Statistical modeling with parameter mc_ok

[miesli]

Expected Behavior

Turning on and off statistical modeling by choosing different corners

Actual Behavior

Turning on and off statistical modeling by choosing different corners PLUS option to change the sigma of the Gaussian function using mc_ok.

Fix

There are multiple options depending on the wanted behavior. The most obvious is:

Replace all mc_ok appearances with Sigma in the ngspice/models/cornerXXX.lib files

Thanks

I report this here after someone else reported this to me.

[dwarning]

No, this is partly a misunderstanding:
In a former version of the PDK lib with different library concept mc_ok was used to switch on/off statistical simulation with Monte Carlo.
Actual the selection followed by mos_tt_stat library parameter. Everything works well as we can see in attached picture:

The problem is that the name mc_ok is not correct anymore. He has now the meaning of Sigma.
The simulation is made with Sigma=6.

[simi1505]

@dwarning Can you share your testbench or at least how you invoke this corner with Sigma=6?

.lib cornerMOSlv.lib mos_tt_stat ...?

I want to try the Monte Carlo simulation in combination with CACE.

What is a suitable value for Sigma = mc_ok? Where do you get 6 from?

Thanks!

Edit: Ok, never mind. Just make .param mc_ok = x.

[dwarning]

@simi1505
Please at first read the ngspice manual (or hspice/spectre) about the meaning of the gauss() and agauss() function parameters.
Assuming the standard deviation values in IHP statistical libraries (I am not sure about this!) are standing for 3 Sigma. Then you get 3 Sigma deviations with mc_ok = 1. Widening the spread to 6 Sigma needs setting mc_ok to 0.5.
A reworked corner/mismatch library concept is pending: #333
Would be interesting to see spec compliance of a practical circuit with this high deviations.

[sergeiandreyev]

Assuming the standard deviation values in IHP statistical libraries (I am not sure about this!) are standing for 3 Sigma.

Yes, these should be 3-Sigma

[simi1505]

Thanks for the fast support!

However, there is something I do not understand. I thought mc_ok equals sigma?

The Ngspice manual defines gauss() as follows:

define gauss(nom, rvar, sig) (nom + (nom*rvar)/sig * sgauss(0))

And in the sg13g2_moslv_stat.lib file it is called with .param sg13g2_lv_nmos_vfbo_mm ='gauss(sg13g2_lv_nmos_vfbo_mm_norm , 0.004, mc_ok)'.

If I now define .param mc_ok = 3, I should get sigma = 3, right? I do not understand, where you get mc_ok = 1 --> sigma = 3 and mc_ok = 0.5 --> sigma = 6?