[No title]

Monte carlo analysis mean cutoff is double the fc from one point analysis in Cadence

I have designed a low pass filter in cadence with cutoff frequency 50Hz.
However when I do a monte carlo analysis it gives mean cutoff frequency as 80Hz.
How is this possible.??
All the parameters are same??
Please help me to understand this.
Thanks

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

The mean value of a MC analysis may well deviate from the "designed" value. At least for a low number of runs - or if there are asymmetric parameter distributions.

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

I have done analysis with 100 sample.
Isnt it sufficient??

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

Yes, this should be close enough. Did you check the parameter distributions for asymmetry? Like those: -20/+10% , e.g. for caps.

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

what do you mean by 'the parameter distributions for asymmetry'?
How can I check that?
Thanks for your help

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

richaphy said:

How can I check that?

Click to expand...

The parameter variations are defined in the statistics section of the MC control file.

Example:

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

Log-normal or other non-gaussian / non-uniform stats
will have means that are not centered in the span.
Nonlinear circuit elements can make response of
uniform stats, one-end-weighted. I have not seen the
form of stats posted by erikl but the std= appears
to make an unequally-weighted distribution?

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

dick_freebird said:

... the std= appears to make an unequally-weighted distribution?

Click to expand...

I think I've seen such asymmetric std= declarations, mainly for litho variations.

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

Today i rechecked everything.
Models are correct.
St. Dev is 11.23 for a sample for 100 no.
But BW mean with monte carlo is 81Hz, and with regular one point analysis is 52Hz.
However, I came to have an explanation behind this.
I am using MOSCAP in this design. And the filter works in weak inversion.
Is it possible that they are reason behind this deviation??
This is the only logical reason I am able to come across.
What is your opinion??
Thanks for your help.

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

richaphy said:

What is your opinion?

Click to expand...

Actually I don't know the difference in setup and speed between the one-point and the MC runs. Do you possibly use different control files for the actual analysis?

Re: Monte carlo analysis mean cutoff is double the fc from one point analysis in Cade

You may want to experiment with the MC setup - for one,
turn off mismatch and check the spread again. Not all of
the variation is "process" alone. And mismatch can really
whack an operating point, slave:master bias ratio varies at
(how many?) mV/decade in subthreshold and if you use
small devices, tens of mV mismatch may yield you (heh)
a poorly controlled setpoint where it counts.

Recommend also that you start keeping data on the
aspects you think may matter - gm of a FET as-used,
the capacitance of the filter cap, resistor as-used if
you do use them, FET VT, etc., etc. Then you could
plot vs mccnt (or whatever index is in your world) and
then change axis for a poor-boy regression of fc against
anything you kept. There lies understanding, of the
"why?".