New ultra low offset opamp champion? - OPA2387

[obuone]

It seems that there is new opamp candidate for Dave's uCurrent.

Accidently stumbled on the datasheet for OPA2387 which seems to be superior to MAX4238/9 and OPA189 in terms of offset and noise performance.

It seems OPA2387 has the same maximum 2uV offset over temperature as that of MAX4238 but the distribution seems to have smaller spread.
As well the noise density is just 8.5 nV/rootHz compared to the 30 nV/rootHz of MAX4238 under the same bias conditions.

So do we have new ultra low noise and offset Opamp champion? I can't find any metric that MAX4238 beats OPA2387 (except bandwidth 6.5 vs 5.7 MHz).

[Kleinstein]

This new OP looks good, though not the absolute lowest noise (AFAIK currently the ADA4523). The max4238/9 is still better with the input bias current and likely also with noise current. The noise current data are always to be taken with some caution, especially if they look really good.

The 30 nV/sqrt(Hz) number for the max4238 is for some 1 kHz, the lower frequency noise likely goes up to some 60 nV/sqrt(Hz) to reach the 0.1-10 Hz noise figure.
At least the DS I have seen does not tell directly, but for classical AZ OP it is normal to have the low frequency noise to go up by about a factor of 2 in the range of the chopper frequency.

The combination of voltage and current noise looks very good - in both aspects better than the LTC2057. With respect to the noise figure the relevant number is the product of voltage noise and current noise. The ratio gives the resistance for the best noise figure.
The input bias is also quite good, especailly for an OP with such a low noise level.

For the bandwidth of the µCurrent, there is a way around it, by using a compound amplifier made from 2 different OPs instead of 2 equal OP stages. In the compund stage the 2nd OP should be faster, but does not need to be zero drift. Alternative one could have a fast and low noise main amplifier and the AZ OP only as offset stabilization.

[tszaboo]

This new OP looks good, though not the absolute lowest noise (AFAIK currently the ADA4523). The max4238/9 is still better with the input bias current and likely also with noise current. The noise current data are always to be taken with some caution, especially if they look really good.

It's a Chopper opamp. I found it very difficult to compare them by datasheet. Most of the relevant parameters are not very well defined, or disclosed in the datasheet. For example the chopping frequency is about 100KHz in this one, and the noise specification doesn't cover this range. Those pA bias also looks very good, but then there is charge injection by the chopper to the input. It doesn't show in DC, but there is a lot of current going in and out at higher frequency, that regular opamp doesnt do. The pA bias wouldn't even make sense without dedicated guard trace pins, if we think about it for a second.

When I had to use them, the only way to evaluate them was to solder them into the circuit and measure, and try to look for issues that you dont even expect.
Not really an issue for experienced designer, but it is not as easy to use as a laser trimmed opamp or a regular opamp.

Overall, this seems to be an improvement over the OPA335.
The OPA2182 is even more interesting. I wonder how it stacks up against the LT2057.

[David Hess]

I wish they would include current noise versus frequency graphs.  I know that in some cases they do not because it would reveal that the operational amplifier must be used with low AC impedance source and feedback.

[Kleinstein]

It would be nice if they tell how the current noise was actually measurend. Charge injection depends on the impedance at the inputs. So the measured current noise may also depend on the exact conditions / impedance at the inputs. For a complete picture the current noise has a common mode and a differential component - so it would ideally need 2 numbers.

For the current noise versus frequency, I would not expect much more information from that. The few cases where they actually show such a curve, it's pretty flat in the low frequency range. Expect large peaks at the chopper frequency. The LTC2057 and LTC2058 are some rare examples that show such curves.

I would consider the OPA2387 a slightly improved version of the OPAx388. Expecially the current noise and offset got quite a bit better.

[David Hess]

It would be nice if they tell how the current noise was actually measurend. Charge injection depends on the impedance at the inputs. So the measured current noise may also depend on the exact conditions / impedance at the inputs. For a complete picture the current noise has a common mode and a differential component - so it would ideally need 2 numbers.

That problem comes up with operational amplifiers that use input bias current cancellation if the noise of the compensation currents is correlated.  Linear Technology was kind enough to discuss it and include a graph showing noise for matched and unmatched source impedance versus frequency for the affected parts.

[obuone]

It would be nice if they tell how the current noise was actually measurend. Charge injection depends on the impedance at the inputs. So the measured current noise may also depend on the exact conditions / impedance at the inputs. For a complete picture the current noise has a common mode and a differential component - so it would ideally need 2 numbers.

For the current noise versus frequency, I would not expect much more information from that. The few cases where they actually show such a curve, it's pretty flat in the low frequency range. Expect large peaks at the chopper frequency. The LTC2057 and LTC2058 are some rare examples that show such curves.

I would consider the OPA2387 a slightly improved version of the OPAx388. Expecially the current noise and offset got quite a bit better.

The noise at the chopping frequency is not as flattering as the banner specs. Sneaky marketing.