can be found in the old Motorola Zener diode handbook, which shows the combinations of Zener with one, two, or three diodes that gave the standard voltages of 7.5, 8.2, and 9.1 V nominal.

[JohnnyBerg]

Yes, somewhat better than TL431 but not sure about zener.

I am
Unless you got a selected zener of about 6.2 V, the LM385 (and also the TL431 and LM4040) have a way, way better tempco.

At about 20-30 cents they do qualify so I bought some from Polida2008. Lower cost than 4040a or LM336 but still 10x more than TL431. I had large qty of the others on hand so used them in the initial test. Now I'm planning to use zeners for the next run and just waiting for some buried type to arrive to start that. Because my 10x resolution expander has restricted range won't be able to include any of the 2.5v parts this time around but definitely next run after.

I'd say ditch the zener, unless you can build it in a good oven, and a really good current source (with good tempco) to maintain a constant current through the zener. Also, current needs to be significant higher as the LM385 and LM4040. But also higher then the TL431.

Probably the noise of the zener is better. Noise ( <10hz) for the LM385 is about 15µV pp and about 11µV pp for the TL431.
But I don't think you're aiming for a 10µV resolution

[paulie]

There's no doubt even low end bandgaps perform better off-the-shelf than uncalibrated zeners in terms of both accuracy and stability. However I found that the diodes can be properly adjusted for almost zero TC. At least over the 10deg or so range I've tried. And that's without being directly concerned with complex or expensive current sources but just proper resistance with a simple voltage regulator which is much easier. Not possible with the bandgaps that are designed for stability over wide range of currents.

The biggest issue so far is the extended warmup time due to my low mass (aka low cost) 10ppm SMD resistors. Also experienced with the bandgap references, just not so bad.

But you might be right about wasting time with these glass penny diodes. Discussions in the volt-nut thread and technical papers indicate lots of noise and poor stability compared to buried. However note that before the advent of reference ICs such as LM399 and LTZ1000 companies like Solartron used surface devices, albeit selected, in high end 7 digit or more meters so there is hope. I'm just curious how mush variation there is between the penny zeners and how they compare with buried type. There is almost a thousand to one difference in cost.

[JohnnyBerg]

@Paulie: and how do you solve the constant current?

[paulie]

Constant current is not required directly. A simple resistor with fixed voltage takes care of that. So maybe more correctly referred to as "fixed current" since a constant current circuit is not really needed. This is a little troublesome with the low voltage zeners which required very low ohm resistors but I've pretty much given up on 2.5v zeners for other reasons.

ps. Self heating of the resistor is an issue but handled with some warmup time.

[tszaboo]

Constant current is not required directly. A simple resistor with fixed voltage takes care of that. So maybe more correctly referred to as "fixed current" since a constant current circuit is not really needed. This is a little troublesome with the low voltage zeners which required very low ohm resistors but I've pretty much given up on 2.5v zeners for other reasons.

ps. Self heating of the resistor is an issue but handled with some warmup time.

As far as I see you need more than 5mA for anything below a 5,6V zener, which is really big for this. I would try this at 5mA.
http://www.digikey.com/product-detail/en/BZX84-A5V6,215/568-6787-1-ND/2676827
Datasheet said -2 .. +2,5mV/K tempco for 5mA. Might even work with some tweaking. You can decrease the resistor self heating problem by just using a bigger resistor, which doesnt increase the price that much, or a more stable resistor, which does. Also, there is this old appnote:
https://www.google.be/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0CCAQFjAA&url=http%3A%2F%2Fwww.microsemi.com%2Fdocument-portal%2Fdoc_download%2F14616-zero-tc-reference-diodes&ei=zvIKVcrGL8f2O_CdgbgN&usg=AFQjCNEj2okI2n5PGSzVBpoekY-17RsXpg&bvm=bv.88528373,d.ZWU

[JohnnyBerg]

Constant current is not required directly. A simple resistor with fixed voltage takes care of that.

Whow ... hey .. wait a minute! Fixed voltage? Why go in all the hassle with a zener and not use the fixed voltage?

note1:  a XC6206 is 1% accurate, has a way better tempco then a zener, can deliver more output current and probably costs less.

note2: voltage references are often operated from a battery ..

[Conrad Hoffman]

I've had excellent results using the old circuit (Walt Jung's Opamp Cookbook, I think) where the zener is buffered by an opamp, and the stabilized output of the opamp supplies the zener current through a resistor. With a temp compensated zener, you can tweak both the tempco and the output voltage, though they do interact. Whatever you use as the reference, that scheme will usually make it better.

[albert22]

Some time ago I found this:

I have an older voltage reference, which was based on Design Idea article in EDN Magazine back in the 1970's -its an LM723 voltage regulator chip connected so that the internal pass transistor acts as a heater, and the internal current limit transistor serves as a temperature sensor, so the chip could provide a stable temperature for the on-chip zener reference voltage.

Although I found no schematics or any other references. It seems worth to try. LM723 is cheap and you have the oven as a bonus.

[SeanB]

I used one as a reference voltage to make an offset voltage meter, just using the internal reference to make a simple 7V voltage source. It is more accurate than the analogue meter attached to it could ever be. It runs nicely off a 33V rail derived from a 60VAC tap on the variac.

[Andreas]

Hello,

sorry that I do not take the thread as serious
(otherwise the TO would simply use a LM385 as all other).

So back to the roots:
You can build a zener and a diode (for TC compensation)
with just one double transistor.
The BCV62 would be a perfect candidate with 2 matched transistors in one case.
Simply use one transistor as zener (BE junction reverse biased gives something 5-7V)
and the other as compensation diode. (collector + base connected).
You only have to select the zener for around 5.6V to have minimum T.C.
(Unfortunately I got around 8-9V for the whole cirquit when I tried this.

With best regards

Andreas

[paulie]

Yes, as mentioned in the first post and photo I've constructed several BJT base/emitter references but except for basic voltage check have yet to evaluate in depth. Certainly no long term drift tests yet. So I have a couple questions for you, Conrad, and the others with much more experience than me who have taken time to peek in here:

1. How do these transistor junctions compare to buried zeners, Both in form and function. Would they be considered buried or surface?

2. Same question about the LM723. I have a box full harvested from power supplies over the years.  Chip schematics show the 6.2v component but nowhere can I find a description of the construction. Is it possibly buried type too?

I ask because the cost (main point of this thread) is so low it would be quite a bonus if either of these actually were. If such luck then would absolutely be the real key to a "poor man's" precision reference. After many hours of searching I see lots of info on the net for high end references but pretty sparse for the penny parts. Hopefully somebody out there might have a hint.

[Andreas]

Hello,

perhaps you should read a good book as overview:
https://www.elsevier.com/books/current-sources-and-voltage-references/harrison/978-0-7506-7752-3

but caution there are some mistakes in the book e.g. claiming the MAX6325 as precision band gap.
(but this bug is/was also on the Maxim web page while the datasheet notes correctly a buried zener).

All standard junctions are no buried zeners.
Buried zeners need a ion implant process which is a cost adder.
So your transistor is a surface zener.
Also the original uA723 is a surface (temperature compensated) zener (according to TI datasheet).
But be aware that there are "LM723" clones around with band gap references.

Also the venerable 1N821-1N829 are surface zeners. So you will have to select for low noise.
They were pennie devices in the past. But nowadays they are more several dollars.
And I personally think that the devices nowadays are more noisy than they were in the past.

The cheapest buried zener might be the LM329. (= LM399 without the heater).

With  best regards

Andreas

[paulie]

Thank you for that information. It will save a lot of effort concentrating on the wrong technology. Space (and time) are very limited as I'm only able to keep one oven operational at a time.

IIRC 1N829 are actually parts I referred to earlier used by Solartron in their high end meters so looks like there's hope for non-buried zeners.

LM329 are definitely out of the price range as a low cost reference but janaf has graciously offered high end zener samples so I may get a chance to compare those with the low cost glass types too. I did purchase an LM339 because it was just under my self imposed budget limit and that has arrived so will be part of the testing. I can't decide right now whether to use it with heater as a high quality voltage standard to compare or just test the zener part along with the others.

ps. Interestingly LM399 does not really have a simple zener available but instead part of a rather complex internal circuit as seen in the attached diagram. Looks like the "zener" in LTZ1000 is not available as a real diode either.

[sync]

LM329 are definitely out of the price range as a low cost reference

60 cents are too expensive???

[paulie]

I'm unable to find those for less than $2. Do you have a link? It should restrict to small quantity and include shipping because unfortunately I do have to pay that.

A little explanation of my budget constraints, there are two: cost of parts as described in first sentence page one (pennies) and hobby expenditure of any one piece of equipment or bulk purchase ($5). That last is partly a self imposed limit from deciding not to pay more for a support item than the cost of the original 5 digit Ebay meter that started all this. Also some restrictions imposed by family obligations (aka WIFEY!).

So for example if parts were available for 60 cents ea but only in 10 pc qty it would be just over the limit. But 300 pcs at one cent ea like with the zeners or 200 pcs TL431 for $2 no problemo. May be hard to believe but total parts cost for the PSU/voltmeter tester, the ovens, and the 10x/MUX logger were all under $5.

[electrodacus]

I also need a voltage reference in my application and my option at the moment is TS6001b it is about 1$ maybe less in quantity but probably better than any alternative proposed here.
Do you know anything better around 1$ ?

[JohnnyBerg]

Do you know anything better around 1$ ?

More equal, I think: LT1004 (mouser 0,71 euro)

Better is more money

Why do you want to change, what I see spec is not bad? Brand?

[electrodacus]

More equal, I think: LT1004 (mouser 0,71 euro)

Better is more money

Why do you want to change, what I see spec is not bad? Brand?

Thanks I will look at that one. This is a new project currently on Kickstarter Solar BMS and I did not had time to look in to voltage reference I just used TS6001B in the Bill off material to calculate the cost of the product I had no time to look in more details just used 30min on digikey for a good reference at around 1$. Cost is not so critical for me since is low volume and there are components way more expensive in that product.
Also there are may other factors that contribute to errors so if the spec of the TS6001B is correct I do not realy need better.
I was thinking some of you have use that or selected something else in the same price range.