From: Sainath Nimmagadda ([email protected])
Date: Tue Mar 20 2001 - 10:31:53 PST
Dear Michael,
Thank you! I am glad my two cents were well spent. Correct me if I am
not correct, solutions that use matrices as a way assume closed form
(integrals ?) of equations (discretization and attending computational
stuff assumed). If that is not correct, following paragraph may be
ignored and I am going to learn something new.
If that is correct, the solution methods you mentioned give misleading
answers where the current loops are not naturally closed or imagined.
That is expected via the nature of solution method. However, there are
solution methods based on differential equations that don't use
matrices. For example, FDTD. I wonder if some one out there has some
related experience (FDTD or other) to share. Knowing how those methods
work, my intuition is their experience would be a bit more pleasant.
Thanks again,
Sainath
"Tsuk, Michael" wrote:
> Sainath,I used to believe as you do, that partial inductances are
> useful to obtain some first-cut answers. Over the years, I've changed
> my mind. I believe that the potential for misuse from partial
> inductances outweighs their benefits, and I'm now doing all my signal
> integrity modeling with loop inductances. I'm much happier. :-)Here
> are some of the problems I see with partial inductances:1.) They are
> arbitrary; as Brian Young points out in his wonderful new book, you
> can add any constant you want to the partial inductance matrix without
> changing the physical result. Different techniques for calculating
> partial inductance give different answers --- witness the discussion
> we've just had on this point.2.) When you use partial inductances in
> SPICE simulations, they give you things that look like "ground
> bounce": voltage differences across large sections of your circuit,
> where it is impossible to make a unique physical measurement of
> voltage because of linked flux. Brian Young again points out that
> ground bounce is not unique; it depends on your definition of partial
> inductance. You can be mislead by how chip ground is bouncing with
> respect to module ground in your simulation --- it looks like
> something real, but it's not. When you use loop inductances, and use
> SPICE node 0 to represent local reference everywhere, you can't be
> mislead; there's no node voltage in your simulation that looks like
> ground bounce.3.) If you use partial inductances in your SPICE
> simulations, you have to make sure that all the current in your
> simulation moves from one side of your circuit to the other only
> through the partial inductances. If you have node 0 on both sides,
> for example, you've violated the assumptions under which partial
> inductance is valid. And it can be very hard to avoid node 0
> sometimes, and it appears that having large sections of your circuit
> isolated from node 0 makes convergence more difficult.4.) Partial
> inductances are completely invalid without mutual inductances, but
> there's a great tendency to ignore them as a "first-pass engineering
> assumption". This is natural; all of engineering is about ignoring
> things. :-) But it just doesn't work with partial inductances. At
> best, you're making assumptions about where the return path is (and
> different ways of calculating partial inductances make different
> assumptions); at worst, you miss the entire point of the exercise.
> Without partial mutual inductances, there's no reason to put power and
> ground planes close to each other.Basically, my feeling now is that
> partial inductances are a wonderful tool for calculating inductance in
> the standard signal integrity situation where the full loop is not
> completely known (package without chip or board, for example). But I
> think now they should remain a computational tool, and that the models
> that are eventually generated should be based on loop inductances.I'm
> working on a paper explaining these points in more detail and talking
> about how we've been using loop inductance rather than partial
> inductance for package modeling here at Compaq. I hope to present the
> paper at EPEP'01 here in Massachusetts. I would appreciate any
> comments people might have.--
> Michael Tsuk
> Compaq AlphaServer Product Development
> (508) 467-4621
**** To unsubscribe from si-list or si-list-digest: send e-mail to
[email protected]. In the BODY of message put: UNSUBSCRIBE
si-list or UNSUBSCRIBE si-list-digest, for more help, put HELP.
si-list archives are accessible at http://www.qsl.net/wb6tpu
****
This archive was generated by hypermail 2b29 : Thu Jun 21 2001 - 10:11:17 PDT