Voltage Reference, precise, but cheap...

[Doctorandus_P]

I'm looking for ideas for a voltage reference, that is both accurate, affordable and readily available. These factors bite each other, so how do you get them in the same room?

My current thought is to use the good old TL431 (KA431, TLV431 etc) and then ovenize it. To keep BOM cost low, the oven will be just a BJT + power resistor, and the rest is done by the microcontroller, temperate sensor is going to be the BE junction of the transistor.

Is this a reasonably good approach, or are there better alternatives for a 50ct BOM cost for the reference?
Some of the xx431 variants claim some 200uV of stability in the 50 to 70 degree Celcius temperature range, and I'm aiming for about 100uV overall stability for the reference. (Not accuracy, that's another topic).
I don't know much about long term drift. (But there is a topic for this in the Metrology section I believe).

Some more background:
I've already got some power supplies, but affordable power supplies don't have accurate meters, which is quite a shame. Buying a power supply is no fun either, designing & building one yourself is much more fun. There are some very cheap power supplies (I.e. the EUR5 Hiland kit 30V 2mA-3A) and there are a few open sourced SMU like power supplies (but those are quite complex and have a high BOM cost).

One of my long term goals is to design and build a programmable power supply, I want to put a high emphasis on low BOM cost, and therefore integrate as much as feasible in a single microcontroller. It would be fabulous If I could sell these in kit form. It's doubtful that would ever happen, but a man needs a dream, and keeping BOM cost low is a prerequisite for a kit. I'm aiming for a total BOM cost (PCB and all parts on it, but exclusive the raw power supply) for around EUR10. 

The heart will be an STM32 (I need a project to get started with those) with filtered PWM as output to the power section, and it's own ADC for voltage and current read back (Datasheet of STM32G030 claims 16 bit resolution with hardware oversampling, which I find hard to believe, but I'm curious enough to write the software).

For the power, I'm aiming at 0 to 20V and a maximum of somewhere between 200mA to 500mA output current. In a later stadium I may be aiming for a higher powered version, possibly with a SMPS pre-regulator.

This thread is NOT for the whole project. I've already seen, collected and studied 100+ different power supply designs. This thread is only for how to get the most stable 5V reference for a < 50ct BOM cost.

[PGPG]

I want to put a high emphasis on low BOM cost, and therefore integrate as much as feasible in a single microcontroller.

Use microcontroller with buildin ADC and voltage reference.

[PCB.Wiz]

I'm looking for ideas for a voltage reference, that is both accurate, affordable and readily available. These factors bite each other, so how do you get them in the same room?

You can't. If this is a kit you are best to target a couple of choices, and give users control of simplicity.

My current thought is to use the good old TL431 (KA431, TLV431 etc) and then ovenize it. To keep BOM cost low, the oven will be just a BJT + power resistor, and the rest is done by the microcontroller, temperate sensor is going to be the BE junction of the transistor.
Is this a reasonably good approach, or are there better alternatives for a 50ct BOM cost for the reference?
Some of the xx431 variants claim some 200uV of stability in the 50 to 70 degree Celcius temperature range, and I'm aiming for about 100uV overall stability for the reference.

Sounds complicated and Hail Mary. The variations between units would be a wild card.
If you want 100uV on 2v5 ref, over 20'C you are already down into low digit ppm.

[iMo]

.. This thread is only for how to get the most stable 5V reference for a < 50ct BOM cost.

The cheapest 5V Vreg I am aware of which will fit your spec is for example an LT1021-5 (1-4ppm/K, needs at least 9V Vcc) in dil epoxy package purchased off ali/ebay (around 1Euro/pc. some time back). No need to ovenize it for your app.
All others are more expensive, afaik.
Here is my first find https://www.ebay.com/itm/391345708954

[tszaboo]

I admire the dedication, and please do it so we can learn from it.
Or buy an LM4040 for ~30 cents.

[thm_w]

I don't think you want a 5V reference, for STM32G0 ADC you'd want 2.5 or 3 or 3.3V.
If you want cheap, oversample away. If you want 16-bit accuracy, get an external ADC.

I would buy whatever gets you off the ground LM4040B suggested is probably a good one, build in the oven circuitry, get the thing up and running, then start working on tweaking BOM cost. And see if you are even able to measure the accuracy requirements.

https://www.eevblog.com/forum/metrology/is-tl431-good-for-reference-in-19-999-counts-voltmeter-any-variants/msg1821833/#msg1821833
https://www.eevblog.com/forum/projects/low-cost-voltage-reference-experiment/

[The Soulman]

The STM32G030 requires a 3V V-ref not 5V.
I'd use the best V-ref that would fit the bom, I think within the budget you can only have reasonable stability and each finished product would need individual adjustment. (User calibration)

[Doctorandus_P]

If you want 100uV on 2v5 ref, over 20'C you are already down into low digit ppm.

Yes, exactly, that is the problem!
Because of that the built in reference of the uC as PGPG suggests is not good enough.
But it's not hopeless. 1ppm of 5V is 5uV, so 100uV is 20ppm, and if that is combined with a reference that is stable in a decent temperature range and stabilized temperature, then it's achievable. A lot of datasheets I've looked at have tempco's that are relatively flat in a decent temperature range.

The variations between units would be a wild card.

For my budget, compromises have to be made. One time calibration (after burn in) has already been accepted as unavoidable. User calibration is also not a problem for a kit. I recently bought a BM869s, and that is very likely the most accurate DMM I'll ever buy. My goal is to get near the specs of this DMM. People who build the kit can then calibrate it to their own DMM or best reference they have.

LT1021-5

Ali / Ebay is fun for DIY tinkering, but not for kits. Digikey prices are between USD5 and USD17 that's above budget.

Or buy an LM4040 for ~30 cents.

Yes, you probably are right. I had already been looking at the LM4040, it's one of the most common voltage references. Tempco varies wildly depending on brand and sub model. If you search digikey for LM4040 with 5V output, you still have 134 different chips to choose between. But looking at specs, most seem to benefit from an oven. Some have an almost flat section in the tempco between 50 celcius and 80 celcius, and that is a pretty ideal temperature for an oven. Some variants of the xx431 have a similar flat section in that temperature range, and that's what made me look closer at the xx431 in the first place.

I don't think you want a 5V reference, for STM32G0 ADC you'd want 2.5 or 3 or 3.3V.

Yes I do want 5V. Goal is to use use the PWM of the uC, and buffer it first with HCT TTL. (which is powered from the reference).  This needs 50% less amplification after the averaging. 3V3 for the uC is also not clean enough. Also thanx for the links. Those are the ones I had vague memories off. I've archived them and read them again soon.

And there we get to another big problem. Opamps with both high input impedance and low offset also get expensive quickly.
A few years ago I've been looking at the "Scullcom" series. Sort of metrology equipment, but an expensive BOM list and minimal effort arduino software. My goal is to turn it around. Low BOM cost and attempting software (+ hardware) compensation in a sensible mix.

Also, don't hold your breath. I'd be happy if this project is in a working condition two or three years from now.

[magic]

I want to put a high emphasis on low BOM cost, and therefore integrate as much as feasible in a single microcontroller.

Use microcontroller with buildin ADC and voltage reference.

... and ovenize it

[Doctorandus_P]

I want to put a high emphasis on low BOM cost, and therefore integrate as much as feasible in a single microcontroller.

Use microcontroller with buildin ADC and voltage reference.

... and ovenize it

Yes, that has also crossed my mind.
Before I finish this project, I also have a side project in between for a temperature controlling gadget. I'm thinking of using a BJT in TO220. Both as a heater element and as a temperature sensor. Controlled by a uC of course. Goal is to slap it onto some other object to be able to heat it to a specified temperature. Once I have this, I'm a step closer to automated temperature cycling and testing.

[Phil1977]

If you just build for yourself and not for series development, then you easily can record and calculate a temperature compensation table for the microcontroller.

[tszaboo]

My goal is to turn it around. Low BOM cost and attempting software (+ hardware) compensation in a sensible mix.

Also, don't hold your breath. I'd be happy if this project is in a working condition two or three years from now.

As I said, do it, and please document it.
I think you will have a of of 1/f noise which is caused by the oven cycling or PWM noise. But maybe not.
IMHO not everything needs software, and I actually prefer doing projects that need as little or zero software. But I also get the appeal of it.

[ArdWar]

And there we get to another big problem. Opamps with both high input impedance and low offset also get expensive quickly.

I'm not sure what your target here is but if you're willing to do manual calibration then most low cost [[C]MOS]FET opamps are still usable here. Their drift is still in low uV/C range even if they have tens of mV offset.

[youngda9]

I admire the dedication, and please do it so we can learn from it.
Or buy an LM4040 for ~30 cents.

/thread

[PGPG]

I want to put a high emphasis on low BOM cost, and therefore integrate as much as feasible in a single microcontroller.

Use microcontroller with buildin ADC and voltage reference.

... and ovenize it

Yes, that has also crossed my mind.

In 1981 being a student I was building a chrono comparator (device to check electronic watch accuracy based on 32768Hz signal radio-detected from it). One man ordered it from me. It was my first serious money earned on electronic. I got for it about two my fathers month salary. I needed precision frequency source. I have ovenized the quartz TTL generator. I wrapped the 1MHz quartz (in old case 2x bigger in each dimension than HC49) with aluminum sheet and used BD136 as a heater and thermistor equipped with a screw as detector. LM741 was driving the heater. I have put it all (together with TTL 7400 generator) in a made of styrofoam box. It was not small but without it I would probably get around 2s/day accuracy and with it I got about 0.1s/day.
Microcontroller is probably smaller than my quartz alone.
I don't think you need so accuracy when designing power supply. All devices should tolerate at least 10% supply voltage deviations so having supply with 2% accuracy is enough good, I think.

[Andy Chee]

You could potentially use a PTC thermistor as your heating element, as demonstrated in this poor man's crystal oven:

https://youtu.be/KxVJfxiw4ZQ?t=313

[Doctorandus_P]

You could potentially use a PTC thermistor ...

Do you realize that those PTC's with eyelets cost around EUR 5 to EUR7? That eats up more then half the BOM cost for my project. For temperature stability, I'm aiming for a delta of 2 to 4 degrees celcius, and a the same PN junction as the heater transistor is good enough for that.

I'm not aiming for the best I can design / build. I'm aiming for the best I can do at a very low BOM cost.

I don't think you need so accuracy when designing power supply.

For a generic power supply, sure, but I already have those. I am aiming for a SMU.

I admire the dedication, and please do it so we can learn from it.
Or buy an LM4040 for ~30 cents.

/thread

You are probably right. Between the LM4040 and TL431 there is enough room for experimentation. I will probably add some fixed voltage regulators too when doing temperature stability & drift tests.

That means the side projects for this now already are:
* SCPI interface for my BM869S (I'm not buying that EUR80 PC interface). *1).
* Some kind of relay input multiplexer / scanner card.
* TO220 sized "hot plate" for bringing test subjects to specific temperatures.

After I've added the micro SMU, I can do some semiconductor testing and characterization for bigger projects.
It's a bunch of projects, but most are pretty simple, and combined you can do fun things with them.

*1).I know there is already a thread for A program that can read data from many multimeters, and I also want to turn that around. BM869s needs some tinkering to interface it with a PC anyway, so making it SCPI compatible makes it easy to integrate it (or any other DMM after conversion) to integrate with standard (logging) software. With a small Linux SBC you can build a compact test instrument for long term drif testing, log in every now and then to see how it's doing. A scripting interface is much more useful for this then complicated logging software with a fancy GUI and graphing functions.

[Andy Chee]

You could potentially use a PTC thermistor ...

Do you realize that those PTC's with eyelets cost around EUR 5 to EUR7?

Then you're shopping at the wrong electronics parts supplier.  Digikey lists them for USD$1.87 in single units.

[Alex Nikitin]

You can have either precise or cheap, not both at the same time  . One of the options is a reference with an internal temperature sensor, something like REF50 series from TI, for 2.5V it would be the REF5025. It is over your budget but should be good enough even without temperature compensation, and if you need it, you can monitor the temperature directly from the chip. It also should be stable enough in time, which might be a serious problem with likes of LM4040 and TL431.

Cheers

Alex

[ArdWar]

I'm not entirely sure how useful that thermistor (or the proposed BJT) heater arrangement actually will be if you (I think) intent to use the heating element as thermal sensor, and DUT to be exchangeable.

It is usable if DUT is fixed since you can calibrate/compensate for DUT thermal resistance and thermal losses, but will be much less useful if DUT is variable since your sensor then will be thermally too far away from the actual DUT

[PCB.Wiz]

My current thought is to use the good old TL431 (KA431, TLV431 etc) and then ovenize it. To keep BOM cost low, the oven will be just a BJT + power resistor, and the rest is done by the microcontroller, temperate sensor is going to be the BE junction of the transistor.

Since you are keen to ovenize this, and external thermal coupling will be a pain, another idea would be to use a Dual LDO - one part of the same die acts as the heater.

lcsc show many AnaSemi Dual LDOs in the VRDxxyy series that may be suitable, in either SOT26 or DFN6
https://www.lcsc.com/products/Voltage-Regulators-Linear-Low-Drop-Out-LDO-Regulators_387.html?brand=12324

These claim 20ppm typical tempco, and flat temperature curves, and the IQ graphs show some temperature slope  ~ +5uA for 50'C change. A calibrate would be needed for uA -> 'C set point.

An external tiny opamp and load would set the calibrated temperature.

[ArdWar]

A more common idea is to use dual zener and use the second zener as temperature sensor.

The problem here is to figure out if the zeners are actually monolithic with good thermal coupling, or two die with common bond wire.

[PCB.Wiz]

A more common idea is to use dual zener and use the second zener as temperature sensor.

The problem here is to figure out if the zeners are actually monolithic with good thermal coupling, or two die with common bond wire.

Zeners is a whole new rabbit hole to go down... more calibration steps and worse initial accuracy.

More advanced is to use a series combination of Diode + Zener and adjust the bias current for temperature null.

The curves on this part show how that works - tempco nulls at a specific current (which will vary from part to part)
https://fscdn.rohm.com/en/products/databook/datasheet/discrete/diode/zener/rsb6.8smfh.pdf

Or, you can go for the trifecta, and include heater + sense + compensation in one package.
That needs a triple Zener, like BZX84C6V2TS or BZX84C6V2TW (sub 10c). The centre Zener can be the heater and the outside two diodes compensate and sense.
A full setup would need a dual opamp and DAC to give Temperature regulate and current operating point set after calibrate.

Addit : I see lcsc have REF30xx series parts at 0.2% and typ 20ppm/'C, and with good long term drift and hysteresis for  ~20c/150, so the OP may be better to apply KISS

[thm_w]

17c/ea and if you burn them in for 1,000h it drops to 15ppm, nice.

https://www.lcsc.com/product-detail/Voltage-Reference_Texas-Instruments-REF3033AIDBZR_C36658.html

[julian1]

That needs a triple Zener, like BZX84C6V2TS or BZX84C6V2TW (sub 10c). 

It's interesting that triples in the same package exist. Have these ever been tested for noise?