In simulation of BGR(Band Gap Reference) circuit using standard CMOS process,
I use following parameters as temperature effects in Substrate-PNP-Bip's Gummel-Poon model file.
   tlev=1, tlevc=1, tnom=25, xtb=0.00557, xti=3, eg=1.162

But actual measurement result of temperature dependency of our BGR circuit is fairly large than simulation.
Of course, I include all temperature coefficients of other all devices such as MOSFET, resistor and capacitor.
I also tried monte carlo analysis with mismatch variation while temperature sweep.

Even if my design is poor as good BGR, simulation should show large temperature dependency in mismatch monte carlo analysis.
But simulation don't show large temperature dependency at all.

I think accuracy of our PNP-Bip model is very poor especially regarding temperature dependency.

Here my questions are :
(1) I 'm now suspecting that PNP-Bip give more tempereature effects than simulation. Is this reasonable ?
(2) If so, should I change "xtb" parameter ?
(3) Or should I specify more detailed temperature coefficients, such as tis1, tikf1, tikr1, tbf1, tvaf1, tbr1, etc. ?

These parameters are common in any vendor's simulator.

sheldon wrote on Jun 20^th, 2010, 4:38am:
Needless to say, of course.
I also tried to increase mismatch variation parameter for "IS" of PNP-Bip than data provided by foundry since it is fairly small.
I also tried to perturb "area" parameter of Bip.
If I change these very largely but unreasonably, I can get large temperature dependency.

sheldon wrote on Jun 20^th, 2010, 4:38am:
This is not useful and not helpful for me.

Rather the followings are interesting for me.
http://www.designers-guide.org/Forum/YaBB.pl?num=1276142600
http://www.designers-guide.org/Forum/YaBB.pl?num=1276024989

sheldon wrote on Jun 20^th, 2010, 5:08am:
After many trial simulations, we are now suspecting that PNP-Bip give more tempereature effects than simulation.

See my original post http://www.designers-guide.org/Forum/YaBB.pl?num=1277030583/1#1