Zero drift amplifier input bias current

[macaba]

I recently started simulating designs that involve DC amplification of high resistance low level signals but there is a lack of data and discussion regarding the input bias current behavior of zero drift amplifiers. This is a classic case of where reality would not match simulation so I wanted to do some measurements.

I had ADA4523 and OPA189 on my bench, both are not particularly ideal for the task (I probably want the other group of AZ amplifiers - higher voltage noise, lower current noise, potentially better behaved inputs) but thought it would be an interesting place to start. Both amplifiers have datasheet typical values in the 50-100pA range.

Using a Keithley 617 and the built-in voltage source, I measured the input bias current over CM voltage, with various values of input capacitance (SMT C0G). See attached images.

OPA189 is horrifying (with any value of input capacitance) and ADA4523 is better behaved (1nF and above, though 100nF & higher is impractical due to dielectric absorption). The OPA189 datasheet has a chart (Figure 8-5) that suggests it could work for high resistance inputs but I suspect this is pure theory.

My next step is probably to order some more appropriate AZ amplifier parts or design a DMM-style discrete JFET auto-zero frontend, but I'd like to get some ideas/feedback at this early stage.

[MasterT]

I've seen similar misbehavior many times, AZ getting upset by caps on any inputs or output, ussualy it's exposed as abnormal offset voltage or low freq. noise.  Cure is isolation by resistors, try to insert R in between cap & input, and inverting input is no exception

[Kleinstein]

The voltage dependence of the input bias looks really bad.


The microchip appl. notes suggest that one shoud aim for a similar capacitance for the inverting and non inverting inputs. I would extend this to a similar impedance in the 1-100 MHz range. This only makes sense if one also really takes care about the decoupling as the inverting input often gets connected to the output and not directly to ground.  A series resistor can be a good idea, but for the very low noise types like the ADA4522 or OPA189 this may also add significant noise. It would however be less of an issue with higher voltage noise / lower current noise types that are more suitable for a high impedance source.  The high impedance AZ amplifiers tend to be low supply voltage (e.g. 6 V max) only. So in this case the setup would be with a bootstrapped supply for the AZ amplifer. So at least the voltage dependence is not a problem. The capacitance dependence gets a low less with series resistance. I have rel. good experiance with AD8628 and MCP6V66 / MCP6V76.


The Keithley 617 may also have additional input capacitance and may react to higher requency pulses.

[iMo]

Now, many are using the 4522/23 and OPA189 in classic wiring with 7V->10V amps, with capacitance at the input against ground.
How does this issue affects the performance actually? How to improve it?
Below the typical wiring I've been talking about..

[David Hess]

Input current noise versus frequency tends to be poorly documented in chopper stabilized amplifiers, perhaps because it makes them all look bad or because they were never intended for high impedance applications.  I am not sure about newer lower noise devices, but in the past the rules for getting good results included equalizing the DC and AC impedance at each input, which applies to non-chopper parts where the input current noise is correlated between the inputs, and operating with low AC impedances at the inputs.

That last rule can be found in low noise two path amplifier designs where the chopper amplifier inputs have a 1 megohm DC input resistance, but say 0.1 or 1 microfarads of capacitance to ground for the non-inverting input and same same between the inverting input and output.

There are newer parts which claim to have solved all of these problems, but I have not tested them yet to see how they perform with high impedance sources.  I am working toward a test setup for characterizing low frequency noise.  There are now some outstanding precision JFET parts from TI, so they would have to perform very well indeed to be worth using.

[iMo]

Well, based on the above to cure the issue with for example OPA189 we have to wire it as follows:
the 3k3=10k||5k and with the C6/C7 1n caps it equalizes the impedancies both inputs see.
Is that correct?

[Kleinstein]

With AZ amplifiers there is no need to balance the DC impedance. The DC bias is anti correlated and thus no compensation and a lower resistance is preferred. A balance would only help for high temperature (e.g. > 60-100 C) with more conventional correlated leakage current.
With conventional OP-amps it also depends if an added series resistor to get balance is worth it. They still add noise and with internal compensation of the bias the correlation is no longer that good.
This was mainly a point with old BJT based ones (e.g. NE5534, 741).

For the higher frequency part I see the slight problem that the output sides usually has some series impedance. So it is not just balancing the capacitor at the non-inverting input and the feedback.
It may help to additionally load the ouput with an RC snubber (e.g. 50-100 ohm and 1-10 nF). To some degree one may be able to use the capacitance at the input to trim the input current. So more like individual try and error.

There are a few examples where they show the frequency dependence of the current noise: What I found so far they showed a flat curve towards low frequencies and the expected extra near the chopper frequency range.
The current noise specs in general are poor and quite some numbers look too good to be true (e.g. a lower bound from the input bias current).

[dietert1]

Some years ago somebody explained about the spikes from the opamp inputs generated by the chopper. Since then i started to place a capacitor to Gnd also on the inverting input. Of course one can only put 500 pF or 1 nF there without instability, but that seems to work well. For similar reasons i think these capacitors are important in a circuit having digital signals. I never had the time to quantify the advantage of the measure, though.

In the 7V => 10 V schematic of imo the OPA189 is likely stable with an additional 1 nF capacitor from the inverting input to Gnd ("C6 prime").

Regards, Dieter

[Kleinstein]

A capacitor to ground for the inverting input can work as part of a compensated divider. For the 7 to 10 V step there is not much gain and thus quite some capacitive loading. So 1 nF is likely too much - maybe 50-100 pF could work. With a higher gain the capacitance should be OK.

[dietert1]

If one puts a 1 nF cap C6 prime onto the inverting input, it forms a capacitive divider 0.5 with C7, while the resistors make a 0.7 divider. Not exactly compensated, but roughly. Capacitive output load will be 500 pF and the OPA189 does that.

The point was the balance of these caps should be more important than the 3.3 K resistor R10. C6 could increase to 2 nF.

Regards, Dieter

[macaba]

Cure is isolation by resistors, try to insert R in between cap & input, and inverting input is no exception

Thanks, this was a good clue.

The voltage dependence of the input bias looks really bad.

I agree, when I saw this, I also thought the same as you - there is still value in bootstrapping an AZ amplifier to make the input bias compensation (current injection through high value resistor) easier. Do you have links to the microchip application notes?

How does this issue affects the performance actually? How to improve it?

You have to build the circuit and measure it. No amount of simulation or discussion will help here. (this is coming from someone who really likes simulation!)

I am working toward a test setup for characterizing low frequency noise.  There are now some outstanding precision JFET parts from TI, so they would have to perform very well indeed to be worth using.

I eagerly await your results. Do you mean OPA140 and friends? (or JFE2140?)

Some years ago somebody explained about the spikes from the opamp inputs generated by the chopper. Since then i started to place a capacitor to Gnd also on the inverting input.

This is the same thing I had in mind, and I've seen it on datasheets too. However, as the following will show... this does *not* apply to OPA189.

Continuation of experimentation

With all those comments in mind, in an attempt to tame the OPA189, I explored various topologies involving R and C and C on the inverting input too. All of them failed except one - simple resistive-only isolation of both OPA189 inputs. It hates any kind of additional capacitance on the inputs (which defies the 'taming charge injection spikes' hint we've all seen before).

See attached image.

I think the only reliable rules so far are:
- You have to measure reality. Discussion or simulation no good here.
- Once you've picked & measured an AZ amplifier for your design, you cannot substitute it for another part unless you perform all the same testing again.

[dietert1]

Thanks for the reminder to check things. I remember doing something similar with the Advantest R6581T input. The picoamperemeter was a HP 3478A with a 1 MOhm "shunt".

Regards, Dieter

[tszaboo]

Input current noise versus frequency tends to be poorly documented in chopper stabilized amplifiers, perhaps because it makes them all look bad or because they were never intended for high impedance applications.  I am not sure about newer lower noise devices, but in the past the rules for getting good results included equalizing the DC and AC impedance at each input, which applies to non-chopper parts where the input current noise is correlated between the inputs, and operating with low AC impedances at the inputs.

That last rule can be found in low noise two path amplifier designs where the chopper amplifier inputs have a 1 megohm DC input resistance, but say 0.1 or 1 microfarads of capacitance to ground for the non-inverting input and same same between the inverting input and output.

There are newer parts which claim to have solved all of these problems, but I have not tested them yet to see how they perform with high impedance sources.  I am working toward a test setup for characterizing low frequency noise.  There are now some outstanding precision JFET parts from TI, so they would have to perform very well indeed to be worth using.

What I've seen, the bias current was correct, but it was an average bias current, with huge swings from it when the AZ was happening. So if you had a high impedance source it was super noisy. I had to order a bunch of different Zero drift amplifiers and test them. Probably not surprising, but the Linear Technology part was like 100 times better than the equivalent TI and AD part, and you couldn't see this at all from the datasheet.
Yes, and forget simulations and theoretical calculations.
BTW I think I used the OPA189 in a recent design, maybe I'll take a look at it's noise, since i haven't investigated that.

[David Hess]

I am working toward a test setup for characterizing low frequency noise.  There are now some outstanding precision JFET parts from TI, so they would have to perform very well indeed to be worth using.

I eagerly await your results. Do you mean OPA140 and friends? (or JFE2140?)

That is right.  The OPA140 is:

Input Offset: 120 microvolts maximum
Input Offset Drift: 0.35 uV/C typ 1 uV/C max
0.1 to 10 Hz Noise: 42 nVrms
Input Bias Current: 0.5 pA typ 10 pA max
Common Mode Rejection Ratio: 120 dB min 140 dB typ

Bipolar parts can achieve an input offset drift about 10 times better, and chopper parts 2 times better again, but 1 uV/C is still as good or better than the best matched JFETs.  That low frequency noise level is about the same as an OP-07 or LT1001, and about twice as good as a low input current super beta bipolar part which the OPA140 would replace for even lower input bias current at room temperature.

I am in the process of moving to New Hampshire and still setting up my home lab.  I am looking for the best way to measure noise down to 0.01 Hz.  Spot noise measurements are straightforward, but maybe I will end up with something more sophisticated.  A low frequency low noise digitizer might be ideal, so I am investigating whether a high resolution multimeter could be used that way, which is how I have made accurate spot frequency measurements in the past down to 0.1 Hz.

[David Hess]

What I've seen, the bias current was correct, but it was an average bias current, with huge swings from it when the AZ was happening. So if you had a high impedance source it was super noisy. I had to order a bunch of different Zero drift amplifiers and test them. Probably not surprising, but the Linear Technology part was like 100 times better than the equivalent TI and AD part, and you couldn't see this at all from the datasheet.
Yes, and forget simulations and theoretical calculations.
BTW I think I used the OPA189 in a recent design, maybe I'll take a look at it's noise, since i haven't investigated that.

I do not remember who published an article about it, but it might have come from Bob Pease in a discussion about how datasheets lie, or a discussion about operational amplifier noise.  Old datasheets also often confuse common mode rejection versus frequency with power supply rejection versus frequency because they used the wrong test for one of them.

[Kleinstein]

The JFET input noise may have a different spectral composition than the precision BJT parts.  Often one is indeed interrested in the frequencies lower than 0.1 Hz. The 0.1 - 10 Hz range is more like choosen to be easy to measure (AC coupling still works and 2 decades is large enough so that the exact filter shape is not that critical (still a factor to cause condusion). For the lower frequency noise a relatively fat DMM (to work well in 1 PLC mode) should be a reasonable choice. If not with a low of DC offset like a voltage reference the amplification can be quite large in a DC coupled configuration.

A point to whatch for is thermal stability: temperature fluctuation combined with a TC may look a lot like 1/f noise.

The current noise specs are not just a problem for the AZ OP-amps, but also for many BJT based OP-amps. To get similar good looking numbers the wrong way (splitting the resistor to both inputs), e.g. used with the OP27 is also used with many other OP-amps. This should give a values for the current noise that is 30% to maybe 50% too low if the current noise is not fully in phase correlated. Already the definition is a bit tricky as there are 2 (or even 3 if the inputs are different as in CF parts) parts to the current noise as an independent and correlated part.

[iMo]

For example MCHP on the MCP6V76 input glitches:

The inputs should see a resistance on the order of 10Ω
to 1 kΩ at high frequencies (i.e., above 1 MHz). This
helps minimize the impact of switching glitches, which
are very fast, on overall performance. In some cases, it
may be necessary to add resistors in series with the
inputs to achieve this improvement in performance.

TI on OPA189

Charge injection from the integrated switches on the inputs can introduce short transients in
the input bias current of the amplifier. The extremely short duration of these pulses prevents the pulses from
amplifying, however the pulses may be coupled to the output of the amplifier through the feedback network.
The most effective method to prevent transients in the input bias current from producing additional noise at the
amplifier output is to use a low-pass filter such as an RC network.

ADI on ADA4522 input glitches or charge injection - none.

[dietert1]

The interpretation of macabas plots depends on the input capacitance of the current meter. One could also try to look at the input current with a transimpedance amplifier and a scope. Certainly the image will also depend on the common mode voltage.
As far as i understand missing capacitor(s) on a chopper stabilized opamp like the OPA189 won't generate offset current but an unstable input offset voltage (extra low frequency noise). Similar to the errors a SAR ADC produces without some buffer cap on its input.
I have been using the OPA189 and ADA4522 for filtering references and in reference gain stages with good results. But of course, if there are quantitative comparisons to other parts, i am interested to see them.

Regards, Dieter

[Kleinstein]

Ti has some article / set of slides on the input current using a TIA to measure the actual current.
The charge injection of CMOS switch chips is known to slightly depend on the capacitance, especially in the range of small capacitors. It helps if both sides see a similar capacitance. So to me it is no surprise that the input current is somewhat effected by the capacitance.

[macaba]

The Keithley 617 is a TIA. I also have a Keithley 428 (a high bandwidth TIA) so there’s the possibility of seeing dynamic behaviour but I am not optimistic - AZ switching frequency is usually above 100kHz, and the Keithley 428 bandwidth depends on the current range (with most sensitive range having the least bandwidth). Worth a quick attempt though.

[iMo]

Interesting paper on on-chip glitches and ripples reduction technology used in AZ/choppers opamps written by ADI designer Mr. Kusuda.
ADA4522 is using the technique depicted in Fig.13.

[Kleinstein]

The dynamic is very fast. Ti in there paper on this used a really fast OP-amp (OPA657) in the TIA to get high speed. The current spike is on the order of 20-100 ns wide.

Attached is a copy of the Ti appl. note, that for some reason is hard to find on the website.

[MegaVolt]

Probably not surprising, but the Linear Technology part was like 100 times better than the equivalent TI and AD part, and you couldn't see this at all from the datasheet.

Can you tell what amp it was?

[dietert1]

Interesting paper on on-chip glitches and ripples reduction technology used in AZ/choppers opamps written by ADI designer Mr. Kusuda.
ADA4522 is using the technique depicted in Fig.13.

How do you know? How to find out which ones are [8] and [9] (charge injection balancing)?

Regards, Dieter

[iMo]

Do click on the link and you will see the reference.