[SOLVED] Are Correlations Between Parameters or Devices Supported in HSPICE?

[cereal]

Hello everyone,
In the Cadence Virtuoso, the correlations between parameters or devices can be easily done by following definitions.

statistics{
	process{
		vary Lb dist=gauss std=1.46n/3 percent=no
	}
	
	correlate param = [Lb_W1 Lb_W2] cc=0.9
	mismatch{
		vary Lb dist=gauss std=2.5n/3 percent=no
	}
	correlate dev = [I10 I13] cc = 0.8 // cc is coefficient.
       }

The correlations are considered due to spatial locations of devices which result in the process variations.
However, I would like to know whether the correlations can be defined in HSPICE? And how?

Thanks a lot...

 

[keith1200rs]

I cannot check at the moment but I think the keyword you need to search for is the LOT parameter (unless I am mixing it up with Pspice). Look for TOL, MATCH and DEVICE.

Keith

 

[cereal]

I cannot check at the moment but I think the keyword you need to search for is the LOT parameter (unless I am mixing it up with Pspice). Look for TOL, MATCH and DEVICE.

Keith

Very appreciate for the advice. I think LOT and DEV parameter may be of great use.
Here is the LOT definition in PSPICE Manual.


Here is the definition of LOT&DEV in HSPICE Manual.
LOT/n, DEV/n
(Monte Carlo) Which of ten random number generators numbered 0
through 9 are used to calculate parameter value deviations. This
correlates deviations between parameters in the same model as well
as between models. The generators for DEV and LOT tolerances are
distinct: Ten generators exist for both DEV tracking and LOT tracking.
N must be an integer 0 to 9.

They are similar and this is really close to what I want!
However, the last problem I have to solve is LOT&DEV provides limited number of random generaters...
How can I use these for defining a coefficient of correlation?

Thanks a lot indeed.

 

[keith1200rs]

Sorry, I cannot remember and it will be Monday before I can check. My memory is of being able to specify something like DEV=50% LOT=2% on the DEVICE parameter line although I seem to remember being able to also specify a LOT name which allowed you to specify which parameters tracked (a bit like the LOT number in the clip you posted). It is a long time since I have had to use it!

Keith

 

[cereal]

Sorry, I cannot remember and it will be Monday before I can check. My memory is of being able to specify something like DEV=50% LOT=2% on the DEVICE parameter line although I seem to remember being able to also specify a LOT name which allowed you to specify which parameters tracked (a bit like the LOT number in the clip you posted). It is a long time since I have had to use it!

Keith

Thank you very much for this information.
In my view, "DEV=50% LOT=2%" perhaps is the standard deviation of random variables.
I will try to use DEV/n /distribution and LOT as well.
Please let me know if you have any information about this.

Best regards.

 

[keith1200rs]

You need to check the documentation - some software consider the percentage to be 3SD, some expect you to express it as a fraction e.g. 0.02 and expect it to ne 1SD. Also, you can usually specify the distribution e.g. Uniform or Gaussian.

Keith

 

[cereal]

You need to check the documentation - some software consider the percentage to be 3SD, some expect you to express it as a fraction e.g. 0.02 and expect it to ne 1SD. Also, you can usually specify the distribution e.g. Uniform or Gaussian.

Keith

Emm, I have checked the documentation, the standard deviation in HSPICE is a bit tricky. However, in the DEV and LOT context, the std is 1. And uniform is an default setting, I can specify gaussian by use keyword 'gauss'. However, I am puzzle at the LOT/n where n can be a media to correlate the other parameters. According to the documentation, the number of n varies from 0 to 9, I am a bit lost at what if I have more than 10 devices that should be correlated...

Many thanks.

 

[keith1200rs]

I am not sure if you have the correct manual so I have attached part of it for reference. The following example shows the use of LOT & DEV:

.model m1 nmos level=6 bulk=2 vt=0.7 dev/2 0.1
+ tox=520 lot/gauss 0.3 a1=.5 a2=1.5 cdb=10e-16
+ csb=10e-16 tcv=.0024

You normally don't need many LOT variables. Mostly just one. All devices on the same wafer, for example, would be part of the same LOT. If you wanted to group them for some reason, then you might have two groups or LOTs.

The best way to check it out is to set up a simple simulation with a few resistors and try it out.

Keith.