OPA2156: HiFi in CMOS

The OPA2156 is a new, high-performance op-amp from TI. Normally, that's not newsworthy because higher performing parts are coming out all the time. But this time it's a bit more interesting. The specs on the opamp are really good, almost as good as the OPA1612 (and better in some ways that will be covered below). 

But more importantly, it's CMOS. 

To date, bipolar opamps have ruled when performance is critical. JFET can be used to help with bipolar input shortcomings. But largely, designers have needed to suffer with input limitations on bipolar if they wanted to enjoy the noise and distortion benefits afforded by the technology.

CMOS in general has a lot of benefits--more acres of CMOS are probably fabricated each year than any other process. Aside from its scale and maturity, one of the biggest benefits is that once you have a design built for CMOS, it can be integrated with larger blocks of logic. That is why sigma delta data converters have been so important in the evolution of cell phones. It used to be that the analog processing (voice ADC and earpiece DAC) was captured and played back on a separate codec inside the phone, and that codec was overwhelmingly analog and realized in a mixed process (often bicmos) that was expensive. But once sigma delta converters arrived, it became easy to integrate the codec on the same low-cost cmos silicon as the processor because the codec was all gates just like the processor. But more importantly, the cost of the codec began tracking Moore's Law because its implementation was digital instead of analog. As your main processor jumped to the next smaller geometry, so did your codec. 

This was very important in the evolution of the modern SOC.

Now, CMOS still faces challenges in making very high performance analog circuits. The transistor matching isn't great. CMOS opamps exhibit a lot of temperature sensitivity. They have a lot of problems with offset. Their 1/f noise is generally not good and the "knee" continues way past 1 KHz--the heart of the audio band.

Some of these, like offset and drift, can be fixed by integrating digital. Just as getting the audio codec into a CMOS sigma delta design marked a fundamental change, the ability to integrate logic alongside the opamp means interesting things can be done to combat offset and drift. A chopper circuit is one example. Small logic blocks that measure temperature, retrieve settings from factory calibration data and make small adjustments to the opamp based on the temperature is another. In other words, logic and smarts can fix a lot of shortcomings and can allow an inferior process with smarts to zoom past a superior process without smarts.

The OPA2156 didn't just appear in a vacuum. TI has had a predecessor part in its CMOS lineage that is almost as good as the OPA2156. The OPA1652 came out in 2011 or so (and for reference, there's not a 36V high performance CMOS opamp out there really prior to 2011). But the OPA2156 has made some pretty solid improvements over the OPA1652. The '2156 adds (over the '1652) rail-to-rail input (the '1652 was RRO only), big wins to offset, 40% more GBW,  4X more slew rate and lots more drive current.

Overall, the top level details in some key amps are below:

But how about the measured performance? The quickest test is to replace both the output and input stage on the left channel of a QA401 and compare it to the right channel. 

The first plot  (yellow is left channel, red is right channel) looks at the noise with the inputs shorted. The concern here should be noise at lower frequencies as the 1/f knee for CMOS is much higher than bipolar. From the measurement, we can see that there's about 0.4 dB degradation in noise in the 20 KHz bandwidth. But this doesn't tell the full story, because we're really limited here by the converter noise floor, not the opamp noise floor. 

With a 50 ohm source, the OPA1612 should come in around -134 dBV noise (20 KHz BW) and the OPA2156 should come in around -124 dBV. That's huge--the bipolar opamp wins. And that is the biggest reason to still favor bipolar. If the input of the QA401 offered any gain (instead of just buffering) it would make it hard to use the OPA2156 in this application because the noise performance would hurt a lot for low-Z inputs. But in this particular application (and 99% of the less demanding applications out there) the noise performance of the CMOS is fine. 

Next, let's take a look at the noise performance with a 10K input impedance. This is easily done by just engaging the QA401 attenuator. 

This is a big difference, and fully expected. With a 10K source impedance, the OPA2156 standalone should have a noise output of -114 dBV, while the OPA1612 would have -110 dBV in 20 KHz. In other words, the CMOS opamp is handily beating the bipolar opamp for noise with a higher source impedance. And this is fully expected: This win comes from the dramatically lower input noise current density of the OPA2156. While the calculations show a 4 dB win using typical data sheet numbers, the measurement is showing 7.7 dB. 

Finally, we take a look at distortion. We'd not expect the OPA2156 to beat the OPA1612. Instead we'd expect it not to degrade it by much. But in this plot below, we're seeing 2H from the OPA2156 @ -122 dB, and 2H from the OPA1612 at -115 dB. Whoa nice! But be careful, at these levels very, very small differences can readily change things. Sometimes for the better and sometimes for the worse. The key point is that the THD didn't degrade by several dB.

Notice also in the above plot where the yellow trace is around 20 Hz. It appears to be about 5 dB worse than the red. Again, not unexpected given the 1/f noise in CMOS.

Summary

In short, the OPA2156 might mark the beginning of the end for bipolar and jfet audio opamps. Or maybe that die was cast with the OPA1652. In any case, I think with the OPA2156 we might be at the point where a CMOS opamp can demonstrate it's "good enough" in most applications out there. And it's quite a bit less expensive too. And as an added benefit, it simplifies source impedance concerns and delivers some impressive load drive capabilities. It does need just a bit more current than the OPA1612.

As an input stage, the OPA2156 appears to start winning whenever your source impedance crosses about 1 K ohm. If you are living in a 50 ohm or 600 ohm  world, bipolar still wins. But beyond 1K ohm source impedance, bipolar is getting challenged hard with this opamp. 

-Matt

 

Like what you just read? Join our mailing list at the bottom of the page. 


2 comments

  • Hi Demian! Yes, for sure something is going on there that needs to be understood. There is indeed more hash and situational sensitivity from the CMOS opamp than the bipolar. With an external 10K impedance the CMOS seemed to have a lot more susceptibility, even to the point it would pick-up something on my desktop running at a 100 Hz rate (maybe a keyboard being scanned). And it wasn’t there in the bipolar input channel. But that study will have to come in the future.

    Matt
  • The CMOS performance is quite remarkable.
    Looking at THD+N and comparing to the noise measurement it seems the noise floor is significantly higher in the THD+N even adjusting for a 17 dB reference reduction. And the THD+N is the same for both opamps. Maybe the other noise at HF is limiting the performance?

    Demian Martin

Leave a comment

Please note, comments must be approved before they are published