“Discrete is an old audiophile hobby horse,” said Lars. “It proves nothing. Consider writing a few words.”
Yes, precisely. Doing things discretely has no inherent value. The normal procedure is that you have a set of requirements and you go looking for a solution. These days, that solution is usually available as a chip, particularly for something as bread and butter as an op amp. Duplicating that with separately packaged transistors will add exactly zero value. You only do that if you want something that chip makers aren’t offering. It’s the circuit that matters, not whether it’s on a die or on a circuit board.
The specific challenges lie in the application. We’re looking for an op amp to be used in a low gain buffer stage.
- Input impedance should be high, or it wouldn’t be a buffer. The op amp must be used in a noninverting configuration.
- It should also be linear. If the source impedance and the impedance of the feedback network are not matched, non-linear capacitances of on-chip ESD diodes can cause more distortion than the entire op amp. That problem goes away if you can match the impedances on both inputs. If the circuit is inverting, the inputs don’t swing so the problem goes away too. But as I said, we’re looking at a noninverting buffer and we can’t second-guess the source impedance.
- The output must be able to drive a heavy load without causing distortion to increase. The noise of a low-gain buffer is dominated by thermal noise in the feedback network, not the op amp input noise. So, to get low noise the feedback network has to have a pretty low impedance.
- We also don’t want to resort to class A. The idea of an input stage that has greater idling losses than the attached power stage does not appeal. This means dealing with distorted currents flowing through the power pins and that distortion couples magnetically into the inputs. What muppet came up with the idea of putting an op amp’s power pins as far away from each other as possible? I despair when I see people market discrete op amps that simply copy that pinout. If any IC people are reading: please put the power pins next to each other in the future. The sum of both currents is undistorted so all you need to do is run the +/- power traces directly on top of each other.
- Finally, the traditional op amp has no explicit output reference. Crudely put, the output voltage is GBW times the integral of the input voltage, but that’s relative to what potential precisely? It’s whichever rail the transimpedance stage sits on. Noise on that rail is coupled directly to the output, and it’s left to the feedback loop to remove it. And so, op amp chips have two PSRR curves. The worst of the two is essentially equal to loop gain. I like to see an explicit reference pin that I can tie to ground and that way get much better PSRR. That reference pin should be directly adjacent to the output pin (a shout-out here to the OPA1622 which does have an explicit ground pin, and power supplies sensibly close together – it has other shortcomings unfortunately).
That’s quite a shopping list, and since there doesn’t seem to be a chip around that ticks all boxes, discrete it is.