Beginner level: DIY an accurate picoammeter (measure picoamps!)

[SilverSolder]

Looks good!

[NeverDie]

Here's the test circuit that will go inside the test enclosure. 



I'm laying it out here because it may be easier to see this way than when it's folded up inside the test enclosure.

The BNC connectors will connect to the passthru connectors.  As configured, both the uCurrent Gold output and the picoammeter will share the same ground.  Hopefully that won't be a problem.  If it turns out that it is a problem, then I can install an isolated BNC passthru on the lunchbox to keep their grounds separate.

Test Plan:  Initial voltage will be set to 1v, which should translate into 1 nanoamp.  Therefore, both the uCurrent Gold and the picoammeter should be able to measure it.  From there I'll reduce the voltage in steps and record both measurements with each step down.

I'm aware that the buck converter is a potential source of noise, but in previous tests it didn't seem to bother the picoammeter, possibly because of its high switching frequency.  Nonetheless, I will take the minimum precaution of resting it on some 1/8" thick unclad FR4 so that it doesn't make direct electrical contact with the lunchbox through its metal standoffs. 

One nice thing about the buck converter is that it has current limiting, which I set to 10ma.  Therefore, if a short were to occur, maybe that would help prevent a catastrophic failure.  Ideally I would set CC even lower, but I'm not sure how well the chinese design and components would behave if operated near its limits, so I gave it some extra headroom just to ensure that it won't trip over itself.

[SilverSolder]

The uCurrent will respond to the noise, up to hundreds of KHz in my testing...

[NeverDie]

Good to know. 

What exactly is the chart illustrating?  I'm lacking the full context. 

[SilverSolder]

Good to know. 

What exactly is the chart illustrating?  I'm lacking the full context.

It is called a Bode plot.

The frequency is along the bottom axis, goes from 500Hz  to 500KHz (logarithmic)

The top trace, "Amplitude", shows the loss through the uCurrent in dB (compared to DC).  So, you can see, as the frequency increases, at some point the output of the uCurrent starts to drop off.   At -6dB, the output has fallen to half (not shown).

The bottom trace, "Phase", shows how many degrees behind the sine wave coming out of the uCurrent is compared to what went in.

To make Bode plots like this is super useful to understand all kinds of electronic components.   There are all kinds of cool tools available to do that these days, many scopes have it built in,  or you could use something like an Analog Discovery 2.

[NeverDie]

The immediate problem turns out not to be the buck converter but instead the picoammeter sharing the output ground with the uCurrent Gold.  With the picoammeter powered on but both with or without power to everything else, what appears to be a positive feedback loop of some kind raises the voltage on the output of the picoammeter until it's about 4.5v (hmmm..  half of 9 volts powering the picoammeter?).  It happens whenever I connect the uCurrent Gold BNC connector to the second BNC passthrough.  It happens even if I remove both the battery and the buck converter from the enclosure.

So, I guess this means I may need to use an isolation BNC conector to keep the grounds separated.  However, if I do that, there will be nothing to stop outside noise/interference entering into the enclosure from the DMM intended to measure the uCurrent Gold, so I'm guessing that won't be a good idea. 

I think maybe maybe putting  an arduino (or a logging DMM) inside the enclosure to log the voltage output of the uCurrent gold when the lid is closed might be the next step.  That would allow the uCurrent Gold's output voltage to float, which seems to be the way it was designed to operate.

Anyone have any other suggestions?

[NeverDie]

Well, OK, probably the easiest way around this will be to take the picoammeter and uCurrent Gold measurements separately rather than at the same time.  Maybe not ideal, but worth a try.  I'll give it a go.

[NeverDie]

I did it, and the results are interesting:

1.  The uCurrentGold is probably good enough for most picoamp measurements.
2.  The DMM is the weak link in this test setup: because it is outside of the enclosure, it is still influenced by electrostatic fields.  There's just no denying it: I can see it with my own eyes.

Therefore, to get repeatable single digit picoamp measurements, either a lot of external ESD protection needs to be in place (?) or else the DMM needs to take its measurements inside the test chamber enclosure.  Maybe both.

What follows are the details. 

This is what I had planned to use as the test setup before encountering the grounding problem described above:



Therefore, to take the uCurrentGold measurements, I disconnected the picoammeter and shorted the uCurrent Gold's negative input lead to ground using an oscilloscope alligate clip, as shown next:



Here are the measurements.  The uCurrentGold was switched to the nanoamp setting.  Input voltage is just what was reported on the face of the cheap buck converter (for this first set of measurements, I maybe should have checked the voltage with a voltmeter, but for expediency's sake, I just used the voltage that was displayed on the LEDs of the buck converter):
0.1v input, 0.23mv output
0.2v, 0.33mv
0.3, 0.43
0.4, 0.53
0.5, 0.63
0.6, 0.73
0.7,0.83
0.8, 0.93
0.9, 1.03
1.0, 1.13

So, to the nearest 10 picoamps, the results were remarkably consistent.  From eyeballing the numbers, it looks as though there was a 0.13v offset that showed up in all the measurements.

[NeverDie]

Then, for the picoammeter readings to approximate the same test conditions, I wired the uCurrent Gold in series with the picoammeter and let it run, even though I disconnected the uCurrent Gold's output for the picoammeter measurements:



The picoammeter measurements (see the following) were quite consistent with those of the uCurrent Gold above:

0.1v, 115mv
0.2v, 215mv
0.3v,315mv,
0.40, 412
0.5, 516
0.6, 613
0.7, 0.715
0.8, 0.813
0.9, 0.918
1.00, 1.015

This was just a quick first pass, but the results seem quite linear.  This time the measurement offset appears to be only about 15mv.

In conclusion, I'd say either the uCurrent Gold or the picoammeter is probably good enough for measuring to the nearest 10 picoamps.  If I were to repeat the measurements and strive for greater accuracy,  I'd want either 1. the DMM to do its measuring inside the test enclosure to see whether it cuts down the noise any further or 2. for the DMM to be properly shielded (just as the picoammeter already is) if the DMM is to remain external to the enclosure.  I'd be interested to hear ideas on how #2 might be accomplished while still keeping a visible display.  Otherwise, option #1 becomes the choice by default.

With those improvements, I'm fairly confident that more precise measurements of one form or another should be possible.  However, even without those improvements, the existing setup is probably already good enough for meeting my near-term measurement needs. 

 

I hope posting the above measurements enriches the discussion.

Thank you everyone for your helpful comments and suggestions.

[SilverSolder]

That is super interesting - it seems that your theory that the uCurrent suffers from not being in a shielded box, might have something to it!

[Gyro]

Nice test. 

A couple of things that come to mind:

1. The uCurrent gold uses a shunt resistor + voltage amplifier, rather than the transimpedance amplifier Picoammeter. The Picoammeter will attempt to keep its input at 0V, wheras the uCurrent will show a voltage drop (voltage burden) across its shunt resistor.  The uCurrent is designed to keep its burden voltage as low as possible by using lots of voltage gain, but it is still there. Your method of generating the test current, using a low output voltage divider and series resistor, will tend to exaggerate the effect of the burden voltage (it would be different if you were using a higher supply voltage and much higher series resistor value).

2. Those little SMPS modules are pretty noisy. That will probably affect the uCurrent more (yes it would benefit from a shielded box due to its high voltage gain - and probably smaller terminals too  ).


P.S. You should be able to calculate the effect of the voltage burden on the uCurrent by including its shunt resistor value (10k on nA range?) as part of your resistor network.

[NeverDie]

I'm willing to remove the smps from the equation.  What do you recommend I use in it's place?  A 9v battery with potentiometer to dial-in the voltages?  Or something else? 

One alternate I can think of would be to use a large supercap with a low ESR to create a voltage source by just bleeding it down to different test voltage levels.  It would be a bit more finicky to setup up though because of supercap tendencies to "regenerate" after voltage drops: I'd have to let it sit for quite a while for it to stabilize after adding or subtracting a large chunk of charge to/from it.

What do you recommend for an adjustable lowest noise voltage source?

[SilverSolder]

A battery or supercap would both work - as long as you know what its voltage is,  you can work out mathematically what you should be reading from the test, and compare with that...  no need to adjust the voltage or anything. 

The main thing is that the voltage has to be reliably stable (which, over a period of a few minutes of testing at low current, a 9V battery will provide, and probably a supercap too).

[NeverDie]

Some of the adjustable LDO's are billed as "ultra low noise."  Any favorites for that category?

On the other hand, Dave Jones did a whole video on how linear regulators aren't actually all that different from switched mode power supplies.  Or, at least, not as different as you might think.

[SilverSolder]

I don't recall the Dave video, but was he saying that a linear regulator cannot quite clean up a switch mode power supply?  -  if you use it to regulate a battery or supercap, you'll be fine!

[Kleinstein]

For the test signal I would use a battery and pot to set a value. At the current test level the relative resolution is limited anyway, so a battery is stable enough if not loaded very much. For low voltage / low power a reference chip could make a good low noise regulator.

An LDO or similar regulator can help a little with the ripple from an SMPs, but the effect is limited. No need for high efficiency for a test current in the nA range.

Higher frequency EMI could effect both the µCurrent and the TIA pA meter. So it is usually not such a good idea.

[NeverDie]

LT3042 was the best ultra-low noise adjustable LDO that I could find:  https://www.mouser.com/datasheet/2/609/3042fb-1271662.pdf
It's adjustable all the way down to 0 volts.  If it has no advantages though, I'm happy to use just a 9v battery and trim pots, since that is easier to build.

[SilverSolder]

@Kleinstein is right, just go with a pot and trim it in.  Simple and good and good and simple!

[NeverDie]

If you want to get exact, the resistor values aren't nice clean numbers because you have to account for the parallel resistances:

 R2=9980, and R4=9891.  Assuming the 9v battery is exactly 9v, then R1=7271.  R1 would need to be adjusted, depending on what the true voltage on the battery actually is.

This way, with the potentiometer set to 1000 ohms, it generates 1na = 1000pa of current at the output (ammeter A5 in the schematic).  i.e. 1 picoamp for every ohm on the potentiometer, making it's easy to dial-in an exact pa current if it's a 10-turn 1K potentiometer with a dial-indicator on it.

A bit of a hassle to construct, but it passes SPICE simulation.

If I'm not mistaken, attaching a current mirror to the output would effectively create a picoamp current source that you could also use in other applications.   

In my case, though, instead of 9v, I'm going to use an REF102CP precision 10-volt reference, just for convenience.  REF102CP outputs exactly 10 volts to within plus or minus 0.0025 volts, so plenty good enough for this endeavor.  It means not having to worry about or recheck the source voltage level, as it will always be 10v pretty much exactly.  REF102CP has practically no noise: just 5uVpp at the chip, and that number will get divided by 10 million after it goes through the voltage dividers.  https://www.ti.com/lit/ds/symlink/ref102.pdf

Edit1: The other advantage of making use of a 10v reference voltage source is that with it, each stage, if equiped with a trim potentiometer, can be precisely set by applying a precise 10v to it and then, measuring with a dmm, tweaking the potentiometer until the correct voltage division is achieved.  This, then, gives an exact result to the accuracy of the DMM, which is more accurate measuring voltages than resistances anyway.  No need to calculate the effect of parallel resistances, because those will be accounted for implicitly when tweaking the voltage dividers.  In this way, only one precision resistor is needed, namely, the final 1K ohm resistor at the very final stage.  All this is good, because with this approach I won't need a 4-wire resistance meter for precisely measuring resistances.

Of course, none of this would account for temperature drift on the resistances.  Other than doing recalibrations, I'm not sure how I would adjust for that.  Hopefully there won't be any meaningful temperature drift over the short duration of doing the measurements.

Edit2: To mitigate against temperature drift, I'll use high resistance in the voltage dividers, so that very little power will get dissipated into the resistors.

[NeverDie]

The attached circuit should do the trick.
 voltage_divider_v002.pdf (17.72 kB - downloaded 69 times.)
The first voltage divider takes the 10.00 reference voltage and creates 1 volt across a 10-turn 1K ohm potentiometer.  That voltage is then divided by 1000 two times by the following two voltage dividers.  Finally, a precision 1K resistor (0.01% tolerance) converts that voltage into picoamps.  The trim pots are 25 turns each.  The trim pots and test points, together with a quality DMM, are used to accurately set the voltage division at each stage by automatically accounting for the parallel resistances attached to it. 

As stated above, this should allow me to precisely dial-in anywhere from one picoamp to 1000 picoamps, in single picoamp increments, with each "click" of the 10-turn 1K ohm potentiometer corresponding to one picoamp.  Of course, this assumes that the potentiometer is perfectly linear, which it may not be.  Before installing the 10-turn 1K potentiometer, I'll measure it with my DMM to see just how linear it is or isn't.  If it turns out not to be good enough, then I'll either order a proper high precision 1K ohm potentiometer to replace it with (0.1% linearity is readily available).  If worse comes to worst I'll simply replace the 1K potentiometer with a four decade resistor box that's accurate enough for the purpose. 

My 10-turn 1K ohm potentiometer arrives tomorrow, so I'll build it then.   

This should serve the current purpose of creating accurate very-low-noise test currents for the picoammeter and the uCurrent gold.  Thus, it is intended to replace both the buck regulator and the existing less accurate voltage divider in the lunchbox test enclosure.

It may turn out that even if the uCurrent Gold can measure picoamps its construction allows too much leakage of picoamps for it to measure picoamps accurately in the single digits.  We'll see soon enough.

[Kleinstein]

To test the pA meter / TIA it would take a test current in the 1 nA range - the value depends on the resistor used in the feedback.
Ideally one would use a large resistor to set the current. Something like 1 M sounds reasonable, as one can get reasonable stable and accurate resistors of 1 M. Also measuring is still possible at 1 M with the usual DMMs. If available a good 10 M resistor would also be an option.
There is no need to get exactly 1 nA,  so there is no need to have a trimmer or even more to set exactly 1 mV. If at all it would be about a few test point's (e.g. 1, 2, 0.5 mV), not to rely on only one case.
So it could be just a divider like 150K and 100 Ohms with an 1.5 V battery to generate some 1 mV, that are measured with a DMM. A 1 M resistor than sets the current.  One may have to take into account the meter resistance (e.g. 10 K shunt with the µCurrent, or isolation resistance at the TIA input, e.g. some 10-100K).
I would do the test with both polarities and zero voltage, to account for a possible offset voltage (e.g. at the DUT).

Depending on the available DMM, one could use accurate resistors for the divider and measure the 1.5 V level.

For pA level currents, one usually does no care so much about the absolute accuracy down to the 0.1% level. Leakage current often change quite a lot with temperature and maybe humidity, so even +-5 % is usually OK.

Using much smaller resistance makes the DUT resistance more important and also is sensitive to offset voltages from the measuring device. It also gets more noisy.

[NeverDie]

Thanks.  You raise a good point about needing to take into account the shunt resistance of the uCurrent Gold.

So, I guess my original test circuit was closer to what's desirable.  If I use that together with a 1.5v AA battery that's divided to 1v, together with a way to divide that 1v by 1000, so as to have any value from 0.001v up to 1v, then that would be good enough.

I'll drop one stage from the circuit diagram so that the final resistance drop is 1M 1% tolerance instead of 1K 0.01% tolerance, and I'll replace the 10v precision voltage with a 1.5v AA battery.  If I'm understanding you correctly, that should put us on the same page.  I'll post the revision today.

[NeverDie]

How about this?
 voltage_divider_v003.pdf (10.71 kB - downloaded 84 times.)

The first stage simply creates a 1v source, divisible into 1000 lesser voltages by a 10 turn 1k potentiometer.  The second stage divides that voltage by 1000 and from that voltage emits a current from a 1 megaohm resistor as the output.

The benefits over the circuit I previously used in the testing are: 1. using a 1.5v AA instead of a potentially noisy buck converter, and 2. a bit more precision on the voltage dividing, and 3. ability to dial-in anywhere from 1pa to 1na in 1pa steps using the 10-turn 1k potentiometer in the first stage.

Or, shall I ditch all the trim pots as well and replace them with just a few 1% tolerance fixed resistors?  Now that we're relying on the 1% tolerance of the 1 Megaohm resistor, maybe 1% everywhere else will be good enough as well.

[Kleinstein]

There is no real need for all the trimmers. If at all it would be 1 trimmer to get the 1 V point right.
The divider could be a single string of some thing like  50 K (possibly adjustable)  - 100 K and 100 Ohms.
If possible one could consider more than 1 mV, so maybe 10 mV and 10 M or 5 mV and 5 M.

In addition to the divider I would include a switch or two, so one could change the polarity (e.g. at the battery) and also have an off state.

[NeverDie]

I'm still not understanding as to why having the option to change polarity is a good idea.  I'm sure you're right, but I just don't understand why.  Under what circumstances would it be needed?

Edit: it would be easy enough to do 10mv increments with a 10M resistor by using the 10v voltage source if that's your preference.  I guess you're concerned about the 10k shunt resistor on the uCurrent Gold?