[Magic-dev] RE: cap values in magic vs. MOSIS process data

[Jeff Sondeen]

probably the 4 af/um is for a set of parameters for a 'dense' layout
style, for which the side cap (Cc) factor would be dominant (which
magic would capture as long as spacings are within sidehalo!).  that
is, since side caps "shield" fringe cap (to conductors below), the
fringe factor incorporates this (assumed) shielding by being
small. it's as though the sum of fringe+side caps were a constant, but
the sum isn't, rather the sum defines a "humped" C-versus-spacing
curve, the shape of which depends on layer (M1 M2 etc).

the main improvment to magic extraction would be to let side caps
"shield" further side caps, so sidehalo could be increased without
getting too many bogus sidecaps for closely spaced wires.

but beyond this, the fact that magic doesn't see matel (say M3 to M2)
"catercorner" caps, etc. shows the need for a better cap model.

i think the Mosis data is a "red-herring" unless you're drawing
isolated 50um square metals!

/jeff

R. Timothy Edwards writes: > Dear Wei,
 > 
 > >From Jeff Sondeen:
 > 
 > > area = 5.28e-03/.4 ff/um**2 = 0.0132 ff/um**2
 > > fringe = 2.94e-02 ff/um
 > 
 > Okay, let's translate everything to aF and um.  Now
 > we have for m2->substrate:
 > 
 > 1) magic's lambda=0.15 extraction:
 > area cap = (0.354 / (0.15 * 0.15)) = 15.7 aF/um^2
 > perim cap = (0.6 / 0.15) = 4 aF/um
 > 
 > 2) MOSIS test data:
 > area cap = 19 aF/um^2
 > perim cap = 60 aF/um
 > 
 > 3) vendor data from Jeff:
 > area cap = 13.2 aF/um^2
 > perim cap = 29.4 aF/um
 > 
 > 
 > Some of this can be accounted for by the rather large
 > difference in fringe capacitance as seen in the
 > vendor data posted by Jeff according to the difference
 > in spacing of structures.  Nevertheless, there is
 > a factor of 15 difference between magic's value of
 > perimeter capactince and the MOSIS test data.  Don't
 > you agree that's a rather HUGE discrepancy?  The
 > value of 4 aF/um doesn't come close to matching any
 > vendor data for any process that I can find.  These
 > in turn are roughly backed up by calculating fringe
 > from a half-cylinder model, knowing the thickness of
 > the metal2 layer and the height above the substrate.
 > 
 > (See equation from Yuan and Trick, on page 17 of
 > http://www.stanford.edu/class/ee371/lectures/lect_02_updated.pdf)
 > 
 > Ultimately, magic should either attempt to incorporate
 > a more detailed model of capacitance extraction than
 > it currently has, but even with a simple model, one
 > would hope that the values are "ballpark".
 > 
 > 				---Tim