EEVblog #1318 - What's State-of-the-Art in µCurrent Opamps?

[SilverSolder]

Supply current for a pair of MAX4239s is 1.2 milliamps?  That can be reduced to 1/10th pretty easily and an extreme design could be 1/50th.  How much bandwidth is required when the measuring instrument is a multimeter?

The instrument is most often a 'scope, in my case at least.  So the current uCurrent current  [ ] bandwidth of about 250KHz is about right.  Could live with a little less, but it is nice to see the "shape" of the current consumption.

For the new design I'm targeting around 4-5 times that bandwidth, and lower noise. Comes at the expense of battery life though, hence the 9V vs coin cells.

Larger bandwidth and lower noise is interesting but I'd rather "spend the improvement" on a 1000x amplification option, than increased bandwidth... There are times when 100x is not quite enough to overcome the low sensitivity of the scope input, if the current being measured is extremely small.

I have never actually encountered a situation where the bandwidth of the existing uCurrent was not large enough for its intended use.  Most power rails have capacitors on them that would limit the bandwidth of the supply current to something lower than the uCurrent can already do, in any case? 

That said, if there was a way to have cake and eat, I would soon think of a use for the extra bandwidth!    Bode plotting, capacitor ESR testing, component testing, etc. etc.

[Kleinstein]

[...]
With the low burden voltage the dynamic range is naturally reduced. So the µC current is not the right starting point to measure the highly variable current of a µC. This would be another piece of equipment, e.g. using voltage regulation with a FET and measure the current at the other side of the FET.

I am struggling to understand.  Isn't the current (and therefore also the dynamic range) the same on both sides of the FET, so the problem of measuring it is the same?  (I do see that you could use a larger shunt resistor in front of the FET,  but I don't see a change in dynamic range of the current)

The other side of the FET could use a larger shunt. It is the low voltage drop that limits the dynamic range.

For low currents a trans-impedance amplifier is a real option. One can get a low burden and large dynamic range. It also makes the protection easier.

[SilverSolder]

[...]
With the low burden voltage the dynamic range is naturally reduced. So the µC current is not the right starting point to measure the highly variable current of a µC. This would be another piece of equipment, e.g. using voltage regulation with a FET and measure the current at the other side of the FET.

I am struggling to understand.  Isn't the current (and therefore also the dynamic range) the same on both sides of the FET, so the problem of measuring it is the same?  (I do see that you could use a larger shunt resistor in front of the FET,  but I don't see a change in dynamic range of the current)

The other side of the FET could use a larger shunt. It is the low voltage drop that limits the dynamic range.

For low currents a trans-impedance amplifier is a real option. One can get a low burden and large dynamic range. It also makes the protection easier.

Ah, the dynamic range of the voltage across the shunt is of course higher with the larger shunt resistor....   dooh, clearly I did not have enough coffee! 

The concept of the uCurrent is to use an amplifier to make up for the fact that you are using a smaller shunt.  As long as that doesn't become too noisy, we are in good shape?  It makes the uCurrent more versatile that it is not part of a larger device or power supply.

I like the idea of a trans impedance amplifier, but now the properties of that amplifier becomes part of the circuit that it is connected to - so we may run into bandwidth issues etc?  - with a shunt, the circuit still works even if the amplifier reading the shunt can't keep up, shifts phase, or whatever...

[Kleinstein]

The TIA would need a resistor at the input to isolate it from a capacitive input. The other point is to have a suitable parallel capacitor in the feedback to limit the BW to a little less than what the OP could provide. This should result in a reasonable well behaved impedance.

The TIA idea is good for the smaller currents (e.g. < 100 µA), so that the current from the OP does not have to be so large and one can tolerate some 100-1000 Ohms at the input for isolation. The shunt + amplifier solution is limited to small currents, as the shunt will have noise of it's own. So instead of a 10 K shunt, I would definitely prefer a TIA (e.g. with some 1 M in the feedback).  The shunt + amplifier solution is limited by noise and the DC offset, especially with a small burden and using 1:1000 steps. With a maximum output of some +-2 V of the µCurrent one would have 20 mV max at the shunt and thus only some 20 µV at the shunt just before the step to the next larger shunt would be possible - this is really small with not much resolution left.

Even with a low drop, I am not so sure that automatic range switching is such a good idea. The change in shunt resistance will have an effect on the circuit and sometimes one would still need a manual mode to get the right range before an expected jump to a higher current.

To get more dynamic range for the output with a limited supply, one can use an active output a little like a bridge driven amplifier instead of a simple fixed virtual ground. So if the output is positive the other output terminal and input side virtual ground could be closer to the negative side. This could nearly double the range, though with a slight limit for fast transients. There is no need to switch manually, it can be done with a not so critical inverter circuit.

[David Hess]

For the new design I'm targeting around 4-5 times that bandwidth, and lower noise. Comes at the expense of battery life though, hence the 9V vs coin cells.

I wonder at what point a chopper stabilized low noise bipolar amplifier has a lower noise for the same power draw than only a chopper stabilized amplifier.  Power draw was never a consideration when I designed them.

My intuition is that the flat low frequency noise of a chopper stabilized amplifier means that a micropower part could be used, and since bipolar parts have lower broadband noise for the same power, the combined noise would always be better.  Plus only a single chopper stabilized part would be required to correct the first stage.

The only caveat is that for bandwidth comparable to CMOS parts which benefit from lower transconductance, decompensated bipolar operational amplifiers would be required which somewhat limits parts selection and even so is a trade off unless external compensation is supported which is practically unknown now.

So the ultimate low power low noise wide bandwidth design would use a discrete bipolar input stage, allowing for maximum decompensation, with low Rb transistors (1) in parallel with a micropower chopper stabilized amplifier.  This would also have the benefit of allowing the input voltage noise and input current noise to be adjusted for lowest total noise given the source resistance of the input divider.  Conveniently this also does away with the need for the most expensive high performance operational amplifiers.

(1) Expensive IC based transistors would be best here like the MAT series but cheap low noise audio transistors like the BC327/BC337 would also perform well.  The various old Zetex "Super E Line" transistors currently made by Diodes, Inc. are likely even better but not characterized for this sort of application.  Fairchild (bought by On Semiconductor) also makes these types of parts now.

[SilverSolder]

The TIA would need a resistor at the input to isolate it from a capacitive input. The other point is to have a suitable parallel capacitor in the feedback to limit the BW to a little less than what the OP could provide. This should result in a reasonable well behaved impedance.

The TIA idea is good for the smaller currents (e.g. < 100 µA), so that the current from the OP does not have to be so large and one can tolerate some 100-1000 Ohms at the input for isolation. The shunt + amplifier solution is limited to small currents, as the shunt will have noise of it's own. So instead of a 10 K shunt, I would definitely prefer a TIA (e.g. with some 1 M in the feedback).  The shunt + amplifier solution is limited by noise and the DC offset, especially with a small burden and using 1:1000 steps. With a maximum output of some +-2 V of the µCurrent one would have 20 mV max at the shunt and thus only some 20 µV at the shunt just before the step to the next larger shunt would be possible - this is really small with not much resolution left.

Perhaps the entire configuration of the amplifier can switch between TIA and Shunt mode, for different ranges?

Dave is talking about using +/-4.5V or even +/-9V in the new design, which would help the shunt amplifier solution generally?  Perhaps the DC power jack could accept 12VDC, now the situation is 1 decade better...   

If the amplifier could be switched between x100 and x1000, the range steps could be made smaller as well...   

So many engineering compromises/design decisions to think about! 

Even with a low drop, I am not so sure that automatic range switching is such a good idea. The change in shunt resistance will have an effect on the circuit and sometimes one would still need a manual mode to get the right range before an expected jump to a higher current.

Ideally there would not be any range switching at all, but that would require a logarithmic response...   which seems surprisingly difficult, there isn't a simple/clean solution (other than an oven, which might be an energy pig) and even if the problems are resolved, it is harder to interpret the numbers afterwards...

Really we can live with selecting a range in advance, but somehow we have to dynamically "short circuit" the shunt if the DUT overdraws current for that range.  I have used the "old" uCurrent with a diode across it to handle shunt overload, that worked well enough that it seems credible that range switching can be made to work - as long as it works as fast as a diode across the shunt would.

To get more dynamic range for the output with a limited supply, one can use an active output a little like a bridge driven amplifier instead of a simple fixed virtual ground. So if the output is positive the other output terminal and input side virtual ground could be closer to the negative side. This could nearly double the range, though with a slight limit for fast transients. There is no need to switch manually, it can be done with a not so critical inverter circuit.

That's a great idea -  One problem is that the negative output would then float with respect to ground - sometimes, both the DUT and the measuring device (scope) are connected to the same ground.  It would work great with a floating instrument or a floating source, of course.  Perhaps it could be made a switchable x2 range extension feature, where the negative is plain ground unless range extension is engaged.  With the 9V battery it would be an amazing range!