Arpad-
Thanks for the very clear question.
The short answer I offer to your question is that at any frequency at
which C_Comp is relevant, there is no such thing as ideal ground. So it
doesn't make sense to connect a circuit element to a node that doesn't
exist.
I resisted the temptation to respond to Walter's e-mail yesterday, but
now I'll offer a broader perspective to explain the answer above.
The debate between single point grounding and mesh grounding has raged
for decades, so I don't suppose this e-mail is going to end it. In fact,
the receiver in one of the earliest fiber optic systems (early 1980's)
attempted to use the single point ground concept, resulting in a design
that was outright silly. There are probably still a lot of power
subsystems that are being designed using the single point ground concept.
The central problem we're trying to solve is to account for the return
current of the circuit elements in an electronic circuit. If the charge
went into the circuit element somewhere, it had to come out somewhere
else. There was a time when the conventional wisdom was to *control the
path of the return current* so that it was always returned to a single
point, thus minimizing noise in the circuit. This single circuit node
was called "ground", and it's been a comforting concept ever since.
The single point ground approach to circuit design assumes that you
*can* control the path of the return current. If each circuit element is
equipped with a terminal for the return current, then one can connect a
wire to that terminal and route that wire back to a single point which
we can then call "ground". That's what we do with AC wiring in a
building, for example. Works great. Always has.
This approach breaks down, however, when the frequency gets high enough
that the inductance in the return current wires becomes unacceptable.
Then we start making wires wider, such that they merge into some form of
conducting surface. We call such a surface a ground plane; however, it
would be more accurately called a "return current plane". There are
voltage differences between different points in the return current
plane, as well as some impedance between these points; so the return
current follows the path of least resistance, a behavior that is nigh on
impossible to control. You can dub some point (on the return current
plane or not) to be "ground" but you will find that it doesn't actually
help you solve the circuit.
Above about 20kHz, skin effect forces the return current to flow on the
surface of the return current plane, and other electromagnetic effects
force the return current to flow near the signal conductor. We end up
assuming that a portion of the return current plane is part of the
circuit element, and that where circuit elements connect to each other,
part of that connection is a joining of two adjacent areas of the return
current plane. This turns out to be a very convenient convention in that
it automatically makes the return current for each circuit element match
the signal current for that circuit element.
Circuit solving at DC for circuits that include a return current plane
turns out to be a lot more subtle problem. The return current is not
constrained to flow near the signal conductors, and so the return
current plane becomes a circuit element in its own right. The only
reliable solution I've found has been to structure the equations in a
way that forces each circuit element connecting to the return current
plane to match its return current to its signal current. I tried
approaches that depended on a defining a single reference point as
"ground" and at least the approaches I tried were numerically unstable.
At least for me, the net result is that I've had to relegate "ground" to
that class of concepts that I found comforting when I was young, but
have since had to give up in order to deal effectively with a complex
World. No, Virginia, there is no "ground".
Thanks for reading this far.
Mike S.
On 06/30/2015 10:42 PM, Muranyi, Arpad wrote:
This is a request that was made today in the IBIS-ATM meeting
to all IBIS member EDA vendors. We need your reply so we can
move forward with the cleanup process in the IBIS specification
regarding "ground".
As you may know, the specification is not very consistent and
not always correct when it comes to the usage of "ground" and
its synonyms. We need to clean this up in the specification
because this situation creates confusion with the already
existing power integrity capabilities in the spec, and the
situation will only get even worse with the upcoming new
Interconnect/Package modeling proposal.
The question for which we would like to get a reply from all
EDA vendors is whether we can all agree that C_comp should not
be connected to the ideal ground (or node0) of the simulator,
but that is should be connected to the [Model]s "local ground"
(or pulldown reference).
Please reply to this email with an answer to that question.
Thanks,
Arpad
=====================================================================
---------------------------------------------------------------------
IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/
IBIS Macro reflector: //www.freelists.org/list/ibis-macro
To unsubscribe send an email:
To: ibis-macro-request@xxxxxxxxxxxxx
Subject: unsubscribe
