[sdiy] discrete SSM2018?

[jh.]

> Well, it's inversely proportional to temperature, for one thing. gm =
> Iabc*q/2kT or Iabc/.052v at 25 deg C - that's where the 19.2*Iabc comes
> from.

That's from the differential pair on the input. (In the mirrors, temperature
effects of the log and expo should cancel).

With a simple differential pair, we have a rather tight gm. It's a function
of the Iabc current and temperature, but (almost) not of the transistor
parameters. Even with a bad Vos, the gm would be well defined at
its steepest point of the transfer courve.

The conclusion of this is, when a OTA is spec'ed with a gm
tolerance of 1:2, this must come from the current mirrors.

The mystery is, when the mirrors are so badly spec'ed, how
can the npn mirrors be well tracking with the pnp mirrors
(If they weren't, the gain
would be different for different input polarity.), and how
can all the mirrors, with +/-10mV mismatch allowed inside
*one*mirror (coresponding to 1:2 current ratio)
nevertheless be matched for the two opamps of a dual package?

I'm repeating myself. (;->) But there must be a logical error
somewhere in this, and I'd like to find it.

JH.