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

[Scott Bernardi]

This is a correct view.  Current gain (Beta) only comes in as a source of
error; the collector of the input transistor is reduced from the input
current by the two base currents (i.e., 2*Iin/Beta) in the case of a
base-collector short on the input transistor. Connecting a third
transistor from the collector of the input transistor to the two bases
reduces this error by another factor of Beta (i.e., 2*In/(Beta^2)).
Multiplication of current can occur by ratioing the size of the emitters.
For example, if the "output" transistor of the current mirror has an
emitter twice the size as the "input" transistor, the output current will
be twice the size as the the input current. You could also obtain this by
hooking two transistors in parallel for the output.   Mismatch in the
emitters will also result in an error - the offset effect.
For more precise mirroring, matched resistors can be placed in series with
the transistor emitters - this is called emitter degeneration, because the
linear IR voltage drop reduces the exponential dependence of Ic on Vbe. It
also greatly increases the output impedence of the mirror (reduces Early
effect). This is used in fixed current mirrors, not in OTA's where the
current is constantly changing (and hence the IR drop).

pnp current mirrors are typically much poorer performers in bipolar IC's
because lateral pnp transistors have poorer specs. The have lower Beta,
which decreases rapidly as Ic is increased, and they have a much lower
Early voltage. They usually have poorer matching also. That's why you can
probably create a better performing OTA with a set of THAT-140's
(dielectrically isolated npn / pnp pairs, no lateral pnp's) than you can
get with a monolithic IC.



jhaible at t-online.de wrote:

> > Well, the current transfer ratio of a mirror depends basically
> > on current gain.
>
> Is this true ? The way I understand a current mirror (at least
> the simple 2-transistor version) is this:
> *Vbe* of both transistors is the same, so if they are identical,
> and if the Early effect is neglegted, they have the same collector
> current. (The remaining error depends on beta, as one side of the
> mirror feeds both base currents.) But basically, the mirror function
> depends on the collector current as a function of the base-emitter
> *voltage*, as this voltage is forced identical on both transistors.
> The overall current mirror is formed by a exp(log()) function.
> If one transistor needs a different vbe for a certain collector
> current (the same thing that appears as offset voltage in a
> differential amplifier), this voltage offset (addition) is transformed
> to a current ratio (multiplication). That's how I always understood
> it.
> And the 3-transistor mirrors will minimize the beta influence and
> also the Early effect, but the principle would otherwise be identical.
> So this large signal Ic(vbe) funtion and its matching for two
> transistors
> would be the main effect for both, the offset of a differential pair,
> or a transfer rate unequal 1 for current mirrors.
> I'm aware that this must be a somewhat simplified view - but where is
> the error ?
>
> JH.

--
Scott Bernardi
sbernardi at home.net