MAX6226 voltage reference

[exe]

Hi there,

It seems there is not much info about MAX6226 voltage reference. So, I decided to start a thread about it.

I bought two pieces of this reference, and today I finally powered it up.

I attach the plot of the very first start. If I read the plot correctly, the vertical scale is 10uV, or 4ppm (still learning how to do plots). The disturbance at around 21:40 is when I put the reference into a plastic jar.

Measurements are done by my (un)trusty Keithley K2000 which I bought on this forum. It has unknown calibration, might be recapped and has some repairs, but to the best of my knowledge it is stable I warmed it up for 2h before starting measurements. Current room temperature is 22.5C. I set NLPC to 10 ("slow" update mode).

Video of the first start: https://www.youtube.com/channel/UCvTzuQaA_jxHFvox4VnlLnQ ("reaction" video, not much is happening there).

[Kleinstein]

10 µV per tic sound reasonable. The plot program and raw data should tell. There should not be guesswork with the scale.

The time is so short than the warm up from self heating of the reference can be a significan part of the curve. So a large part of the observed drift could be just a temperature effect.

Ideally one would also record the actual temperature at the refrence and measure the temperature dependence as a seprate experiment.

[exe]

Here is data over night, from 21:15 yesterday till 9:15 today. The voltage didn't change from yesterday, it still shows rock-solid 2.49994V. The input voltage might dropped by ~80-100mV over night (it measures now 3.27). That's because my DIY power supply is not precise and that's why I'm looking for a better voltage reference .

The temperature dropped by 2C from yesterday. Unfortunately, I don't have any logging thermometer. I'll try to resolve that. In the worst case I'll record my thermometer on webcamera and manually add data .

I said that my K2000, but I actually don't have any evidence to support this. I mean, it's stable up to some 100uV because it agrees with another reference, but that's not enough for this experiment. I also didn't power my K2000 since a long time ago.

I plan to do two more experiments:
1. Change input voltage and see how this affects output voltage. Say, in 2V increments from 3V up to 12V. Probably I should let it set for, say, 30minutes after each change?
2. Do thermal cycling to see hysteresis. Say, up to some 70C and back to room temperature.

[Kleinstein]

The K2000 has a LM399 reference inside, that by  now is probably well aged. The LM399 refrence can still show some random effects (e.g. jumps) comparable or slightly smaller than the max6226. So the meter can still contribute.

For testing a few supply voltage there is no need to wait that long. Some 2-5 min should be good enough.

[exe]

I tested line regulation. The settle time is surprisingly long. Between 3.2V and 5V it settled much faster than between other voltages. Finally, at 13V (that is absolute max for the IC) it was doing a weird thing: see the plot. I'm a bit surprised. Also, the regulation is exceeding 25uV/V specification, unless we use the "box" plot. Might it be I damaged the ic?

PS vertical scale is volts, one tick is 100uV.

[Andreas]

I tested line regulation.

It is a mix of line regulation and tempco due to self heating.

The manufacturers test the line regulation with short pulses only
(leaving the chip temperature constant at 25 deg C)

but of course this does not help in a "real application" where you have the self heating.

with best regards

Andreas

[Andreas]

Finally, at 13V (that is absolute max for the IC)

Hello,

read the datasheet carefully: the maximum recommended operating voltage is 12.6 V.
So the behaviour above 12.6V is not specified.

Did you check for oscillations of the output in this state?

I would also recommend to use only one connection to the DMM
(and solder the other end to the reference instead of using crocodile clips).

with best regards

Andreas

[exe]

Current voltage is 2.499930, which is 10uV below of what was yesterday (I kept DMM and the reference powered the whole time). I attribute this to testing the line regulation yesterday. I'm a bit disappointed.

Today I want to do two things:
1) clean the flux to see if there is any effect (I used quality flux, but there was also flux in the solder, so not sure how the two mix).
2) Heat it up with a hotairgun to, say, 61C for a few minutes, then let cool down and let's see how this affect the voltage.
3) re-arrange DMM crocodile clips to see if this makes a difference on the voltage. That's because ground lead carries some 0.4mA of quiscent current and I wonder if longer/shorter ground lead would affect the reading.

I wish I had a thermal chamber. I have a bottom heater, but it has up to 20C overshoot and poor regulation, so, not exactly suitable for our purposes

read the datasheet carefully: the maximum recommended operating voltage is 12.6 V.
So the behaviour above 12.6V is not specified.

Did you check for oscillations of the output in this state?

I would also recommend to use only one connection to the DMM
(and solder the other end to the reference instead of using crocodile clips).

13V is just 0.4V above recommended operating voltage, I decided to try it . I have no intention to power the reference with voltages above 3.3V in a real circuit. I didn't check for oscillation with a 'scope, but looking at the readings, I think we'd see that on the plot if it was oscillating. There is only one DMM connected to the reference, and I know why you recommend this .

As of soldering, I don't get it. Could you please elaborate more? The power and DMM are connected with crocodile clips. I don't have DMM leads that I could solder.

[exe]

Hey, I reattached negative lead of my DMM and got my 10uV back! Wow, I need to be careful with wiring. Yesterday I re-arranged wires a little bit, may be that's why I got that 10uV drop. That's because I tried to put negative lead as close to the IC as possible. I'll try to reduce length of wires. The downside is, it will be some thermal shock for the IC.

So far the results are not too bad. Next I'll clean the flux and do some very crude thermal cycling (more like thermal shock).

[Kleinstein]

The crocodile clips are not ideal for precision measurements. They may cause some thermal EMF if the is a thermal gradient and the wires may cause mechanical stress to the reference. The air-wired connection to the capacitors is good in keeping the mechanical stress low - it may be better than practical use on a PCB.
I would suggset soldering wires for the supply and meter to the caps, so to get even even less disturbance and possible bad contracts. Just simple banana plugs can cause a few µV of thermal EMF from not that much temperature difference (e.g. after touching them).

[exe]

I cleaned the IC from flux. It recovered to previous voltage pretty quickly. However, then it gradually dropped by 10uV again  I thought it might be due to mechanical stress, so I knocked on the chip with a guitar pick just in case, but it seems this didn't affect it. So, I had another creative idea how to relief the stress, I used a freezing spray to do a thermal cycling to -55C and back to room temperature  . I've got a 10mV jump, which in 5 mins leveled off. But the thing is, the voltage keeps going down. So, it's now at about 2.499918 V, which is 22uV below what I had yesterday. I'll probably leave it alone for the next 12h to see how it goes.

Next I'll try to make a crude temperature logger with a raspberry pi and ds18b20.

[Kleinstein]

cooling to -50 C can be quite some stress and there may be some hysteresist from this.

[exe]

This is data over night. However, the power supply sagged by ~50mV. Considering previous data showed line regulation 100uV/V, this perfectly maps to 5uV change that we see on the plot. So, at least I need to fix that before making any conclusions. I guess all the previous data was affected by that too.

I have a question: is power supply noise important in this case? Asking because I consider using lm317n, which has noise, according to datasheet, about 8uV in 10kHz BW (it's unspecified if it is p-p or rms).

[Kleinstein]

A LM317 regulator for the supply should be good enough. It is not very low noise, but the power supply noise should be attenuated similar to the line regulation. There can be slight differences because of the thermal effects, but the order of magnitude is usually similar for the lower frequency part one usually cares about.

[exe]

The reference (let's call it IC1) settled at around 2.499901, which is like 40uV difference to initial voltage. I guess thermal shocking wasn't the best idea.

For the experiment with a stable supply, I deployed my second MAX6226 (let's call it IC2). The initial voltage is 2.5002.

This time I tried to get the wiring "right". I also increased output capacitance to 1uF. I used a polyester cap because it was lying on my bench (kemet R82DC4100DQ60K). I hope it's fine (I mean ESR and ESL). I didn't install cap on NR pin (noise reduction) because I'm lazy. According to the datasheet, it should lower the reference noise from 75uV/sqrt(Hz) to 45uV or so, and hope my K2000 doesn't care about it.

I didn't find a raspberry pi to do 24x7 voltage and temperature monitoring, so I ordered a new one, raspberry pi with four cores and 8 gigs of ram. A bit pricey and has more power than I need, but I wanted to get the best for my lab . It will arrive tomorrow. When it comes I'll try to solder the output wires to crocodile clips.

[Kleinstein]

A capacitor at the NR pin is filtering some of the higher frequency ( e.g. > 10 Hz) noise. A slow meter, like the K2000 does not care much about the higher frequency noise. So for the test there is no real need for the cap at the NR pin.
The ouput capacitance is also mainly about the higher frequenies. So it should not make that much difference here.
One may still need some capacitance to avoid oscillation of the reference by itself.

The curve from the 2nd chip looks quite a bit better.

[exe]

The voltage dropped more than 15uV over night, which is about -6ppm. I wonder if this is due to different wires I used to connect the DMM. I think I'll replace them just in case. Temperature logging would probably help with the investigation.

[exe]

On ic2 the output voltage keeps going down. It lost ~24uV since first report.

During next month I won't be able to access my lab and I think I'll shutdown the equipment while I'm absent.

[d-smes]

If you can do so safely, I'd leave the reference powered while you are gone.  It will allow the reference to 'burn in' and hopefully stabilize it a bit.

[exe]

Unfortunately, didn't have time to arrange that. On a positive side, this gives me time to make a proto-board with multiple references. So I can test, say, four references instead of one. Also I'll add a temperature sensor. Also, now max6226 is again available on digikey and will be soon in mouser.


PS any ideas how could I make a simple multiplexer for, say, four references? Would CD4051B work? I now have raspberry pi, so I have digital io that I can integrate into my measurement script.

[Andreas]

PS any ideas how could I make a simple multiplexer for, say, four references? Would CD4051B work?

Hello,

Could work if the supply + digital input levels are 3.3 V and input voltage is max 3.3V.
And the input impedance of your Instrument is >> 10 Meg.
For my ageing box I use CD4067 (16-fold) multiplexers.

One for the Reference output and one for the Gnd-Signal
(which differs by up to 30 uV depending on position of the reference).

But since in my case I have a 14V positive supply (since the references are up to 10V) i need level shifters from the 5V digital level to the 14V input level of the multiplexers.

For my LM399 ageing box I prefer MAX4051A (A-Grade).
These have lower leakage current and also level shifters included.

In any case I have 100 nF bypass capacitors at the output of the Multiplexers to avoid EMI effects.

with best regards

Andreas

[Kleinstein]

The CMOS MUX work for signals within the supply range.  So the cheap CD4051 , 4052 and so one work only to the logic supply. For higher voltages I would consider DG408 with seprate supply (up to some +-20 V) for the analog part.

With an isolated meter and having the ground level near the negative supply the CD4051 could also work for 5 V or 7 V references. Just have a -5 to -8 V and +3.3 V supply and the ground level near the negative supply. So the Raspberry ground would not be the reference/analog ground.

[exe]

Thank you very much. Since my references are only ~2.5V, I think I'll try to use 74LV4051/74LV4052 with a signgle 3.3V supply, same rail that I'm using to supply the references.

How important it is how have a separate switch for ground terminal? I'm not looking for absolute precision, I'm only interested in stability (initial drift is some tens of uV anyway). I was hoping to get away with a solid ground poor and not to have an additional multiplexer. Although, probably, it won't be a big problem to do kelvin connection.

[Kleinstein]

For precision circuit a ground pour is poor. One has little control over the actual curren paths and variable supply current at some places can cause extra offsets.
So one should better have dedicated ground traces, ideally to a star ground. Also EMI wise, the left over parts are usually not a valid ground plane. It only needs a few slits to destroy a ground plane.

Depending on the circuit, if there is a good ground, one may get away without separate mutiplexers for the low side. It still depends. If the ADC is single sided, there would be no seprate MUX, but only some mux channels used for ground.

[Andreas]

How important it is how have a separate switch for ground terminal?

Hello,

as long as the power supply current (and the load) are constant it is simply a position dependant offset.
On the other side: when having a 8 channel multiplexer and only 4 references you can also use the spare channels to sense the gnd pin levels. (Quasi-differential with 2 sequential measurements).

with best regards

Andreas