[sdiy] Quad transistor sources

[Magnus Danielson]

From: WeAreAs1 at aol.com
Subject: Re: [sdiy] Quad transistor sources
Date: Sun, 27 Jun 2004 22:22:00 EDT
Message-ID: <1e6.23c7b49e.2e10dac8 at aol.com>

> 
> In a message dated 6/27/04 5:42:43 PM, cfmd at bredband.net writes:
> 
> << The MAT-04 gives you four transistors. The MAT-02 is basically the same
> silicon but with a different metal layer hooking them up in two-and-two
> (in parallel) such that temperature gradients is cancelled (to some
> degree). >>
> 
> I've often wondered if there might be some practical advantage to doing this 
> with discrete transistors.  That is, connecting several of the same type 
> together in parallel so that their individual differences are kind of cancelled out 
> (or rather, averaged out).  I imagine such a thing could work, but who knows 
> how many you'd need to parallel in order to get the desired result.  I once 
> read that the National "super-matched" precision transistor pair (LM394) is made 
> this way, and that it has 100 paralleled transistors on each side of the 
> "pair", for exactly this purpose.  Is this true?  When you parallel a bunch of 
> trannies, does the internally paralleled Base input require more overall current 
> in order to properly drive the transistor?  (that is, does the effective 
> resistance of the Base junction change when you link all the internal transistors 
> together?)

I doubt that you get as good results as when you do it on a single chip. The
point is that they are intrinsicly very closed matched just by being done with
the same geometry on the same piece of silicon (going through the same process
at the same time, so their deviations in process-dependence is very small).
Also, they have a very tight thermical coupling as it is, which is also hard to
acheive. It's not that it is magic, but it is hard to acheive it as well with
other methods.

> I saw a schematic once for a mic preamp in an Amek Angela mixing console, and 
> they had several (maybe eight or ten) discrete transistors paralleled 
> together in ther preamp's differential input "pair" (but no explanation why in the 
> service manual)  One can only guess that if the transistor pair is better 
> matched, then its common mode rejection and noise cancelling function might be 
> better (especially if you also use matched resistors in the circuit).  I wasn't 
> aware that this was a problem in typical two-transistor/one-opamp mic preamps -- 
> most of them just use an unmatched pair of fairly cheap devices.  Any thoughts?

This is a classical way of reducing the input noise. When you double the
transistors the signal will raise 6 dB since it is coherent but the noise
will only raise 3 dB since it is non-coherent. This means an improvement of the
signal-to-noise of 3 dB. 8 transistors in parallel is 9 dB och improved signal
to noise.

Cheers,
Magnus