[comp.lsi.cad] Capacitance extraction with dracula

dauber@flush.Eng.Sun.COM (Jeff Dauber) (01/15/91)

I am trying to extract capacitances using dracula and I have come up
with a problem concerning fringe capacitances.  Currently the method
that I am using concerns oversizing each of the layers by some "fudge"
factor and then extracting the capacitances in the usual fashion.  The
problem comes in when I try and use this technique with our new
process.  Fringing is so significant that when we oversize our layers,
they merge into one big layer.  I cannot think of a way around this
using dracula, and was wondering if anyone had any suggestions.


Jeff

mpe@shamash.cdc.com (Mike Ebsen) (01/15/91)

The best way that I know to extract fringe cap is to downsize 
the layer in question by a very small amount (i.e. 0.001um).  The value
that you use should be at least one order of mag less than your minimum
layout geometery - that way merging is not an issue.

This new layer is then used to generate the difference of the original
layer minus the downsized layer to produce a very small skeleton layer 
which outlines all of the fringe area that you want extracted.  To better
imagen what this is, think of a children's coloring book which has a 
vacant inside (where the person colors) and a dark black border which 
confines the region.

Next extract the cap of the black outline region.  Use an cap/area multiplier
of 2/0.001um or whatever was used in step #1.  The values which dracula
will return will be the perimeter +/- 4*0.001 (or whatever value was used in
step #1.  A simple proof:
      Think of a square, 
            area = L * W
            slim-area = (L-s) * (W-s)

            area - slim-area = LW - [LW - sW - sL - s^2]
                             = sW + sL + s^2

            above/s = (sW + sL +s^2)/s
                    = W + L + s

            Note: the final s term is dependent on the shape of the 
                  structure (i.e. # inside and # outside angles).  This
                  term will introduce a small error in the estimate of 
                  the perimeter of the structure but since it is one
                  order of mag less than any feature size of interest -
                  "who cares".

            perimeter = 2 * (W + L + s)
                      = 2W + 2L + 2s


To produce usable cap values change the term used above from
    2/0.001um to    cap_unit_length_fringe*2/0.001um.

Now dracula is extracting only fringe cap.