The previous message and attachment, including subject line, had several
significant errors related to dates of previous meetings. This version has
been corrected.
(attaching a text version of the minutes for ease of archiving)
======================================================================
IBIS INTERCONNECT TASK GROUP
http://www.ibis.org/interconnect_wip/
Mailing list:
ibis-interconnect@xxxxxxxxxxxxx<mailto:ibis-interconnect@xxxxxxxxxxxxx>
Archives at //www.freelists.org/archive/ibis-interconn/
======================================================================
Attendees from March 9 Meeting (* means attended at least using audio)
ANSYS Curtis
Clark*
Cadence Design Systems Bradley Brim*
Cisco David
Siadat
Intel Corp. Michael
Mirmak*
Keysight Technologies Radek Biernacki*
Mentor Graphics Arpad Muranyi*
Micron Technology Justin Butterfield*,
Randy Wolff*
SAE ITC Maureen
Lemankiewicz, Logen Johnson
Signal Integrity Software Walter Katz*, Mike
LaBonte*
Teraspeed Labs Bob Ross*
University of Aveiro in Portugal Wael Dghais
Michael Mirmak convened the meeting. No patents were declared. Michael
called for comments on the minutes of both the Feb. 24 and March 2 meetings.
No comments were made. Brad Brim moved to approve the minutes of both the Feb.
24 and March 2 meetings. Curtis Clark seconded the motion. The minutes were
approved without objection.
No opens were raised.
Brad summarized offline discussions regarding the shortcut approach. He noted
that Walter Katz made several good points regarding mistakes that can be made
with a subcircuit wrapper and also a shortcut. Walter agreed with Brad's
summary, stating that both approaches can be abused. Radek Biernacki noted
that there may be additional restrictions like global node 0, but in terms in
functionality, there should be no difference between a wrapper and the
shortcut. Brad replied that, if you use the wrapper, the designer will know
exactly how it will be used in the simulator (it's a netlist). Radek responded
that we are trying to control netlisting of the subcircuit. Brad replied that
we want to know the correct connections for *every* circuit simulator out
there. We put a wrapper around the data when we want to be sure.
Brad and Walter agreed that "node" and "terminal" are not synonymous. Next
time, Brad will complete his presentation and the team will resolve the
shortcut approach.
Walter summarized changes he proposed on the reflector to the terminal
description in the Interconnect proposal. An S-parameter can be used directly,
as a shortcut if the signal_name on the reference terminal shall be a
signal_name on any of the port terminals. "Return port" rather than "return
terminal" should be used; Eric Bogatin does not like the word "reference", as
opposed to "terminal". The second rule is that all of the ports shall
reference the same signal_name as the reference terminal. These rules are
"good" rules for S-parameters generally, but the same errors being checked here
could be made within an IBIS-ISS circuit.
Arpad Muranyi asked about the phrase, "Terminal N+1 (reference terminal) shall
be connected to a Pin with a model_name of POWER or GND". How can one connect
reference of two-terminal differential interconnect? How about 3-terminal
power/ground/signal? Walter replied that, if you can't satisfy the rule, then
you cannot use a wrapper. Arpad suggested his second example violates this
rule.
Radek reviewed his very simple RL example (a series RL path with no reference
terminal). If the purpose of a shortcut is to have a single S-element in the
specification, rather than in the ISS subcircuit separate file, then having an
extra rule is superficial. The N+1 approach makes sense here. If you use only
N terminals, you must use global 0 and we don't want that. Walter replied that
the shortcut is a way to address how S-parameters are being used today.
Brad used the analogy of a small town in Montana, with a bypass that supports
cars but not trucks. Cars can go quickly through, trucks must take a longer
route. The solution is by definition and applies to the shortcut; not all
circuits can be described through the shortcut.
Arpad asked about a 3 pin, 3 pad circuit with separate reference, in an s6p.
Walter asked how one would measure this circuit. Where would you put the
probes? Brad suggested that one cannot measure a "reference" with respect to
global node 0. Your "universe" has 6 nodes, because a reference would use a
7th node. Radek replied that this is a separate issue. If you have 2N
terminals, you can have 2N-1 ports, and everything is fine.
Bob Ross asked if there is a ground (e.g., for s6p) assumed for S-parameter
extraction. Is the ground implied? Walter replied that the problem is the use
of words like "ground"; think about the second terminal of probe. Arpad asked
if it is possible to remove the phrase "shall be connected to a Pin with a
model_name of POWER or GND". Radek replied that, if Walter's example is
understood, it can be removed. Arpad asked whether this approach is for the
EDA vendor, or the model-maker. Radek replied that the model-maker provides
all the details.
Michael asked if the reference terminal, node, and port clarifications were
already done for Touchstone and IBIS-ISS already.
Arpad responded that ports are ports in Touchstone; no connectivity is defined.
Radek added that IBIS-ISS took its connectivity from commercial SPICE.
Arpad asked whether connectivity assumes the same (reference) nodes on both
sides. Walter replied that there are two uses of S-parameters. This describes
traditional use in signal integrity, not power integrity, where you supply a
specific voltage to each buffer.
Arpad asked, if we cannot use the same reference on both sides, then how can we
instance an S-parameter with one reference? Radek replied that this question
is asked all the time, and the 2N reference is allowed but not recommended.
Arpad followed up by asking if the 2N reference isn't recommended, why isn't
ground allowed? Radek replied that one can be limited in available data and be
effectively underdetermined. If you use network analyzer, you have test
fixtures at opposite ends. Walter added that return terminals get "sucked up"
by node zero, and everything is handled properly.
Arpad asked whether the "shall be" phrase can removed and the issue left up to
the EDA vendor implementing the circuit. Walter and Radek replied that this is
up to the model maker, not the EDA vendor. Walter added that accurate power
integrity analyses cannot use the N+1 method. Arpad asked whether the number
of ports must equal the number of pins and the number of pads.
Radek summarized several points, which will be addressed next time in detail:
1) N-port usage is redundant. This has been discussed extensively and is
the focus of meeting
2) What is syntax for unconnected terminals? Something unconnected should
be open-circuit. The meaning of the termination may change with context.
3) The reference terminal must not be unconnected; the reference must be
explicitly identified.
Arpad stated that, if "N+1 Terminals" language remains in specification, then
he accepts a requirement of connecting POWER/GND as a reference terminal.
Bob asked about RLC circuits. Brad replied that converting RLC to S-parameters
causes loss of information. Bob suggested that in a circuit with two ports,
two terminals plus reference, which side contains the reference? Radek
responded that two ports with four terminals is a problem in terms of
under-determination. For three terminals, you are covered.
Arpad asked for confirmation that a 3-pin, 3-pad package uses 5 port data.
Radek confirmed this, stating that one port is the reference. Bob added that
one never get 5 ports in practice. Brad asked how many ports one would see: 4,
5, or 6?
Brad proposed that the team next time discuss the intent of the shortcut. He
suggested that the objective was to automate (the "A" in "EDA" is for
automation) so that wrapping the S-parameters in a subcircuit isn't needed.
The shortcut approach will not work without some restrictions.
Arpad moved to adjourn. Brad seconded the motion. The meeting adjourned.
Attachment:
image002.jpg
Description: image002.jpg
