[sdiy] Temperature compensation results

[Ian Fritz]

Rene --

At 12:56 PM 6/13/2003, René Schmitz wrote:

>>This argument is fine as long as the mismatch offset is independent of 
>>T.  If it comes from different T dependences in Is1 and Is2 then the 
>>situation is much more complicated.
>
>This is how the Temperature dependance of Is is modelled in the Gummel 
>Poon Transistor model: (used in SPICE)
>
>IS(T) = IS(T0) (T/T0)^XTI  exp[ EG/VT ((T/T0) - 1) ]
>
>EG is the bandgap voltage
>XTI is usually 3.
>VT = kT/q
>T0 is the temperature at which the model parameters were extracted.

So you assume the same model equation for both transistors. I don't see the 
justification for this -- small offsets could be caused by small deviations 
from the equation.

>I think XTI is a mere modelling parameter, its a question which physical 
>meaning it has. I know that the number of carriers depends as 
>T^(3/2)*exp(-EG/2VT). Which would also explain the form of the equation.
>
>At least we see that Is(T) depends on Is(T0) in a linear fashion.
>If Is(T) were itself some nonlinear function of Is(T0) then bad things 
>could happen. For example you could find two transistors with an offset of 
>zero which aren't really matched, just happen to have the same Ic at some 
>specific temperature.

I don't follow you here.  IS(T0) looks like just  a number that is factored 
out. The rest of the expression collapses to unity at T=T0.

>Its the crucial question here if EG and XTI would be the same for say two 
>randomly picked 2N3904s.

I believe EG is just the Si bandgap, which is a fundamental number.  Why do 
you think it would change?

>Or the transistors inside a CA3046 or 2SC1583. So we are back to the point 
>"geometry" vs. "defects". But thats only natural, since a transistor model 
>can't answer those questions. Only experiments can.

I think the crucial question is whether ideal model behavior can be assumed 
when you are looking for small effects that could cause mismatch..

>>Gradients in themself may not be a show stopper, as long as they are 
>>reproducible and non-fluctuating.  Turbulent air flow and drafts are what 
>>need to be controlled, I think.
>
>Agreed. But then, I think you would need to do your calibration in the 
>target environment, where you have the final gradient.

Right.  But I think you can take care of this simply by putting the circuit 
in a small enclosure, like an Al or plastic case.

BTW, do you know that all the mail to your address uzs159 at uni-bonn.de is 
bouncing?

   Ian