MAX6226 voltage reference

[Kleinstein]

There is quite some correlation between the references. Unless there is a prblem with the supply, chances are this could be an effect of the meter or maybe thermal EMF from the cables. What meter is used for the measurement ?  Looks like it has a good ref. with low popcorn noise.

The dead bug mounted refs look good.

[exe]

The meter is K2000 which was probably re-capped by previous owner(s). K2000's manual says it has 1ppm/C coefficient on 10V .
I too noticed correlation, and I don't know if it because of the meter, because all references exhibit similar tempco . Or may be leads are not very good.

As of IC1 and IC2, those were subject to freezing well below zero (I used a freeze spray), then hotair, then I physically knocked it with quite some force (I wanted to see how sturdy is ceramic carrier), they were also resoldered at least once... May be that's why IC2 is not doing well .

I kept the lab with closed window over night, so  there is not much temperature and humidity variation. Tomorrow I'll probably leave the window open to see how that will affect measurements. The downside is, this can create drafts.

After about a week or so I plan to load one or two references with a 2.4K resistor (~1.04mA current) to see load regulation.

[Andreas]

K2000's manual says it has 1ppm/C coefficient on 10V .

Hello,
where exactly do you read this?

I read:
for 18-28 deg C the tempco is within uncertainity specs.
outside of 18-28 deg C the TC is 2 ppm of reading/K (2,5V in your case = 5 uV/K) + 1 ppm of range/K (10V = 10uV/K) or a total of up to 15 uV/K for this measurement.
I fear you will have to do some T.C. measurements of the instrument first.
See:
https://www.eevblog.com/forum/metrology/project-pimp-a-keithley-2000/msg1106829/#msg1106829

with best regards

Andreas

[exe]

2Andreas: oh, right. There is no spec for it. Well, then I can't really say anything about precision of this instrument. Even if I find a cal lab (probably I'll have to send it to Germany as local labs turned me away because I'm not a company), I guess they charge me extra for climate chamber.

[exe]

Current state of references.

[exe]

Currently logged data. I averaged all values with a sliding window of 1000 measurements. It's still hard to see individual voltages, so I'll re-plot each reference individually.

On Sept 6 I loaded ic1 ic3 ic6 and ic8 with 2.4k resistors (~1.04mA). Vertical black dashed line indicates that moment.

On Sep  1 05:37:45 PM (UTC) there was a 30s power outage in my lab, but I don't see any impact on measurements.

IC2 is a wreck .

[exe]

I'm attaching latest data (extremely averaged so the trend is clear). There is a 5-day gap because script crashed.

From what I see, long-term drift is in spec ("max 11ppm/1000hr Long-Term Drift"). Tempco looks fine too.

I'd say it's good-enough for 4.5digit or may be even 5.5digit portable instruments if environment conditions are stable.

What should I do with it next? Shall I switch it off for one day and then on again to see if there is any hysteresis?

[Andreas]

What should I do with it next? Shall I switch it off for one day and then on again to see if there is any hysteresis?

Hello,

if it is related to humidity one day might be not sufficient.
(Its more like 1-2 weeks off time needed here).

I had one reference (ISL21009) which had after a short power interruption (minutes) a nearly 300 uV offset which needed several days to settle again.

with best regards

Andreas

[exe]

I unplugged my board for two weeks, then plugged again. The DMM was off as well, but I let it warm up for two days before starting measurements.

I don't see any significant drift except ic2.

One thing I was curious is how bad my cables are. So, when the board was unpowered, it was showing 0.47-0.48mV voltage on output. Touching/moving cables would cause -100uV change, with recovery to within 10uV or so within some tens of second. Not sure if this observation means anything other than my cables suck. 10uV is pretty large value comparing to drift of references themselves.

[exe]

Latest data. Periods with less disturbance are when I was absent. I don't know what happened to ic2. It seems it recovered and doesn't suffer excessive drift anymore  .

[dietert1]

After three months of operation one could try correlation plots of each channel agains temperature in order to derive a TC, e.g. for the last data taking period. The resulting TC values will contain a common contribution from the DVM you use. If you then plot the residuals after numerically subtracting the average TC effect for each reference, unexpected changes will be much easier to pin down. I'd guess all your references together can make a sub-ppm voltage standard.

Regards, Dieter

[exe]

My friends, on the latest plot I missed that ic1 went nuts. This is weird. I only see the jump correlates with drop of humidity. May be hermetic sealing is broken or something... Idk, 8 ppm sudden change is very suspicions, considering other references don't behave like that.

[b_force]

Bit of an older topic, but I was just looking for some kind of affordable voltage reference myself.
There doesn't seem to be a whole lot of information shared about these things unfortunately.

I would still like to add some important points to the discussion, especially on the first couple of comments.
It all depends if you're looking for the best possible vs worst case scenario.

Unless I can really specify something, I would go for the last one.
Looking at the long term stability, this will eventually lead to say something like 15-20ppm or so.
Which is an equivalent of 0.002%.
Some devices seem to have an asymptote here, while others just seem to drift forever?

So instead of trying to bring that number down, one can also just use that number as a starting point.
Everything less will be just better in performance.
Obviously it depends on your application, but 0.022% total error is pretty good or very good in most cases.

Although there are two questions from my side;

1 - It seems a little unclear what they sometimes mean by the max accuracy?

According to this article; https://e2e.ti.com/blogs_/archives/b/precisionhub/posts/ic-long-term-stability-the-only-constant-is-change
They see it the rated accuracy as the max accuracy after x many years (see end of the article). While others sometimes seem to use this term as the initial value?
So in the case of the MAX6226, this 0.02% is worst case after many years.

2 - I assume that the long-term stability rating counts for when the device is powered on? Or is there also some natural drifting, even when the device is not being used at all? Certain epoxies and other bonds age over time as well for example.

[exe]

Hi b_force!

Designing around worst-case scenario is a safe bet, but more often than not leads to unsatisfying guaranteed performance. This is why I made this experiment - to check what I can get in practice. As of ageing, its a good question. I contacted their support for that.

[Andreas]

As of ageing, its a good question. I contacted their support for that.

Hello,

please let us know if you get a better/more detailed answer than the typical 11 ppm/1000hrs from the data sheet.

But keep in mind: this ageing drift is done under lab conditions (25 deg C constant over time).
So practically you might have different conditions.
An in my ageing experiment I see sometimes sudden drift steps which I cannot explain.

with best regards

Andreas

[b_force]

Hi b_force!

As of ageing, its a good question. I contacted their support for that.

Extremely curious about this, please keep us updated!

[b_force]

As of ageing, its a good question. I contacted their support for that.

Hello,

please let us know if you get a better/more detailed answer than the typical 11 ppm/1000hrs from the data sheet.

But keep in mind: this ageing drift is done under lab conditions (25 deg C constant over time).
So practically you might have different conditions.
An in my ageing experiment I see sometimes sudden drift steps which I cannot explain.

with best regards

Andreas

I was also thinking more in the practical sense of aging.

What happens when those IC's are in storage for a long time, do they need to be kept at constant temperature/humidity level etc etc?
I can't find any information about it with actual data, only anecdotal and personal experiences.

[b_force]

Well, an update from my side, since I had to ask some other things anyway'

1) 0.02% initial accuracy is the worst case scenario one can expect.
 
2) Yes, the stability counts when the device is powered On
 
3) No the device does not need to be stored in a certain way

[Svgeesus]

Sorry for the necro-bump but this thread seemed the best place to put this.

Context: I have built a small voltage reference board using the ceramic package MAX6226_50 together with a dual low-drift, low-offset op-amp, OPA2186D. One half of this is used as a unity gain buffer (voltage follower) while the other half provides unity gain inversion. This board will interface with two other DAC boards; the AD5781ARUZ DAC requires both a positive and a negative voltage reference. This is a "fit it and forget it" application; after burn-in and measurement, the circuit is to be used without further measurement or adjustment for some years.

Circuit: ±9.5V power from linear regulators. MAX6226_50 has 1μF and 100nF input caps, 100nF noise reduction cap, 1μF and 100nF output caps. The 100nF are C0G and the 1μF are X7R.

Finding: The reference has been powered on for four days now and is settling nicely. Measured with Keysight 34465A at 100PLC 4.999727V sd 6µV. However, it has twice made sudden jumps in output: one of about 30µV jump when it became less noisy, then hours later -30µV jump and became more noisy again. Measurements are at the output of the voltage follower.



(The photo was taken when I was monitoring the inverting output, but this post is regarding the positive output)

[Kleinstein]

In the times when noisy the reference may be oscillating. The ground part of the layout is not visible an may not be that great. E.g. adding inductance to the supply decoupling or the load side capacitance.
The AZ type OP amp has some current spikes that can cause ringing at the reference. So ideally one would have a little filtering between the reference and the voltage follower.
The OP amps is also noisy on the supply side and ideally has good decoupling with series impedance in the supplies.

It is not in the DS, but usually a voltage regulator or similar output wants a mix of a relative low ESR capacitor and some larger lossy capacitor, e.g. with series resistance of a few ohms.
The DS example curves with 10 µF load capacitor already show some ringing on the transisents. Some 1-10 ohm of series resistance to the 1 µF cap may help.  For a test one could also use an electrolytic instead of the X7R.

The resistors for the inverter don't look right: 100 K is too high and add quite some noise and the OPA2186 may not be that happy with such a high impedance. The right choice would be more like a 10K/10K or maybe 5K/5K resistor pair as an array to get good matching (e.g. as div23). At least the resistor should be good thin film ones, possible as 2-4 in parallel each (some types, like Susumu RR are cheap in quantety) for better averaging / matching by numbers.

[iMo]

Also do not use the solder-less breadboard (and jump wires as well) when chasing uVolts..
A slight knocking on your desk may cause your issues easily..

[Svgeesus]

I was worried that the op-amp was oscillating; I had not considered that the reference could also be oscillating.

Not visible from the photo, but the entire back side of the board is a ground plane (minus a couple of power tracks, as this is a two-layer board).

Thanks for the hints regarding series impedance. I haven't designed with inductors, what sort of ballpark values am I looking to use here?

Regarding the resistors for the inverter: I had planned to use a pair of 10k resistors here, Susumu RG2012V-103-P-T1 10k 0.02% 5ppm/C 0805 which are $2.57 each if you get 25; I had some left over from another project.

When I built up this board I couldn't find them, however, so I substituted Susumu RG2012N-104-W-T1 100k 0.05% 10ppm/C which I had on hand. As you say, the value is too high.

I had also looked into the Vishay MPM divider resistor pairs, but Mouser and Digikey don't stock the best F and Z grades (0.025% and 0.01% ratio matching, respectively) and they have to be special ordered in lots of 1000. So the individual P grade Susumu RG2012N was better specified.

Thanks again for the rapid and helpful feedback.

[Svgeesus]

I can try soldering the wires that go to the meter; but I suspect the oscillations are the big culprit here.

[Kleinstein]

Oscillation of the OP-amps can be an issue with capacitive loading of the output. For a test some 100 ohm in series, close to the PCB could be a good idea.

The PCB takes care with drive and sense at the output side, but the ground part is not ideal. For the given circuit it is not too bad as there is no other DC GND current than the reference chip. The GND side for the reference should be ideally taken from close to the ref. chip, e.g. from the via.

[Svgeesus]

I made another test with a second sample of the voltage reference. The differences to the first experiment:

• no op-amp is fitted
• measurement wire soldered directly to the junction of the force and sense traces
• no breadboard

Here are the measurements of the first few minutes after power-on, and then a two-hour measurement.

As there is no AZ op-amp here and the results are similar to the other board: 5.00014V sd 10µV it seems that the reference is oscillating?

(Sadly my oscilloscope, Owon SDS 7102 which has been in storage a couple of years, appears to be busted. It switches on, the screen lights up, there is no trace and none of the menu buttons do anything. I plan to get a Siglent SDS1104X-E once I have the money. So not possible to verify oscillation directly.)