Designer’s Guide Community

horizon

Junior Member

Offline

Posts: 16

montecarlo eldo spectre standard deviation
Apr 6^th, 2006, 4:34am

Dear All,
I am comparing result I have of the standard deviation of a bandgap ouput voltage (I am using spectre) and results a collegue had using same model, but with eldo.
The models are identical for both of the simulator.
The spreading is higher with eldo.
The guy was using an option with eldo : "MCLIMIT"
I would like to know if one of you knows more about this option and especially if this option is changing the original spreading written in the model. I read the eldo manual but it 's not clear for me.
Does somebody knows the equivalent command in spectre ?

thank you very much for your help
have a nice day
� Horizon

Back to top

« Last Edit: Apr 07^th, 2006, 12:38am by horizon »

IP Logged

achim.graupner

Community Member

Offline

Posts: 51

Re: montecarlo eldo spectre standard deviation
Reply #1 - Apr 12^th, 2006, 12:04am

HI there,

I do not know anything about eldo. But I may suggest two things to consider:

In the spectre model paramter file (in general called mcparam) there is a parameter called truncate. It truncates the probability density functions by multiples of their standard deviation:
statistics {
truncate tr=4.0
}

A second problem migth occur when you use multiplier > 1. The multiplier is not necessaryly regarded in the model card when computing the device's area. May the spectre and elso netlisters are doing the same.

Good luck,
achim

Back to top

Achim Graupner
ZMD AG, Dresden, Silicon Saxony, Germany

IP Logged

Paul

Community Fellow

Offline

Posts: 351
Switzerland

Re: montecarlo eldo spectre standard deviation
Reply #2 - Apr 12^th, 2006, 1:35pm

Hi,

are you using subcircuits for all devices in Eldo. If you are using model cards, the Monte-Carlo is not performed correctly, because the same parameter variations are applied to all devices with identical dimensions. Indeed, Eldo does not know "inline subcircuit" statements. In your case, does Spectre give larger spread than Eldo? This would be a hint on the fact that the Eldo computation is not correct.

Paul

Back to top

IP Logged