[sdiy] Matched transistors - critical parameters?

[ASSI]

On Wednesday 08 February 2012, 09:05:50, Matthew Smith wrote:
> The DMMT5551 (and DMMT5401 for PNP) from Diodes Inc., are "intrinsically
> matched," which is explained as being built on adjacent dies from a
> single wafer.

This might be a marketing term for "we have no reason to believe otherwise, 
but we don't bother to actually measure it".  Check the datasheet for what 
they _guarantee_ for the matching (I'd be very surprised if Vbe matching was 
guaranteed to be better than about 10mV, for instance).  Actual measurements 
of matching properties are rather expensive since they take time.

> So, with the constant improvement in fabrication
> technologies, I wouldn't be at all surprised if electrical matching were
> not as good as, if not better, than what might be considered dinosaurs,
> in silicon terms.

That's OK, just as long as you are also not surprised if that assumption 
turns out to be wrong.

Matching is a function of the technology type and maturity, but even then 
there are limits: all other things equal, matching improves with sqrt(area) 
and with less distance between matched features on the wafer (obviously 
since that feature takes up space, there's a conflict between those two 
rules).  If any variable that bears on the matching has a gradient over the 
wafer or package, then the matched features need to be on an isoline of that 
variable.  That's the reason those expensive matched transistors are not 
only made on the same die, but are much larger than their single-transistor 
brethren and have interdigitated and common centroid layout - that cancels 
all linear gradients.  With higher orders of symmetry you can cancel 
nonlinear gradients of increasing order if you really care about that.

> As far as heat matching is concerned, and I realise I could well be
> wrong on this, I can't see how the two dies, encapsulated in an area of
> about 1.5 square millimetres or less, wouldn't be isothermal as near as
> dammit. Not sure about the thermal conductivity of the lead bonding but,
> even with emitters connected externally, there is still a thermal path
> going that way.

Two transistors on the same die are in better thermal contact than two 
transistors on separate dies in the same package by at least an order of 
magnitude.  This can become readily apparent if the power dissipation of the 
two transistors is not similar or if there are thermal gradients on the 
board.  If you want to to do a nice stunt circuit, put a heater left and 
right from the transistor pair  and trim the offset by adjusting the power 
of the heaters...

If you are replacing a circuit that usually can be built by gluing two hand-
matched transistor packages together, then those non-monolithically dual 
transistors are a much nicer option: easier to work with and probably even 
cheaper.  Don't forget you can still select the best matching ones according 
to your needs if you want to wring out some better performance.  Just don't 
think that you've found a replacement for a MAT03.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Factory and User Sound Singles for Waldorf rackAttack:
http://Synth.Stromeko.net/Downloads.html#WaldorfSounds