Voltage Reference Newbie Questions

[Sam__]

hello voltnuts,

I've just gotten back into hobby electronics and would like to have a go at building a reasonably good voltage reference. Mostly to practice my fundamentals and to have something that may be useful in the future.

I've been reading around the last few days and would like to ask some theory questions to get your expert answers and make sure I'm understanding what I'm reading. Apologies for it being a bit of a brain dump but I feel like I'm trying to process a lot of information at the moment and not sure yet how it all comes together.


The main variables we're trying to control with a voltage reference are noise and drift. Is there anything else that needs controlling?

1. The major factors contributing to noise are the components themselves and "outside world" interference. Which is dominant?

2. The major drift factors are time, temperature and humidity. Are there any others?


Design considerations:

Supply voltage: The supply to the voltage reference should be stable. Would using another voltage reference with a suitable output current be a good option for this?
 
Temperature control: If one were to try and control the temperature with power resistors for example. Using a micro controller controlling a current source. Would this potentially introduce noise beyond the usefulness of maintaining temperature control?

Buffering: What are the advantages of buffering the output of a reference? Surely this just introduces more potential noise and error sources?


Practical implementation:

I've seen slots used in PCB design to minimise physical stress on the reference part while also isolating from other PCB signals & potential noise sources. Is this always a good idea?

EMI: Is the easiest way to deal with this throwing it in a metal project box?

PCB ground plane. Is this a good idea around the reference to reduce noise?

Trace length and width - longer traces will pick up more EMI. Wider traces have more impedance and capacitance. So keep them short and thin?

Thanks in advance.

[Kleinstein]

The main noise source are usually the reference elements itself. There are contributions from an ouput buffern and supply noise not getting perfectly suppressed, but usually these are small contributions.
With most reference chips it is sufficient to use a standard voltage regulator (e.g. LM317, 7815 or similar). Some circuits use bootsstrapping, using the votlage reference to regulate the supply.

A regulated temperature is sometimes used.  A µC for regulation is possible, though many such circuits are from times before µCs got popular. Because of the power consuption modern circuits seem to prefer more a correction than temperature regulation. Still for ultimate performance it is hard to beat temperature regulation.

For EMI a metal case with suitable filtering is the obvious way. It may still help to also have the circuit inside still a little tolerant. So the case alone is not enough. The EMI level can get quite high near a mobile phone.

A PCB ground plane can help with EMI, but it can also cause trouble with no so well defined ground return paths and voltages in the µV range.
Slots in the PCB can help with board stress, but it is not perfect. The slots can increase the suceptibilty to EMI e.g. as the ground plane gets holes.
A ground plane with lots of slots can be quite bad WMI wise.

For the drift there can also be an effect of mechanical stress on the chip - here it can be indirectly humidity causing the stress, but other effects are possible too ( e.g. creep in solder).

[Grandchuck]

As one option, Dr. Frank (a contributor here) has designed a circuit based on the LTZ1000 that has been built many times and gives good results.  A circuit board is available.

https://oshpark.com/shared_projects/dGYcQAgn

[Sam__]

The main noise source are usually the reference elements itself. There are contributions from an ouput buffern and supply noise not getting perfectly suppressed, but usually these are small contributions.
With most reference chips it is sufficient to use a standard voltage regulator (e.g. LM317, 7815 or similar). Some circuits use bootsstrapping, using the votlage reference to regulate the supply.

A regulated temperature is sometimes used.  A µC for regulation is possible, though many such circuits are from times before µCs got popular. Because of the power consuption modern circuits seem to prefer more a correction than temperature regulation. Still for ultimate performance it is hard to beat temperature regulation.

For EMI a metal case with suitable filtering is the obvious way. It may still help to also have the circuit inside still a little tolerant. So the case alone is not enough. The EMI level can get quite high near a mobile phone.

A PCB ground plane can help with EMI, but it can also cause trouble with no so well defined ground return paths and voltages in the µV range.
Slots in the PCB can help with board stress, but it is not perfect. The slots can increase the suceptibilty to EMI e.g. as the ground plane gets holes.
A ground plane with lots of slots can be quite bad WMI wise.

For the drift there can also be an effect of mechanical stress on the chip - here it can be indirectly humidity causing the stress, but other effects are possible too ( e.g. creep in solder).

Thank you so much for the information. Good to know the supply can be as basic as an LM317.

I'm tempted to include thermal regulation in whatever I end up designing. Just to give me the option. Especially as my office can vary from 15C to 30C.

Thank you for the information on ground plane and EMI trade offs. Seems like I need to learn more on that front.

I didn't realise solder creep was a thing! That's interesting.

As one option, Dr. Frank (a contributor here) has designed a circuit based on the LTZ1000 that has been built many times and gives good results.  A circuit board is available.

https://oshpark.com/shared_projects/dGYcQAgn

Thank you for the link. I will have a read of that thread and see what I can learn.

[Andreas]

Temperature control: If one were to try and control the temperature with power resistors for example. Using a micro controller controlling a current source. Would this potentially introduce noise beyond the usefulness of maintaining temperature control?

Hello,

I would not use (a single) power resistor. Since it generates a single hot spot and thus temperature gradients on the PCB.
It is better to have a spreaded heater, at least around the reference and eventually the amplification resistors.

Other design considerations:
- use a device with buried zener they have the lowest ageing rates
- use a hermetically sealed package like TO-99 or CERDIP to avoid influence from humidity
- a larger package has less influence from the PCB (avoid SMD packages)

As beginner I would rather start with one of the other buried zeners and not the LTZ1000 which is very sensitive to EMI influence and needs some care with chosing resistors.

So either start with a LM399 (heated), AD587, AD586, LT1021, LT1236 or similar.

In this thread I am doing some EMI-Measurements and countermeasures (Capacitors, common mode chokes) also output buffering with the "right" OPAMP helps.
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2684070/#msg2684070

Some references also can be passively temperature compensated in a small temperature range.
Here a example with AD587 and some NTCs.
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg1449487/#msg1449487

Trace length and width - longer traces will pick up more EMI. Wider traces have more impedance and capacitance. So keep them short and thin?

Wider traces have less impedance. (What you want if drawing current from the reference).
You want them short and with a low (loop) area between signal and ground to reduce EMI pick up.
The width should not be too thin.

with best regards

Andreas

[Kleinstein]

Solder creep was at least an issue in the transition from leaded to non leaded solder. The softer leaded solder allows better relaxation of stress between pins.  AFAIR this was an issue with the durability of large BGAs - this was some 25 years ago, so the durability issue is likely solved, but precision parts may still react to the stress with drift.

The LTZ1000 and LM399 references include temperature regulation on the chip - so likely no need for external regulation.

Heating a reference circuit has another effect: the relative humidity will go down (roughtly half every 10 K) and this possibly also helping with humidity effects, at least if powered 24/7. If only powered now an than the change in humidity can also be a problem.

[Sam__]

Temperature control: If one were to try and control the temperature with power resistors for example. Using a micro controller controlling a current source. Would this potentially introduce noise beyond the usefulness of maintaining temperature control?

Hello,

I would not use (a single) power resistor. Since it generates a single hot spot and thus temperature gradients on the PCB.
It is better to have a spreaded heater, at least around the reference and eventually the amplification resistors.

Other design considerations:
- use a device with buried zener they have the lowest ageing rates
- use a hermetically sealed package like TO-99 or CERDIP to avoid influence from humidity
- a larger package has less influence from the PCB (avoid SMD packages)

As beginner I would rather start with one of the other buried zeners and not the LTZ1000 which is very sensitive to EMI influence and needs some care with chosing resistors.

So either start with a LM399 (heated), AD587, AD586, LT1021, LT1236 or similar.

In this thread I am doing some EMI-Measurements and countermeasures (Capacitors, common mode chokes) also output buffering with the "right" OPAMP helps.
https://www.eevblog.com/forum/metrology/emi-measurements-of-a-volt-nut/msg2684070/#msg2684070

Some references also can be passively temperature compensated in a small temperature range.
Here a example with AD587 and some NTCs.
https://www.eevblog.com/forum/metrology/ad587lw-10v-precision-travel-standard/msg1449487/#msg1449487

Trace length and width - longer traces will pick up more EMI. Wider traces have more impedance and capacitance. So keep them short and thin?

Wider traces have less impedance. (What you want if drawing current from the reference).
You want them short and with a low (loop) area between signal and ground to reduce EMI pick up.
The width should not be too thin.

with best regards

Andreas

Good thinking on the temperature gradient. Maybe an array of resistors carefully placed, or a TEC, or a resistor+heatsink combination could work to create a more even temperature?

I have an LT1236 around here in a box so will experiment with that first!

Thank you for the link. I will have a read.

Doh! on the trace width, yes of course.

Solder creep was at least an issue in the transition from leaded to non leaded solder. The softer leaded solder allows better relaxation of stress between pins.  AFAIR this was an issue with the durability of large BGAs - this was some 25 years ago, so the durability issue is likely solved, but precision parts may still react to the stress with drift.

The LTZ1000 and LM399 references include temperature regulation on the chip - so likely no need for external regulation.

Heating a reference circuit has another effect: the relative humidity will go down (roughtly half every 10 K) and this possibly also helping with humidity effects, at least if powered 24/7. If only powered now an than the change in humidity can also be a problem.

Interesting information about the history of solder. I still use leaded at home so hopefully that will help minimise the impact of solder creep.

I'm thinking, if this ends up being fairly low power I could leave it powered on all the time on a shelf, so maintaining lower humidity could be an advantage.

[Andreas]

I have an LT1236 around here in a box so will experiment with that first!

Ok.

If it is the A-Grade I usually measured a (near linear) T.C. of +2 .. +4 ppm/K near room temperature (10-40 deg C) on different devices.

with best regards

Andreas

[YetAnotherTechie]

Since this is a newbies thread, i'd like to ask: how can you measure a reference without having a much better reference? as i understand it, not everyone here that builds them has a room with a 3458A at 23*C to check against, but still manage to do it. How to separate the drifts(aging, thermal, humidity) of the meter from what you're measuring, just statistics? Can i get any meaningful data from a LTZ1000 or LM399 with a 34401a or keithley 2000?
Thank you.

[Andreas]

Hello,

for temperature it is easy:
use a reference voltage in a constant environment (room temperature or better) and then temperature cycle your DUT to determine the T.C.
for my Keithley I got a T.C. around +0.55 ppm/K
https://www.eevblog.com/forum/metrology/project-pimp-a-keithley-2000/msg1106839/#msg1106839

Now you can correct your readings improving the measurements by around a factor of 10.

Humidity and ageing is much more effort.
on my both self built instruments with a 24 Bit ADC (ADC4 + ADC8) with a LT1027CCN8 plastic package I tried to calculate the compensation values.
With measurements over 3 years and putting a linear fit over ageing and humidity I got around 0.5ppm / % relative humidity and different average ageing coefficients.

Note: to get the "best fit" you have to take the time constant of humidity into account so the humidity values are PT1 filtered e.g. over 5 days to get the influence on the reference.

Of course the ageing coefficients are relative the ageing of the LTZ based reference.  (which should be added).
And humidity compensation is far from perfect. (I only measured once per day instead of integrating up over the whole day).

So humidity compensation is very difficult since you need to know the history of the device at least over the last week(s).
Now I try to avoid the influence of humidity by references with hermetically tight housings and decoupling them from the influence of the still humidity sensitive Epoxy-PCB.

with best regards

Andreas

[YetAnotherTechie]

Mmm, but if i do this with a newly build reference, it will be drifting the most it will, at the same time than i'm temp. cycling the meter...
If so, unless i have an already aged reference, i won't be able to see how the newly build reference drifts, is that correct?

[Kleinstein]

If one wants to see the initial drift, it may be better to first start with a test at constant temperature and do the temperaure cycle later.
The initial dirft can be effected from soldering and the PCB.  Especially initially there can be several processes combined. So the first hours are not that interesting as the fast processes can stabilize quite fast and later don't matter. A reference may drift a lot in the first 10 hours ( especially after rather high temperature soldering) and still stabilize to get good later on.

[Andreas]

Mmm, but if i do this with a newly build reference, it will be drifting the most it will, at the same time than i'm temp. cycling the meter...
If so, unless i have an already aged reference, i won't be able to see how the newly build reference drifts, is that correct?

Hello,

thats right.
You need one pre-aged device which drifts less than one tenth of the expected temperature drift during the cycle (of e.g. one day).
The largest ageing is during the first 2 kHrs. But it may last even more than 5 kHrs until some devices stabilize. I also have devices that never stabilize like my LM399#1.

with best regards

Andreas

[Doctorandus_P]

An observation:
The green and read traces in 20140905_AGE_LM399_1.PNG track each other suspiciously close. It could be that they are both stable, and the reference used to measure those thraces is drifting itself. The light blue rH % (air humidity?) also has peaks and valleys at the same places. So there is a chance both red and green react the same way to changes in air humidity, but that does not clarify the overall downward trend of both red and green, while the first order approximation of cyan is an almost horizontal line.

I feel like using a reference that's 10x better then what you have is sort of cheating. It's nice for quick commercial calibration, but at some point there simply is no reference available that is better then what you want to calibrate.
You can also build a few different references, and then compare them to each other.

Putting a few in a sealed box with desiccant (and a separate humidity sensor?) may also give interesting information for comparison.

[Andreas]

An observation:
The green and read traces in 20140905_AGE_LM399_1.PNG track each other suspiciously close. It could be that they are both stable, and the reference used to measure those thraces is drifting itself. T

Yes you are partly right. I picked an old measurement with ADC4 and ADC8 which are both LT1027CCN8 references so in plastic DIP package. Both have a sensitivity to humidity of around 0.5 ppm per % rH. So part of the swing, around 10 ppm or 35 uV is related to ADC4 and ADC8. The other 10 ppm are from LM399#1.

Later I changed to measurements with ADC#13 which has a AD586LQ as reference.
ADC13 has no sensitivity to humidity.
But LM399#1 has still seasonal changes of around 10 ppm.

with best regards

Andreas

[branadic]

I wonder, if you have an explaination for the seasonal changes on your LM399, as the package is hermetically sealed. Is it an influence of the opamps being used, is it an effect of the printed circuit board.
Stress due to moisture in the pcb is not what I would expect from a TO package, but moisture influencing the isolation of the pcb traces, creating parallel resistance to the copper traces, could be much more likely.

-branadic-

[KT88]

Rth of air is dependant on the humidity.
https://www.electronics-cooling.com/2003/11/the-thermal-conductivity-of-moist-air/
No mentioning whether this was absolute- or relative humidity though.
This might be causing the observed effect....

Cheers

Andreas

[dietert1]

Andreas, you suppressed the first 1000 days in this diagram. Is this because you changed the metering setup?

I think with this kind of setup and after some years of observation one gets to a very comfortable situation. Probably one can reject red, blue and light blue. Green is a drifter that may be useful after subtracting its linear drift. Then with violet, yellow, green (and maybe the reference in the meter) together you have a redundant, stable reference. Probably a calibration every 5 years will be enough to stay within 1 ppm. This is roughly the approach Fluke has been taking. They would reduce error estimates for a reference after having several years of calibration records. It may be a bit frustrating though for a newbie.

A year ago i put some LTFLU 10V references into hermetic boxes with desiccant and separate humidity sensors and already after one year the results definitely look like sub-ppm (see LTFLU thread).

Regards, Dieter

[Andreas]

I wonder, if you have an explaination for the seasonal changes on your LM399, as the package is hermetically sealed. Is it an influence of the opamps being used, is it an effect of the printed circuit board.

Hello,

I am not shure wether LM399#1 is really (still) hermetically sealed.
It was a used part which I have desoldered from an old measurement system.

There are also no opamps involved.
I have only a temperature compensated LM334 based current source connected.
So the same as with LM399#2 which shows no humidity influence.
So I do not have really a explanation for the LM399#1 seasonal changes.

LM399#3 is also unbuffered but has a current source which is supplied from the own reference voltage.

Rth of air is dependant on the humidity.
This might be causing the observed effect....

Also a possible explanation. (together with the "high" raw voltage of 7028 mV which gives a large unheated T.C. compared to the Sweet spot of 6875 mv)
But since this is no explanation for the large ageing rate; I think its rather a lost hermetical seal.

Andreas, you suppressed the first 1000 days in this diagram. Is this because you changed the metering setup?

I think with this kind of setup and after some years of observation one gets to a very comfortable situation. Probably one can reject red, blue and light blue.

Suppression is only done since my visualisation program cannot handle more than 2000 points.
Starting point is somewhere where I added LM399#CH6 + LM399#CH7 to the measurements.

And all measurements are relative to the temperature compensated reference voltage (AD586LQ) of ADC#13.
LTZ#2 was my best reference until I shorted the unbuffered output (event of -5ppm).
Since then the drift is increased. And I have no explanation for the jump back around day 2200.
I really cannot recommend using unbuffered LTZ1000(A) references.

with best regards

Andreas

[branadic]

@KT88

I was more referening to electrical conductivity / isolation due to moisture in the PCB. Some mΩ change of a copper trace due to a humidity dependant parallel resistor to it can have quite a large impact of a few µV.

-branadic-

[Kleinstein]

The LM399 is low impedance, as it is more than just a zener. So added lakage currents are less of a problem. With Cu traces this is even less of an issue. However the LM334 current source may react to humidiy quite a bit.

[Andreas]

Hello,

lets do the math:

the dynamic zener resistance is below 1 Ohm.
we are looking for 10 ppm = 70 uV change of the 7V zener voltage.
With 1 Ohm we have to change the 1 mA by 70 uA.
So 7% or 70000 ppm.
I would say: very unlikely.

with best regards

Andreas