EEVblog #1325 - OPAMP Shootout

[EEVblog]

Part 2 in designing a better uCurrent.
A datasheet shootout between the Maxim MAX4239 and the TI OPA189
A deeper dive into the datasheet specs from the previous video, looking at gain bandwidth product, noise, supply current and voltages, offset voltages histograms, input bias currents, recovery and settling time, and slew rates.

[graybeard]

The input current of CMOS op-amps is primarily due the reverse leakage of the input ESD protection diodes, not gate current.  The opa189/2189 operates at a much higher voltage, thus higher reverse voltage on the diodes than the Maxim part.   That is the source of the higher input current.

I have used the OPA2189  and OPA189 in several designs.  I have done some designs that require the low noise density and high loop gain in the 0.1Hz-100KHz region.  The high gain-bandwidth product of this part is excellent for that.

The first time I used it was for an visible light oscilloscope probe.  In this case the wide bandwidth was both a curse and a blessing.  It allowed the bandwidth I wanted, but it also amplified the switching artifacts of the chopper.  See the video below for details.



Chris

[razvan784]

No, the input current of chopper amps is mainly due to charge injection from the input switches that do the chopping, at least at room temperature. The OPA189 switches at around 400 kHz IIRC, giving it quite a high input current, at least compared to the "classic" old choppers like the LTC1050.

[graybeard]

I did not think about the sampling current, but 10s to 100s of pA is typical of ESD diode leakage currents.  I suspect if you drop the temperature the sampling current will dominate, but the values on the spec sheet are close to what I would expect from ESD diode leakage.

Now I am curious, but I am away from my home lab until next month.  I have some opa189s and opa2189s sitting about 2 feet from my HP4145.  I should measure the diodes and measure the input current vs. common mode input voltage.

The sampling frequency I measured was close to 300KHz.  See slides 16 & 17 of this: http://chrisgrossman.com/yt/0008/0008_Visible_Light_Oscilloscope_Probe_2019-11-29.pdf

[Phaedo]

Hi All,
Bit of a newbie question, I guess.
Why is the bandwidth critical for the uCurrent ? I mean if it is used with a multimeter which samples at about 10sps , would the bandwidth be crucial ? Also even if used for measuring the sleep wake current in a uC , I guess the uC would atleast run for about a mSec or so before going to sleep mode again.
Dave could you do a similar video on ADCs in the future ? Would be really interested in knowing how to use an ADC like the MCP3201 in the single ended mode and differential mode and the difference in noise levels for both.
Cheers and thanks very much

[EEVblog]

Bit of a newbie question, I guess.
Why is the bandwidth critical for the uCurrent ? I mean if it is used with a multimeter which samples at about 10sps , would the bandwidth be crucial ? Also even if used for measuring the sleep wake current in a uC , I guess the uC would atleast run for about a mSec or so before going to sleep mode again.

You want to be able to see the current pulse on a scope.

[Kleinstein]

I don't think the BW is really that critical. One can improve the circuit by using a different 2 nd OP. If used inside the loop of the first OP, the 2nd one does not have to be low drift, which makes it easier to get high speed at reduced supply current.

For use with the scope the noise is an important parameter. For the max4239 it is worth looking at the noise curve: it has the 30 nV/sqrt(Hz) at 1 kHz and higher, but something like twice that at low frequencies  - thus the relatively large noise for 0.1-10 Hz. This is because the max part is an classical, sampling auto zero OP, whereas the OPA189 is a chopper stabilized one, with low noise for the low frequency end up to some 10 or 50 kHz, but some extra noise at a few 100 kHz.

[David Hess]

This is because the max part is an classical, sampling auto zero OP, whereas the OPA189 is a chopper stabilized one, with low noise for the low frequency end up to some 10 or 50 kHz, but some extra noise at a few 100 kHz.

I am not sure it is possible to distinguish the various parts by name anymore.  What Analog Devices refers to as an automatic zero amplifier is what Linear Technology called a chopper stabilized amplifier, and the diagram for the OPA189 sure looks like what we used to call a chopper amplifier, but extended to a differential signal path.

[ebastler]

I do like the "Floating Dave Head", by the way. It works well -- keeps you in the frame, lets the viewer follow easily which part of the datasheet you are currently looking at and talking about, without taking up too much screen real estate.

When the head is looking down from the top, am I the only one reminded of this imagery though? 

[Kleinstein]

This is because the max part is an classical, sampling auto zero OP, whereas the OPA189 is a chopper stabilized one, with low noise for the low frequency end up to some 10 or 50 kHz, but some extra noise at a few 100 kHz.

I am not sure it is possible to distinguish the various parts by name anymore.  What Analog Devices refers to as an automatic zero amplifier is what Linear Technology called a chopper stabilized amplifier, and the diagram for the OPA189 sure looks like what we used to call a chopper amplifier, but extended to a differential signal path.

There are 2 separate types: the classical automatic zero uses sampling the zero. This gives additional low frequency noise. The chopper-stabilized form uses the switches to invert the polarity and a filter, so that internal errors give a triangle like ripple. The result is extra noise at higher frequencies instead of lower. Practical the switching frequency is also different: rather slow (e.g. 1-10 kHz) for the sampling version and fast (e.g. > 50 kHz) for the choppers. Both types have a characteristic noise-spectrum, so one can usually tell from there. The tendency is to have lowest noise as a chopper and lowest bias as sampling auto zero.

There are a few modern ones that use a combination with both sampling AZ and chopping - so the borders do blur a little.
The naming is some times not so unique: there is some mix up in the terms, especially when an OP with zero drift is meant no caring about the internal.

The 2 OPs compared in the video are quite different: the max4239 is at the very low bias end and the OPA189 is one of the very low noise ones and as a consequence with considerably higher bias.

[David Hess]

This is because the max part is an classical, sampling auto zero OP, whereas the OPA189 is a chopper stabilized one, with low noise for the low frequency end up to some 10 or 50 kHz, but some extra noise at a few 100 kHz.

I am not sure it is possible to distinguish the various parts by name anymore.  What Analog Devices refers to as an automatic zero amplifier is what Linear Technology called a chopper stabilized amplifier, and the diagram for the OPA189 sure looks like what we used to call a chopper amplifier, but extended to a differential signal path.

There are 2 separate types: the classical automatic zero uses sampling the zero. This gives additional low frequency noise. The chopper-stabilized form uses the switches to invert the polarity and a filter, so that internal errors give a triangle like ripple.

And that illustrates the problem; you just described the reverse of how Linear Technology named them.

There is at least a third integrated topology which preceded the integrated chopper stabilized amplifier which only worked in the inverting configuration, but that may be because it only corrected the DC error at a common mode input voltage of zero so it was the copper-stabilized type (LT terminology) without LT's refinement.  Early integrating analog-to-digital converters had the same problem in their automatic zero loop resulting in poor linearity.

The result is extra noise at higher frequencies instead of lower. Practical the switching frequency is also different: rather slow (e.g. 1-10 kHz) for the sampling version and fast (e.g. > 50 kHz) for the choppers. Both types have a characteristic noise-spectrum, so one can usually tell from there. The tendency is to have lowest noise as a chopper and lowest bias as sampling auto zero.

That does seem to be the only way to tell the difference now, at least where the noise spectral density graph is provided.

I notice that both types have decreased in noise, but what makes the zero sampling type noisier?

The 2 OPs compared in the video are quite different: the max4239 is at the very low bias end and the OPA189 is one of the very low noise ones and as a consequence with considerably higher bias.

There used to be some other zero offset amplifier typologies but I do not know if they are still produced, and I do not think they corrected flicker noise and probably not common mode or power supply error.  One measured the input offset at startup and corrected it with a DAC so had no charge injection maladies.

[SilverSolder]

I don't think the BW is really that critical.   [...]

I think that is true.  Personally, I have never seen a problem with the current uCurrent's 250KHz bandwidth and I am not aware of anyone that has complained about its bandwidth not being large enough?

Noise is a different matter - that could usefully be lower.  If the OPA189 is making noise at around 300KHz, it might even make sense to intentionally have the gain fall from around 250KHz, just like the current uCurrent, to suppress the chopping noise.

A new uCurrent with far bigger dynamic range (due to higher supply voltages) and much lower noise at lower frequencies (for measuring very low sleep currents etc.) would be winner, winner, chicken dinner with a bandwidth of 250KHz, in my humble opinion.

[bsfeechannel]

As for the floating head, Dave has a predecessor.

[David Hess]

As for the floating head, Dave has a predecessor.

Ha, that occurred to me.  Now we need Dave's floating head to spit out electronic parts.

[EEVblog]

I think that is true.  Personally, I have never seen a problem with the current uCurrent's 250KHz bandwidth and I am not aware of anyone that has complained about its bandwidth not being large enough?
Noise is a different matter - that could usefully be lower.  If the OPA189 is making noise at around 300KHz, it might even make sense to intentionally have the gain fall from around 250KHz, just like the current uCurrent, to suppress the chopping noise.

I am considering that.
Otherwise, 1MHz bandwidth seems doable. Perhaps even a selectable filter?

[SilverSolder]

I think that is true.  Personally, I have never seen a problem with the current uCurrent's 250KHz bandwidth and I am not aware of anyone that has complained about its bandwidth not being large enough?
Noise is a different matter - that could usefully be lower.  If the OPA189 is making noise at around 300KHz, it might even make sense to intentionally have the gain fall from around 250KHz, just like the current uCurrent, to suppress the chopping noise.

I am considering that.
Otherwise, 1MHz bandwidth seems doable. Perhaps even a selectable filter?

Perhaps some jumpers inside the device could set some options like the bandwidth, to keep things simple?

I'd love to see a x100/x1000 jumper to take advantage of the lower noise to measure tiny currents...  (x1000 would be at 100KHz bandwidth, presumably, which is good enough for sleep currents).

[Kleinstein]

I notice that both types have decreased in noise, but what makes the zero sampling type noisier?

.....
There used to be some other zero offset amplifier typologies but I do not know if they are still produced, and I do not think they corrected flicker noise and probably not common mode or power supply error.  One measured the input offset at startup and corrected it with a DAC so had no charge injection maladies.

The sampling type AZ OPs get extra noise from aliasing. The fast sampling brings higher frequency noise back down to the low frequency range. The relatively low switching frequency makes 1/f noise more of a problem.

AFAIK the OPs with digital correction are still made, but they are not true zero drift, as this does not correct temperature drift and aging while working. There are Ti e-trim ones that supposedly do the offset trim once at the factory. There are some Microchip ones (MCP621 and related) that do it at start up or external signal. There also is patent from Phillips / NXP on using the digital correction inside a chopper to reduce ripple.

There is also a topology (commutating auto zero) with more or less 2 full amplifiers switching between error correction (sampling) and working. There was an early Intersil chip, that did not work well and was replaced by the ICL7650 (still in production).

I see limited use of higher amplification. Even the OPA189's noise is still significant after 100 fold amplification. Usually the scopes should be better than some 500 nV/sqrt(Hz).
The dynamic range is more limited by the noise and small input range. So if the dynamic range is a problem one may consider finer steps for the shunts - the 1:1000 steps eat up quite a bit for some current values at the low end. A little higher permitted drop and thus possibly higher output voltage may help here. This may actually need more supply voltage, though not all the way to 30 V.
Still keeping the gain at 100 one could probably go to some 50 mV at the shunt = 5 V at the output. This can even work with a single 9 V supply.

An intentional limit on the BW may be good to limit the noise.

[Cervisia]

The input current of CMOS op-amps is primarily due the reverse leakage of the input ESD protection diodes, not gate current.

Some additional current might come from the switches in the "MUX-friendly" input circuit; see this appnote: https://www.ti.com/lit/pdf/sbot040

[SilverSolder]

The input current of CMOS op-amps is primarily due the reverse leakage of the input ESD protection diodes, not gate current.

Some additional current might come from the switches in the "MUX-friendly" input circuit; see this appnote: https://www.ti.com/lit/pdf/sbot040

The "MUX-friendliness" might come in handy for auto-ranging?

[Kleinstein]

Auto-ranging for current still does not involve high voltages. So the diodes across the inputs would not be a problem - reasonable fast recovery from overload would be more of a thing.
The mux-friendly would come into play when switching between different voltage ranges and thus possibly steps larger than 1 V.

The AZ OPs are CMOS, but there is additional current from the switching part, especially with the OPA189. The protective diode current is more in the pA and below range - this may be a part of the input current of the max4239. I have not seen a curve for bias versus temperature for the max4239. the relativle similar LTC2050 has such a curve, showing the bias to go up with temperature only from some 60 C on - so the temperature dependent part would extrapolate to some 1 pA at 50 C. So the leakage part can be quite small.

[richnormand]

Really enjoyed this video Dave. Thanks.

[asmi]

I remember reading somewhere a while ago that dual opamp chips usually have the same offset voltage on both "sides" of the same chip because they were created in the same process. if true, I wonder if this can be taken advantage of to compensate offset (say have a second opamp "amplify" ground in inverse configuration such that amplified offset voltages would cancel out). Does anyone know anything about this, or ran any experiments to confirm/deny?

[Dr. Frank]

The classic chopper OpAmp was the ICL 7650, and its brother 7652.
It was/is manufactured by Intersil, then by Maxim, and by Linear Technology.
Both have a chopping frequency of 200 or 400Hz.

The 7650 has a low bias current of 1.5pA (typ.), but high noise of 2µV_pp (10Hz), wheras the 7652 has a higher bias of 15pA (typ.), but only 0.7V_pp noise.

bias and noise are the complementary parameters of chopper amplifiers, that is the case also for this comparison.

Please also read in the AoE the description about the HP34420A front end, which is extremely low noise, due to its huge area FET amplifier, but has many nA bias current, which has to be compensated to < 50pA by an external circuit.

Nowadays, I find it difficult to find low bias choppers (around a few pA), most have low noise instead.
Therefore I still use these old designs, or the also old LTC1052.

Frank

[David Hess]

I remember reading somewhere a while ago that dual opamp chips usually have the same offset voltage on both "sides" of the same chip because they were created in the same process. if true, I wonder if this can be taken advantage of to compensate offset (say have a second opamp "amplify" ground in inverse configuration such that amplified offset voltages would cancel out). Does anyone know anything about this, or ran any experiments to confirm/deny?

Monolithic dual and quad linear parts have matched offset because the mismatch depends on process variation.  But that does not apply to zero offset parts because most of their offset comes from thermoelectric junctions.  Even if it did match, external junctions would overwhelm it anyway which is why I do not put a lot of stock in minor improvements in offset voltage for zero offset parts.  Even the offset drift of the best linear parts is only a factor of 2 worse and primarily limited by thermoelectric junctions.

[Kleinstein]

The OPs in dual OPs don't necessary match in offset of drift. They share the same process, but the OP designs itself are made in a way that with a perfectly symmetric process there would be no offset. So the offset of the individual OPs  is already from things like gradients or local defects and also thermal effects at the time when the offset was trimmed at the factory.

The effects of thermal EMF is surprisingly small, at least with things like CMOS switch chips.

Still thermals can be important. So for high precision it helps if the power is low. So the OPA189 may have more trouble at higher supply (and thus more heat).