Montecarlo Simulation (mismatch variation)

[vikas]

Hello all,

I am running a simple montecarlo simulation on two transistor on the
same schematic.When I use mismatch only option in MC tool,result does
not vary.For process variation thought it works.Even defining a
correlation between 2 transistor doesn'nt proudce different results.

Now as i searched out the solution is defining an inline subcircuit(or
some prametrized model inside subcircuit)and so so....

What is this ?
can anybody tell me in a simple way what to add in model file to get
this thing working?

Regards,
--Vikas

 

[Roger Light]

Hi vikas,

I am running a simple montecarlo simulation on two transistor on the
same schematic.When I use mismatch only option in MC tool,result does
not vary.For process variation thought it works.Even defining a
correlation between 2 transistor doesn'nt proudce different results.

Now as i searched out the solution is defining an inline subcircuit(or
some prametrized model inside subcircuit)and so so....

You need to have a monte carlo model (provided by your foundry) and to
actually use these models. In my case, the model libraries are
mcparams.scs (the actual MC parameters) and cmos53.scs. For the cmos53
library I have to set the section to be cmosmc so that the monte carlo
parameters are used - otherwise even if the mcparams.scs is in the model
library list then they aren't used.

Your equivalent to mcparams.scs will contain lines such as

parameters cj50=x.xxxxe-xx

as well as statistics sections with

vary cj50 dist=unif N=x.xxxxe-xx percent=no

You may need to look at your model files to find the exact section to set.

Hope this helps.

Cheers,

Roger

 

[vikas]

Thanks Roger,

I know the section which is "stats" in my case.When I do MC analysis
with Process variation only, it works(taking random values for
parameter and giving diferent results in each run),but when i do
mismatch only it gives the same results for each run.

Somewhere I found out that in order to do the mismatch(variation)
analysis in 'Spectre', I need to surround my circuit with a
parametrized model . Here I do'nt understand how to do this modelling.

I am using TSMC18rf mixed signal lib,and want to to do montecarlo
simulation for an RF front-end.

Do you also use spectre for MC?

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Roger Light <roger.lightDOESNTLIKESPAM@nottingham.ac.uk> wrote in message news:<caucva$dsf$1@oyez.ccc.nottingham.ac.uk>...

Hi vikas,
I am running a simple montecarlo simulation on two transistor on the
same schematic.When I use mismatch only option in MC tool,result does
not vary.For process variation thought it works.Even defining a
correlation between 2 transistor doesn'nt proudce different results.

Now as i searched out the solution is defining an inline subcircuit(or
some prametrized model inside subcircuit)and so so....

You need to have a monte carlo model (provided by your foundry) and to
actually use these models. In my case, the model libraries are
mcparams.scs (the actual MC parameters) and cmos53.scs. For the cmos53
library I have to set the section to be cmosmc so that the monte carlo
parameters are used - otherwise even if the mcparams.scs is in the model
library list then they aren't used.

Your equivalent to mcparams.scs will contain lines such as

parameters cj50=x.xxxxe-xx

as well as statistics sections with

vary cj50 dist=unif N=x.xxxxe-xx percent=no

You may need to look at your model files to find the exact section to set.

Hope this helps.

Cheers,

Roger

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Regards,
--Vikas

 

[Andrew Beckett]

As I think I've outlined recently in this forum, there's a good paper on this
on http://www.designers-guide.com in the modelling section.

It talks about the methodology for modelling for monte carlo.

Here's the full URL to the paper:

http://www.designers-guide.com/Modeling/montecarlo.pdf

Regards,

Andrew.

On 18 Jun 2004 09:44:13 -0700, sharmav@cae.wisc.edu (vikas) wrote:

Thanks Roger,

I know the section which is "stats" in my case.When I do MC analysis
with Process variation only, it works(taking random values for
parameter and giving diferent results in each run),but when i do
mismatch only it gives the same results for each run.

Somewhere I found out that in order to do the mismatch(variation)
analysis in 'Spectre', I need to surround my circuit with a
parametrized model . Here I do'nt understand how to do this modelling.

I am using TSMC18rf mixed signal lib,and want to to do montecarlo
simulation for an RF front-end.

Do you also use spectre for MC?

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Roger Light <roger.lightDOESNTLIKESPAM@nottingham.ac.uk> wrote in message news:<caucva$dsf$1@oyez.ccc.nottingham.ac.uk>...
Hi vikas,
I am running a simple montecarlo simulation on two transistor on the
same schematic.When I use mismatch only option in MC tool,result does
not vary.For process variation thought it works.Even defining a
correlation between 2 transistor doesn'nt proudce different results.

Now as i searched out the solution is defining an inline subcircuit(or
some prametrized model inside subcircuit)and so so....

You need to have a monte carlo model (provided by your foundry) and to
actually use these models. In my case, the model libraries are
mcparams.scs (the actual MC parameters) and cmos53.scs. For the cmos53
library I have to set the section to be cmosmc so that the monte carlo
parameters are used - otherwise even if the mcparams.scs is in the model
library list then they aren't used.

Your equivalent to mcparams.scs will contain lines such as

parameters cj50=x.xxxxe-xx

as well as statistics sections with

vary cj50 dist=unif N=x.xxxxe-xx percent=no

You may need to look at your model files to find the exact section to set.

Hope this helps.

Cheers,

Roger

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Regards,
--Vikas

--
Andrew Beckett
Senior Technical Leader
Custom IC Solutions
Cadence Design Systems Ltd