Dear All,

I'm currently doing an inductor Monte Carlo simulation in Cadence-Spectre (Cadence version: 5.10.41_USR5.90.69; Spectre version: 6.2.0.420). My objective is to see how the inductance and Q-factor (differential-driven) changes over process variations.

I set up a few expressions to convert 2-port S-parameters to 1-port differential equivalent as follows:

monteExpr( "S11" "sp(1 1 ?result 'sp)" )
monteExpr( "S12" "sp(1 2 ?result 'sp)" )
monteExpr( "S21" "sp(2 1 ?result 'sp)" )
monteExpr( "S22" "sp(2 2 ?result 'sp)" )
monteExpr( "Sdm" "0.5*(S11-S12-S21+S22)" )

The plots of first four expressions are working fine all the time, and I can see the effect of process variations. But the "Sdm" expression gives problem, sometimes it gives only one curve, and sometimes (probably without anything changed), it even gives error message as follows:
*Error*   Evaluating expression ((0.5 * ((S11 - S12 - S21) + S22))).
*Error* ("eval" 13 t nil ("*Error* eval: unbound variable" S11))

I don't know what's going wrong. Does this ring any bell for you? Thanks.


Best Regards,
Ian

pancho_hideboo wrote on May 2^nd, 2009, 6:46am:


I actually tried the alternative expression for "Sdm" you've mentioned, and it didn't work either.

I replied this topic on another forum, but will post here for anyone who is interested:

s11, s12, s21, s22 have no meaning to spectre, you must use aaSP(m n) where m,n can take value 1 or 2

try loading these variables into your ADE calculator:

armSetCalcMemory( "Sdiff" "(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2))/2" )
armSetCalcMemory( "Zdiff" "2*50*(1+(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2))/2)/
(1-(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2))/2)" )
armSetCalcMemory( "Rdiff" "real(2*50*(1+(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2)
)/2)/(1-(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2))/2))" )
armSetCalcMemory( "Xdiff" "imag(2*50*(1+(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2)
)/2)/(1-(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2))/2))" )
armSetCalcMemory( "Ldiff" "imag(2*50*(1+(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2)
)/2)/(1-(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2))/2))/(2*pi*xval(aaSP(1 1)))" )
armSetCalcMemory( "Qdiff" "imag(2*50*(1+(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2)
)/2)/(1-(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2))/2))/real(2*50*(1+(aaSP(1 1)-aa
SP(1 2)-aaSP(2 1)+aaSP(2 2))/2)/(1-(aaSP(1 1)-aaSP(1 2)-aaSP(2 1)+aaSP(2 2))/2))
" )

in order: differential sparameter, differential impedance, differential resistance, differential reactance, differential inductance, differential quality factor. even if you define a variable in calculator, if you just type the variable name (ex, Sdiff) into calculator expression it will give error. you have to select the variable from the list of user defined variables in calculator, then in the expression you will see the full equation that the variable represents. for example you can not type in calculator 0.5*Z0*(1+Sdiff)/(1-Sdiff) to get differential impedance. very stupid i know, but it is the way calculator works.

pancho_hideboo wrote on May 2^nd, 2009, 6:54pm:


ahh i see. because in calculator you can find s11, s12, etc as available functions but when you select them the expression editor will enter aaSP(1 1), aaSP(1 2), etc. i always find it annoying that you can define variables but not refer to them explicitly in expression editor.

Thanks again, oermans!

As for a single iteration of SP simulation using Ocean stript, the following functions just work fine:

S11=sp(1 1 ?result 'sp)
S12=sp(1 2 ?result 'sp)
S21=sp(2 1 ?result 'sp)
S22=sp(2 2 ?result 'sp)

Sdm=0.5*(S11-S12-S21+S22)
Zdm=100*(1+Sdm)/(1-Sdm)
Xdm=imag(Zdm)
rdm=real(Zdm)
Qd = imag(Zdm)/real(Zdm)

BTW, aaSP() and VF() can be accessible in the script. Compared to SP(), aaSP() takes more processing time in my observation.

Ian,

Quote:


I believe in some versions it does automatically create SKILL variables with the same name as the monteExpr name. But if you're using too old an IC version, this won't work. A workaround is to do this:

Code:

monteExpr( "S11" "s11=sp(1 1 ?result 'sp)" )
monteExpr( "S12" "s12=sp(1 2 ?result 'sp)" )
monteExpr( "S21" "s21=sp(2 1 ?result 'sp)" )
monteExpr( "S22" "s22=sp(2 2 ?result 'sp)" )
monteExpr( "Sdm" "0.5*(s11-s12-s21+s22)" ) 

I changed the variables to lower case to avoid clashing with auto-generated variables if you end up using a more recent IC version.

Note, I didn't test this - but I think it should work.

Best Regards,

Andrew.