Re: Bird 28.2 - Comments and Suggestions

Howdy, EIA/IBIS folks:

  I have a couple of comments on some earlier posts:

Point 1: [Regarding the signs of the off-diagonal C matrix entries]
----------------------------------------------------------------

  One interpretation of the capacitance matrix is that it relates
the charges Q1, Q2, ... Qn on a set of n conductors to the potentials
V1, V2, ... Vn on those same conductors. For two conductors, we'd
have

            Q1 = C11 * V1 + C12 * V2
            Q2 = C12 * V1 + C22 * V2

(This is just writing out the "matrix" multiplication [Q] = [C][V].)

  The simplest two-conductor situation is a parallel plate capacitor C.
In this case, we know that the charge on the top plate (conductor 1)
is Q1 = C*(V1 - V2); and so expanding the equation out, C11 = C and
C12 = -C. The charge on the bottom plate is Q2=-C(V1-V2), so everything
is consistent if the off-diagonal terms are assigned a negative sign.

  To think of it another way, suppose that we connect conductor 1 to a
constant voltage source, and then we raise the potential on a second
nearby conductor. Additional NEGATIVE charge will appear on
conductor 1 in order to terminate the extra field lines coming from
the second conductor.

  Many field solvers actually compute "charges", and so it is natural
for them to report the off-diagonal entries as negative.

Footnote: The situation is rather different for INDUCTANCE matrices.
In this case, we can arbitrarily assign a certain direction as being
"positive" for current flow. Depending on the orientations of the
currents in two neighboring wires, the mutual inductances between them
could have either positive or negative sign.

Point 2: [Regarding the need for G matrices]
----------------------------------------------------------------

  Here's the problem that I have with specifying the G matrix: suppose
we conduct the measurement of G at DC. Chances are that the dielectrics
used in the package are so good that G is immeasurably low; all the
entries of the G matrix are then zero.

  You _can_ run into nonzero G values at high frequencies. This occurs
because the poor little atoms in the dielectrics can't spin fast enough
(they have some mass) to reorient themselves to the rapidly changing
electric fields. The lag causes some very frequency-DEPENDENT losses.

  So, at what frequency do you specify G? And is data from just one
frequency point good for a transient analysis?

Another footnote: certainly the R and L matrices can change due to
skin effect as we increase the frequency. But the changes in L are
very minor (usu. less than 10%); changes in R are much more
substantial, but at least there is a well-defined non-zero DC value
for R that's easy to measure.

--Eric
Received on Mon Aug 28 07:54:18 1995