Know of any op amps with low input **capacitance**???

[IDEngineer]

Not input **impedance** (i.e. input AC bias current), but stray input capacitance.

This is for a capacitance measuring application. We're finding that the input capacitance of the opamp is affecting measurements in a nonlinear fashion, very much like how a resistor divider is nonlinear as one of its two elements changes. We can't just "subtract out the offset" because the effect varies across the entire dynamic range of the measurements.

Input capacitance isn't a commonly specified parameter. We're going to reach out to some IC manufacturers but I wondered if anyone here has ever had to spec such an opamp before.

Thanks!

[tggzzz]

It would help if you gave a schematic and layout of the relevant parts, or at least specify the how you are using an opamp. We ,may then be able to make specific suggestions as to what is causing the non-linearity.

For example, some capacitors are horribly non-linear, and can lose 80% of their specified capacitance when operated at their rated voltage.

[Gyro]

Also which opamp you are using? There will presumably be differences between CMOS and JFET input (assuming it's not bipolar).

[nctnico]

It would help if you gave a schematic and layout of the relevant parts, or at least specify the how you are using an opamp. We ,may then be able to make specific suggestions as to what is causing the non-linearity.

For example, some capacitors are horribly non-linear, and can lose 80% of their specified capacitance when operated at their rated voltage.

I agree. We need more info on what kind of measurement method is being used.

[IDEngineer]

It would help if you gave a schematic and layout of the relevant parts, or at least specify the how you are using an opamp.

I didn't bother because it's the definition of simplicity: The opamp is in a voltage follower configuration with no passives. The input signal drives the non-inverting input, the inverting input is connected directly to the output. Obviously unity gain, and BW of a couple hundred KHz is more than sufficient (the present opamp is 5MHz GBW product).

For example, some capacitors are horribly non-linear, and can lose 80% of their specified capacitance when operated at their rated voltage.

I am well aware of the behaviors to which you're referring, and I agree for discrete caps, but this is a sensor. It's not measuring a discrete capacitor. The capacitance being measured has a dynamic range of 30-200pF, so small amounts of stray/parasitic capacitance can make a big difference. Some opamp application notes suggest that opamp input capacitance can be as high as 25pF, nearly equal to the bottom of our range. (Plug those values into the classic divider equation and you get a curve that looks exactly like what we're seeing.)

We have characterized the capacitance being measured across its dynamic range using a standalone cap meter and it is linear when measured without the opamp. With the opamp in the circuit the response goes non-linear as described. Remove the opamp and the response goes linear again. Note that the opamp has exactly one connection to the DUT - its non-inverting input - so there aren't a lot of variables here. {grin}

We considered that the opamp's input bias current might be (part of) the problem. The CMOS opamp we're presently using has input bias current of 1pA typical, 50pA max at max temp. Given such low values we do not believe the input bias current is the source of the nonlinearity. If the input current were having an effect, it would also bleed noticeable charge off the capacitance. Testing for that, the input bias current is so low that no voltage droop is visible on the charged capacitance for tens of microseconds (our measurement period) after charging has stopped.

We need an opamp that has minimal input capacitance so its single, non-inverting-input connection to the circuit doesn't add capacitance to the capacitance being measured.

[IDEngineer]

Also which opamp you are using? There will presumably be differences between CMOS and JFET input (assuming it's not bipolar).

Microchip MCP6282 dual, in SMT package, running on a single ended 5VDC supply.

[Conrad Hoffman]

If you can't find a suitable opamp, which seems unlikely, maybe a composite amp using a discrete jfet would work. Linear Systems has some low capacitance parts.

[AR]

The non linear input capacitance effect on distortion  can be reduced by lowering the source impedance driving it, so what is your source impedance ? . An ideal source impedance  is  50 - 100 ohms. You can also use the opamp in inverted mode this configuration is  reduces the non linear input capacitance effect on distortion because the input stage has very little voltage variation because its a virtual earth.

     

[Marco]

Use a JFET unity gain buffer first, bootstrapped cascode if you want to lower input capacitance still more. Though a BF862 and a resistor is probably enough.

[Gyro]

As it's being used as a voltage follower, I wonder if something with bootstrapped protection diodes might have more linear input capacitance.

[IDEngineer]

The non linear input capacitance effect on distortion  can be reduced by lowering the source impedance driving it, so what is your source impedance ? . An ideal source impedance  is  50 - 100 ohms. You can also use the opamp in inverted mode this configuration is  reduces the non linear input capacitance effect on distortion because the input stage has very little voltage variation because its a virtual earth.

Remember that, once the capacitance being measured has been charged, the opamp is being "driven" solely by the charged capacitance. To the opamp that looks like a very high impedance. In fact, the whole reason for having the voltage follower is to create a low-impedance representation of the voltage on the capacitance... if we already had a low impedance version we wouldn't need the opamp! {grin}

For the same reason, an inverting configuration on the opamp will not work. The input impedance of an inverting configuration is orders of magnitude lower than a non-inverting configuration, because the latter is just the input impedance of the non-inverting terminal by itself... there's no resistor network (Rinput and Rfeedback) to lower the impedance. Again, since the opamp is being "driven" by a rather small charged capacitance, the opamp input impedance must be kept as high as possible.

[IDEngineer]

As it's being used as a voltage follower, I wonder if something with bootstrapped protection diodes might have more linear input capacitance.

You mean an opamp with NO input protection diodes? That's a good thought. Hadn't considered the C-vs-V relationship on the input diodes. Hmmm.

[nctnico]

Also which opamp you are using? There will presumably be differences between CMOS and JFET input (assuming it's not bipolar).

Microchip MCP6282 dual, in SMT package, running on a single ended 5VDC supply.

First of all I wouldn't use Microchip in any sensitive analog circuit because they suck at analog.
Secondly I'd use a LCR meter style circuit to measure capacitances. Send a known AC current through the DUT and measure the resulting voltage across the device using a differential amplifier.

[tggzzz]

As it's being used as a voltage follower, I wonder if something with bootstrapped protection diodes might have more linear input capacitance.

You mean an opamp with NO input protection diodes? That's a good thought. Hadn't considered the C-vs-V relationship on the input diodes. Hmmm.

Not just diodes. Obviously MOSFETs have charges kept apart across the junction, and you can expect such junctions to be a little like varicaps.  Whether that is significant depends on parameters you haven't revealed.

Any such effects are irrelevant if there is no voltage change. It might be possible to arrange that by bootstrapping. ISTR there are circuits for bootstrapping entire opamps in The Art of Electronics.

[IDEngineer]

Obviously MOSFETs have charges kept apart across the junction, and you can expect such junctions to be a little like varicaps.

Good point on the input MOSFET's, thanks!

[Gyro]

As it's being used as a voltage follower, I wonder if something with bootstrapped protection diodes might have more linear input capacitance.

You mean an opamp with NO input protection diodes? That's a good thought. Hadn't considered the C-vs-V relationship on the input diodes. Hmmm.

No, some very low input current opamps have their input protection diodes driven by an internal low impedance bootstrap driver to keep their leakages (therefore the voltage across them) balanced. Examples would be the LMC6001 and (I think it uses internal bootstrap) the LMC662, which is a dual and cheap.

As tggzzz says, there are other 'junctions' involved but this might take care of the protection diode contribution. To go further would indeed involve bootstrapping the supplies of the entire opamp - which brings up another point, you haven't mentioned the bandwidth that you need. That could severely restrict the supply bootstrapping approach.

[IDEngineer]

...which brings up another point, you haven't mentioned the bandwidth that you need. That could severely restrict the supply bootstrapping approach.

Obviously unity gain, and BW of a couple hundred KHz is more than sufficient (the present opamp is 5MHz GBW product).

[nfmax]

The common-mode input capacitance of almost any OPAMP will be voltage dependent, because the reverse-biased isolation junction around the input devices acts as a varicap diode: as the voltage changes, the width of the depletion region changes, changing the capacitance.

One solution to this is to bootstrap the OPAMP supply rails so the common mode input voltage with respect to the supply rail does not change as the input voltage changes. You can use the output of the OPAMP itself, suitably level-shifted, to drive suitable fast voltage regulators (or even just emitter followers may be good enough). In theory, you only need to bootstrap the rail to which the isolation junction is connected, but it is often difficult to find out which that is, and if you do find out you still have the issue of maximum supply voltage to worry about.

Bootstrapping also effectively removes the common mode input capacitance - well a good part of it, anyway, in practice. There is still the differential input capacitance, probably dominated by the OPAMP package. Your bootstrap needs to have the full circuit bandwidth, of course.

[IDEngineer]

No, some very low input current opamps have their input protection diodes driven by an internal low impedance bootstrap driver to keep their leakages (therefore the voltage across them) balanced. Examples would be the LMC6001 and (I think it uses internal bootstrap) the LMC662, which is a dual and cheap.

I'm aware of the LMC6001, a truly remarkable component by the way. Reading the spec sheet is amazing.

Just looked at the LMC662. Very interesting. The text casually mentions that the per-pin input capacitance is on the order of 10pF. That's better than the ~25pF I've seen elsewhere. We might swap an LMC662 into the board and see if it makes a difference. Thanks!

[Cerebus]

Obvious candidates are OPAs designed for electrometer type applications:

• LMC662: unspecified C_IN but 2 fA input bias current, 22 nV/Hz^0.5 voltage noise with 2 fA/Hz^0.5 current noise is a big clue to a low value - cheap (£1).
• LMC6001: unspecified C_IN but 10 fA input bias current, 22 nV/Hz^0.5 voltage noise with 0.13 fA/Hz^0.5 current noise - expensive (£10).
• ADA4530-1: 8 pF - more expensive (£17)
• AD549: 1pF differential, 0.8 pF common mode - damned expensive (£30 - 50).

Those are the few I have datasheets lying about for, there are others.

Edited to add: Just realised that the comments might not be obvious to all. Because input capacitance can convert current noise into voltage noise the combined specification for voltage and current noise places a limit on the input capacitance, I = C ^dv/_dt and all that.

[danadak]

Some ref material -


http://e2e.ti.com/cfs-file/__key/communityserver-discussions-components-files/14/5706.boostrapping-techniques-to-improve-the-bandwidth-of-transimpedance-amp.pdf


http://iris.elf.stuba.sk/JEEEC/data/pdf/11-12_101-02.pdf


http://www.analog.com/media/en/technical-documentation/application-notes/742022599AN232.pdf



Regards, Dana.

[amspire]

Is there a reason you have to use the opamp in voltage follower mode? If you use it inverting mode with the + input connected to a fixed voltage, the opamp input capacitance is essentially nulled out.

Even a LM324/LM358 could measure 1pf at 10KHz pretty easily in summing mode without much error.

With a 1M feedback resistor, 1V 10KHz on the 1pF capacitor will give something like 63mV out of the opamp.

[IDEngineer]

Is there a reason you have to use the opamp in voltage follower mode?

For the same reason, an inverting configuration on the opamp will not work. The input impedance of an inverting configuration is orders of magnitude lower than a non-inverting configuration, because the latter is just the input impedance of the non-inverting terminal by itself... there's no resistor network (Rinput and Rfeedback) to lower the impedance. Again, since the opamp is being "driven" by a rather small charged capacitance, the opamp input impedance must be kept as high as possible.

...and the reason it needs to be kept as high as possible is because a low impedance will cause voltage droop as it pulls charge out of the capacitance. One way to think of this is like a Sample-and-Hold circuit... once the cap is charged, it must only see a high impedance or it won't "hold" very well for very long.

[amspire]

Is there a reason you have to use the opamp in voltage follower mode?

For the same reason, an inverting configuration on the opamp will not work. The input impedance of an inverting configuration is orders of magnitude lower than a non-inverting configuration, because the latter is just the input impedance of the non-inverting terminal by itself... there's no resistor network (Rinput and Rfeedback) to lower the impedance. Again, since the opamp is being "driven" by a rather small charged capacitance, the opamp input impedance must be kept as high as possible.

...and the reason it needs to be kept as high as possible is because a low impedance will cause voltage droop as it pulls charge out of the capacitance. One way to think of this is like a Sample-and-Hold circuit... once the cap is charged, it must only see a high impedance or it won't "hold" very well for very long.

You started off by saying this is for a capacitance measuring application, but for some reason, it seems you have to do it the hard way. It gets difficult when we are missing the details. Why are you charging and holding the voltage on the capacitor if you just want to measure it?

The summing mode I mentioned is beautiful in that you can even have a coaxial cable from the opamp to the capacitor and still measure 1pF accurately without any compensating for the coax cable. It is basically the way all the LCR meters work. A very cheap and accurate solution.

You are doing something that is possible but it will be very difficult. Could be expensive. As mentioned before, you may need to bootstrap the supply voltages of the opamp to remove any relative DC voltage changes on the opamp input. That way you do not need a special opamp.

[IDEngineer]

You started off by saying this is for a capacitance measuring application, but for some reason, it seems you have to do it the hard way. It gets difficult when we are missing the details. Why are you charging and holding the voltage on the capacitor if you just want to measure it?

That is a valid question. Without revealing proprietary info that I'm not at liberty to discuss, the short answer is that we did try a frequency based measurement scheme early on. That was our first choice. But it worked very poorly due to the environment in which this sensor must operate, and so we tried voltage based measurement as an alternative. It showed far more initial promise so we proceeded with that.

In the meantime, to improve the voltage based measurement we have discovered and refined certain aspects of the system to the point that *I* now believe the problems of frequency based measurement may have been ameliorated. I'd like to go back to that and see if it can be made to work, and not just for the reasons raised in this thread. But the project is so far along at this point that such a wholesale redesign isn't going to happen until at least the next major revision. Voltage based measurement is working "well enough" to proceed, we're just trying to refine it as much as possible.

Hence the first post: We believe the opamp input capacitance may be contributing to the error, and I was hoping some here may have info on this parameter which is seldom published in spec sheets. I've ordered some LMC662's which we will drop on the board in place of the existing (rather low spec) units and see if that buys us some improvement.

I do really appreciate the comments and assistance!

[Cerebus]

I've ordered some LMC662's which we will drop on the board in place of the existing (rather low spec) units and see if that buys us some improvement.

I assume you are aware, but just in case you aren't, that these are such high input impedance devices that you might have to step up your guarding and shielding techniques  to avoid problems with tiny leakage currents that wouldn't bother lower spec devices. There's a video kicking around somewhere showing the rig that Dave Grohl Paul Grohe and Bob Pease built to characterise its more expensive cousin the LMC6001 which is most instructive.

[Alex Nikitin]

For low input capacitance and less than 5V supply voltage have a look at  OPA355, OPA356 and OPA358 with 1.5pF/1.5pF input capacitance.

Cheers

Alex

[Zero999]

How about the input impedance and bias currents?

If you just want a low input capacitance, the old uA741 has an input capacitance of only 1.4pF.
http://www.ti.com/lit/ds/symlink/ua741.pdf

[xygor]

Do you suppose you could use a bridge circuit, which would cancel the nonlinear part?

[Marco]

People are getting a little too scared of discretes ...

[IDEngineer]

I assume you are aware, but just in case you aren't, that these are such high input impedance devices that you might have to step up your guarding and shielding techniques  to avoid problems with tiny leakage currents that wouldn't bother lower spec devices.

Indeed, and we already have careful guarding and shielding in the design for exactly that reason. We also treat the PCB as an active component and spend a LOT of time working with the PCB Designers to optimize the layout for this type of measuring.

Good reminder, thanks!

[T3sl4co1l]

Put a bootstrapped buffer in front of it?

Phil Hobbs did this for a microchip probe jig, and posted and discussed the circuit.  The basic skeleton is this: Probe --> JFET follower (CCS loaded) --> BJT follower --> (output) --> positive voltage offset (zener) --> BJT follower --> JFET drain.  The CCS-loaded JFET has DC gain very close to 1, the BJT has a low output impedance and fairly high gain (0.99ish), and the follower in the drain circuit bootstraps out Cgd.

Actually I think he used a PHEMT or something, not a JFET, but you might not need that.  BF862 is the usual favorite, though it's EOL now; there's an On Semi replacement out there.

Also place the output node under/around the probe node, to bootstrap the capacitance to space.

He ended up with a fairly small fractional pF input capacitance, which easily met the application requirement.

Tim

[Cerebus]

BF862 is the usual favorite, though it's EOL now; there's an On Semi replacement out there.

2N/MMBF 4117 would probably work nicely here as there's no requirement for a high f_t and it's got a tiny junction with concomitantly low C_iss (3pF versus 10pF for the BF862).

[T3sl4co1l]

'4117 is embarassingly small (kohms Rds(on)), I might be worried about output impedance / drive capability instead.  Sure, there's probably something inbetween, a few mA versus the 8-20mA of a BF862.

A 2N3819 or 2N5486 probably won't give very good noise at these low frequencies... but I suppose if an LM358 is in play, there's no care about noise here.

Which might give the OP some reason to think about how to improve the sensitivity of this circuit: if it's not in the noise floor, it's not really at the limit, is it?

Tim

[BrianHG]

No one has yet recommended to additionally add a series resistor & cap to GND on the - feedback input to create a filtered compensated output response adaptively nullifying some of the sluggish response due to the capacitive load of the + input.

[Henrik_V]

The LTC6268 spec is  .450pF ...
http://cds.linear.com/docs/en/lt-journal/LTJournal-V25N2-00-df-LTC6268-GlenBrisebois.pdf