EEVblog #479 - Opamp Input Bias Current

[EEVblog]

Fundamentals Friday
All about opamp input bias currents.
Dave goes through the theory, and then does some practical measurements and tweaking.

[AlfBaz]

Great vid Dave, I think what you have highlighted here for me is the difficulty in knowing which of these parameters causes unexpected output errors. Not to mention the many other things that can go wrong especially when working in the device extremes, such as precision, HF, small signal etc

[EEVblog]

Great vid Dave, I think what you have highlighted hear for me is the difficulty in knowing which of these parameters causes unexpected output errors.

Yes, that's one of the major things I was trying to convey.

[ChadSeibert]

Great vid Dave, I think what you have highlighted hear for me is the difficulty in knowing which of these parameters causes unexpected output errors.

Yes, that's one of the major things I was trying to convey.

AD's opamp handbook has been demonstrating that quite handily as well . More info about bias current compensation can be found in chapter 1, page 38, for those interested.

[Rufus]

We said in the previous thread the bias currents are not significant enough for the output error you were getting.

You can consider the bias current to flow in 1k//100k and multiply that by the circuit gain. You can roughly consider the bias current to flow through the 100k to get the output offset but you can't say it flows through the 100k and then multiply by the circuit gain.

[kkp]

I'm with Rufus (and Kirchhoff ) on this one. But:

What flux are you using? If you used the water washable kind, either as pen or in the solder, or have been using the water washable kind a lot with the sponge on the soldering station, it could explain the 3nA flowing in the 100k. It is quite conductive.
(hey, that's a good opportunity to revisit guard traces.)

[colotron]

Nice video. One more test to do:
Put a mid-rail voltage on both inputs, before the 1K resistors. With mid-rail a refer to (positive supply - negative supply) / 2. This way we will have a 0V differential input with mid-rail common mode. Does it improve the input bias current? (I don't know, but I will like to see it!!!).
Being a rail to rail opamp this mid-rail voltage is used when we have to measure negative voltages.
If the last works, the input bias problem will still be there for close-to-rail voltages on non-inverting input.

[ftransform]

Are these hardware based solutions to op-amp offsets actually recommended though? Is it not better to eliminate the offsets through software, i.e. hook up a voltage reference to the input, measure the output in steps, then form a polynomial equation to deal with it? I swear I have read something along those lines in an app note but I can't find it.

I believe the app note pretty much said that on the longer term scale trimmed op-amp parameters would end up drifting more then software, like the strain/tempco in the resistors will lead to a summed error with the op-amps drift vs just the op amps drift, and that it is just a messier and more expensive solution over all.

[EEVblog]

Nice video. One more test to do:
Put a mid-rail voltage on both inputs, before the 1K resistors. With mid-rail a refer to (positive supply - negative supply) / 2. This way we will have a 0V differential input with mid-rail common mode.

I was doing that using a split supply.

[EEVblog]

Are these hardware based solutions to op-amp offsets actually recommended though?

Yes, but depends entirely on your system requirements. Either solution can work (or both at once).

[ftransform]

Are these hardware based solutions to op-amp offsets actually recommended though?

Yes, but depends entirely on your system requirements. Either solution can work (or both at once).

What is the advantage of using both? I figure it would be worse because there are more parts that can drift, so its better to use one or the other.

[EEVblog]

What is the advantage of using both? I figure it would be worse because there are more parts that can drift, so its better to use one or the other.

It won't drift if you do it right. Low ppm resistors are cheap, and either way you have to use low ppm resistors anyway to prevent any drift.
Doing both might for example allow "good enough" performance from the factory using default cal values, that could potentially be software tweaked better if really required.

[free_electron]

One thing to stress is that there are opamps that actually DELIVER current.
Dave mentioned that in the first minute but I find this important enough to re-state this here.

So , if you connect a large resistor, say 10 megaohm to such an opamp to ground... you can see a few 100 mV dc across it . have any kind of gain in such a system and it goes banana's...  this is a frequent problem when making systems that need to interface with high impedance sensors. the sensor resistance itself will muck things up.

JFet opamps are especially dangerous. The gate of the JFET is essentially a diode in reverse. pull that pin low and you will pull current out of the gate as the gate diode now conducts !

That is why JFET opamps like TL071 can be 'annoying' to work with. The current source sits on top and the gates of the jfets come out. you can actually pull the entire current bias current out of those pins... if you do that the opamp starts working. As you deviate the current you effectively limit the input swing rate of the opamp.

[free_electron]

in the AD8628 circuit the typical trick is to use a trimpot with its wiper to ground.

you would connect a 2k trimpot with its outer ends connected to _ input and + input and the wiper to ground. this let's you balance the input currents.

Code: [Select]


           ----------|+\
          |          |  >---+-----
          |        --|-/    |
          |       |         |
          |       +-[Rfb]---    
          |       |
           -[2 k]-
             /|\    
              |
            _|_
             GND

         

this lets you trim the effect of the input bias away ( on classical opamps )

if you don't want to use a multiturn trimpot you can take a 200 ohm trimpot and put a 900 ohm resistor left and right.

[Galaxyrise]

Excellent! I believe this video answers my first question in https://www.eevblog.com/forum/projects/keithley-2000-gain-repair/

Thank you

[Rufus]

I'm with Rufus (and Kirchhoff ) on this one. But:

What flux are you using? If you used the water washable kind, either as pen or in the solder, or have been using the water washable kind a lot with the sponge on the soldering station, it could explain the 3nA flowing in the 100k. It is quite conductive.

In the previous video the output offset changed with a change in the decoupling arrangement of the chip power supply. That indicates at least some of what is going on is not related to dc currents and voltages.

[gerrysweeney]

Great explanation Dave, its funny I ran into exactly this problem in the PSU I am designing just a few weeks ago and posted in my vblog. I did not manage to articulate the problem as well as you have in your video but the exact same problem I had to resolve. It was ultimately resolvable for me by putting in that +input resistor to balance input offset current and ultimately better op-amp selection.  My less articulate explanation starts at about 14 mins into the video.

http://gerrysweeney.com/fully-programmable-modular-bench-power-supply-part-13/

Gerry

[kg4arn]

We said in the previous thread the bias currents are not significant enough for the output error you were getting.

You can consider the bias current to flow in 1k//100k and multiply that by the circuit gain. You can roughly consider the bias current to flow through the 100k to get the output offset but you can't say it flows through the 100k and then multiply by the circuit gain.

I agree with Rufus.  The bias current flows through both resistors in parallel, not mainly the higher valued feedback resistor.  This is why the compensating resistor value is Rin || Rf.  But don't take my word for it:  see The Art of Electronics page 194.  Something else is at play here.

[Dave]

I don't think your current measurement with the Keithley 480 was accurate. The input burden voltage of that meter is supposed to be below 200uV. While this figure is very low, it's still way more than those microvolts of error that the bias current caused in the circuit.

[P_Doped]

I think the current measurement was fine.  There was no indication that the ammeter was in the circuit when the output referred voltage measurements were being taken.  The input bias current would be independent of how balanced the inputs were, so the impedance imbalance introduced by the ammeter should not have affected the measurement of the current.

[The Chump]

Another agreement with rufus here. You've applied a rule of thumb and forgotten how to use it properly

Why not just calculate the current in the 100k feedback resistor due to 300uV of output voltage.
You will find that 3nA is far in excess of either input bias current or input offset current specs.

Time to be really really sure that you've built what you think you've built:
have you checked that the resistors are the correct value? (both by inspection and measurement)
have you checked the gain eg with a 100Hz sine input?
have you checked that the split rails are symmetrical? (you are only metering the total rail)
have you checked that the circuit is not oscillating?
have you checked all the other things I've not thought of?

If the error remains after you've eliminated all these other things, i'ts time to ask AD why the parts appear to be way out of spec.

Cheers

Alex

[mrflibble]

JFet opamps are especially dangerous. The gate of the JFET is essentially a diode in reverse. pull that pin low and you will pull current out of the gate as the gate diode now conducts !

Thanks! Do you happen to have more information or reading material on this? Assuming that the exact same issues can occur on TL072's...

[dtweed]

You can consider the bias current to flow in 1k//100k and multiply that by the circuit gain. You can roughly consider the bias current to flow through the 100k to get the output offset but you can't say it flows through the 100k and then multiply by the circuit gain.

Someone said this in the YouTube comments too, but so far, Dave has ignored this point.

You are, of course, quite correct.

[free_electron]

You can consider the bias current to flow in 1k//100k and multiply that by the circuit gain. You can roughly consider the bias current to flow through the 100k to get the output offset but you can't say it flows through the 100k and then multiply by the circuit gain.

Someone said this in the YouTube comments too, but so far, Dave has ignored this point.

You are, of course, quite correct.

but most of it flows through the lowest value ... so that portion does get amplified.

[dtweed]

but most of it flows through the lowest value ... so that portion does get amplified.

No, the "lowest value" has no voltage across it. All of the voltage is across the feedback resistor, and the input bias current flows through it.