current mirrors (was: AW: [sdiy] discrete SSM2018?)

[jhaible at t-online.de]

> Well, I thought that you feed a current mirror with a 
> current. 

Sure, but what is "passed" from one transistor to the other
is a voltage. Beta has some (minor) influence, but it's
greatly reduced in the 3-transitor mirror. 
But the main thing that happens (even with a hypothetical
infinite beta) is current-to-voltage-to-current conversion
with the typical log and expo laws.

> Or in other terms: how much of the emmitter current will
> make it to the collector?

Well yes, but that's all summarized in the "beta" part.
Beta would play a big role if the two transistors were
forced to have identical base current. Instead, they are
forced to have identical bease-emitter voltage. So
if a mirror has a ratio that is much different from 1
(from two discrete transistors, for instance), beta
cannot be the reason for this. Even a perfectly matched
beta would cause an error around 2/beta (or something
similar) for the mirror, and a severe mismatch of beta 
wouldn't cause significantly more, but a vbe "offset"
of 20mV can cause 100% error (twice the current in one
transistor). This is what makes sense (to me) for
discrete transistors at least. But for integrated
mirrors, there might be other tricks (?).


> So I think it is possible that offset will have no or 
> weak correlation with beta. 

So do I, but I think the beta (matching) is not the important 
thing here. It should be the offset, for both the mirror and
the differential amp.

JH.