Monte Carlo - Virtuoso IC 6.1.5

It works so many thanks.. by the way I am using this version [IC 5.1.41]

It works if I choose together process and mismatch variation but when I choose mismatch variations only, the standard deviation is to small in femto scale while the mean in micro. 

Cadence (R) Virtuoso (R) Spectre (R) Circuit Simulator
Version 7.2.0.109

It works when I choose both process and mismatch. But, when I choose mismatch by itself, it doesn't. Maybe as you said, the problem on the mismatch script. How can I solve it?

First of all, I would suggest looking at your device models. If you look at the model files that are included in your netlist (i.e. those referenced from ADE) and check to see if there is a statistics block in the model files. This will look something like this:

statistics {
    process {   // process: generate random number once per MC run
        vary rshsp dist=gauss std=12 percent=yes
        vary rshpi dist=gauss std=rshpi_std // rshpi_std is a parameter
        vary xxx dist=lnorm std=12
        vary uuu dist=unif N=10 percent=yes
        truncate tr=2.0 // +/- 2 sigma
    }
    mismatch {  // mismatch: generate a random number per instance
        vary rshsp dist=gauss std=2     
        vary xisn dist=gauss std=0.5    
        vary xisp dist=gauss std=0.5    
        truncate tr=7.0 // +/- 7 sigma
    }
}

Probably just doing "grep statistics *.scs" would be a good start to find it. Check to see if there's a mismatch section.

Assuming all looks OK there, the models themselves need to be subckts. Mismatch is applied to each subckt and then the models within those subckts can be unique. If the transistors you are using are just direct models (without a subckt around them), then no mismatch will be applied. Some technologies have different components for mismatch than for normal use (for strange reasons).

Not sure if you said, but what technology are you using?

Regards,

Andrew.

Sorry, I took from your time a lot.

I am using 180nm, here what I find in the statistics part:

statistics {
    process {
        vary par1_rf dist=gauss std=1/3
        vary par2_rf dist=gauss std=1/3
        vary par3_rf dist=gauss std=1/3
        vary par4_rf dist=gauss std=1/3
        vary par5_rf dist=gauss std=1/3
        }
    mismatch {
        vary vthnmis_rf dist=gauss std=1/1
        vary dlnmis_rf dist=gauss std=1/1
        vary dwnmis_rf dist=gauss std=1/1
        vary toxnmis_rf dist=gauss std=1/1
        vary vthpmis_rf dist=gauss std=1/1
        vary dlpmis_rf dist=gauss std=1/1
        vary dwpmis_rf dist=gauss std=1/1
        vary toxpmis_rf dist=gauss std=1/1
        }

    }
 
include "rf018.scs" section= mos
endsection mc

What do you think? Does it make sense?

Thanks again and I appreciate your help.

Knowing just the process node isn't what I was after - I wanted to know the foundry too. Anyway, from what you've posted, it looks like TSMC.

Are you using the "_mac" or "_mis" devices in your schematic (I can't remember whether they are called mac or mis in that particular technology)? I suspect not...

TSMC had this tendency to have two different flavours of device, one which supports mismatch and one which doesn't. Most customers use the mismatch (or macro) devices all the time as changing the schematic is not really something you want to do on a regular basis - and with spectre, there's not really a disadvantage of using the subckt models. I think in newer technologies they always need to use a macro model so this isn't necessary.

If you are not using the mac/mis devices, then monte carlo will be varying the parameters (because it has a statistics block), but they won't be consumed by the models, hence no significant variation. Process variation will still work however because that doesn't need the models to be in subckts.

Regards,

Andrew.