[ibis-macro] Re: BIRD 158.2
- From: "Todd Westerhoff" <twesterh@xxxxxxxxxx>
- To: "'IBIS-ATM'" <ibis-macro@xxxxxxxxxxxxx>
- Date: Wed, 22 May 2013 11:14:29 -0400 (EDT)
Ambrish,
Are you forgetting that you were one of the people who requested that we
submit BIRD 158 in the first place? You knew perfectly well what it
entailed when you asked us to write it.
What has changed?
Todd.
Todd Westerhoff
VP, Software Products
Signal Integrity Software Inc. . www.sisoft.com
6 Clock Tower Place . Suite 250 . Maynard, MA 01754
(978) 461-0449 x24 . twesterh@xxxxxxxxxx
"I want to live like that"
-Sidewalk Prophets
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Ambrish Varma
Sent: Wednesday, May 22, 2013 10:50 AM
To: 'IBIS-ATM'
Subject: [ibis-macro] Re: BIRD 158.2
Hello Walter,
I wanted to point out some fundamental issues about your BIRD that you
rightly characterize a 'Shortcut' BIRD to reference touchstone models for
analog buffers through the .ami file.
1) First and foremost, there is already a perfectly legitimate way of
accomplishing everything that this proposal wants to achieve (through BIRD
160).
2) The proposal is based on some 'canned' circuits that model the analog
buffer. This cct represents a particular design but by no means universal.
Any change in these circuits (and there are and will be changes, for ex,
circuits with t-coils ) will mean that either the shortcuts are
meaningless and the model maker has to end up writing a sub-circuit to be
referred through the method prescribed in BIRD 160 OR write up another
BIRD to add/edit the hard coded circuit that the EDA tool has implemented.
This just means more churn in the tool and more meetings to discuss new
BIRDs.
3) This method works only for AMI and any need to include a touchstone
file for legacy simulations will require a sub-circuit.
4) It blurs the boundary between the analog model and the algorithmic
model by including sections of the analog model in the .ami file -
considering there is a perfectly legal and elegant method of achieving the
same outside the .ami file.
What we have done with the analog BIRD is to provide the means for a
sustainable, long term solution that will give the user/model maker the
flexibility and the EDA tool a level of stability for the foreseeable
future. Any 'shortcut' implementation is only going to be, by nature,
short term.
You had suggested template circuits before. I suggest we should go back to
that and work on a library of sample circuits. This will be outside of
the IBIS standard and we can add to the library as and when IP vendors
come up with new IO structures. We think this is a much more valuable
contribution that could be immediately leveraged by the IBIS AMI
community. This will be a much better use of our time instead of constant
perturbation of the spec.
Again, the cct in this BIRD is only particular to a design and there is
nothing universal to it that should be included in a standard.
Best Regards,
Ambrish.
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Walter Katz
Sent: Monday, May 20, 2013 5:20 PM
To: Arpad_Muranyi@xxxxxxxxxx; 'IBIS-ATM'
Subject: [ibis-macro] Re: BIRD 158.2
Arpad,
First, I am including 158.4 which has the graphic for the step response
voltage source as Bob requested.
You are confusing the waveform input to Tx GetWave which is differential
(centered around 0.0V) with the single ended half of a differential
signal. In this example (as in most differential signaling), both the
non-inverting and inverting side of a differential typically go between
~0.0 Volts and ~PuRef. These are the SINGLE ENDED voltage swings. The
differential voltage goes between -PuRef and PuRef. In fact the common
mode voltage goes away in AMI modeling because we only deal with
DIFFERENTIAL impulse responses and DIFFERENTIAL waveforms.
Again you are comparing apples and oranges. In differential signaling
single ended waveforms may and in fact usually do have common mode
components. AMI modeling currently ignores the common mode component in
its algorithmic modeling.
Of course there is common mode to differential conversion in the analog
(e.g. step response simulation) so common mode must be accommodated in the
single ended signals) and in generating the impulse response of the
channel.
AMI makes the basic assumption that the Rx does an excellent job of common
mode rejection. I cannot attest to the validity of this assumption, other
than this is the assumption made in many IC Vendor internal analysis
tools.
Walter
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Monday, May 20, 2013 4:06 PM
To: 'IBIS-ATM'
Subject: [ibis-macro] Re: BIRD 158.2
Walter,
That added sentence doesn't change anything in the
voltage values you defined for the two sources.
"The step response stimulus is a differential step response waveform which
switch from a logic level 0 to a logic level 1."
The next sentence still defines your differential
stimulus with a DC component of Tx_V/2 away from
the axis because the high and low levels of the
sources are Tx_V and 0 V, which averages to Tx_V/2.
So the question is this, do you intentionally want
this stimulus to have a DC component of Tx_V/2 and
want to be inconsistent with the +/-0.5 volt stimulus
defined for the AMI bit pattern?
Thanks,
Arpad
========================================================
From: Walter Katz [mailto:wkatz@xxxxxxxxxx]
Sent: Monday, May 20, 2013 2:50 PM
To: Muranyi, Arpad; 'IBIS-ATM'
Subject: RE: [ibis-macro] Re: BIRD 158.2
Arpad,
In "ANY OTHER BACKGROUND INFORMATION" I state "The rise time of the step
response voltage source should be as close to 0 as possible within the
practical limits of SPICE simulations." (I added as possible)
I changed the text at the bottom of the Transmitter Circuit to "The step
response stimulus is a differential step response waveform which switch
from a logic level 0 to a logic level 1.When logic level is 1, SRC1 V=Tx_V
and SRC2 V=0. When logic level is 0, SRC1 V=0 and SRC2 V=Tx_V. The
transition time between 0 and 1 in the two voltage sources is zero (or as
close to zero as possible within the limitations of SPICE)."
I am working with the graphic artist (Mike LaBonte) to fix the voltage
source symbols.
Walter
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Muranyi, Arpad
Sent: Monday, May 20, 2013 3:26 PM
To: 'IBIS-ATM'
Subject: [ibis-macro] Re: BIRD 158.2
Walter,
Regarding the notation of the voltage sources in the
Tx drawing, I tend to agree with Bob. I does NOT
reveal what you are stating:
"The differential waveform is Tx_V-0 = Tx_V when high and 0-Tx_V when low.
So the differential waveform is centered around 0."
Now that I know your answer, I can force myself to "see"
what you are saying in your notation "Tx_V,0" and "0,Tx_V"
if I think of the HSPICE syntax for their "PULSE source",
where the argument list of that particular source type
contains a starting voltage, a comma, and the pulse voltage
among many other arguments.
Unfortunately it is not obvious that this drawing uses a
portion of the HSPICE PULSE source syntax, and an uninitiated
reader can get confused just like Bob and I got confused
reading this drawing.
The other problem is that the text below the drawing contradicts
even this interpretation of the drawing. For one, the use of
"Vdc" is inappropriate because this is not a DC source. But
more importantly, the text explicitly says that the pulse
is between zero and Tx_V volt for both sources. This doesn't
seem to support the idea that there is a differential stimulus
with levels of +/- Tx_V, centered around 0 volts.
This needs to be cleaned up.
While I am at it, how do you propose to implement a zero rise/fall
time in a time domain simulation? I think the BIRD should say
something about that too.
Thanks,
Arpad
===================================================================
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Walter Katz
Sent: Friday, May 17, 2013 9:36 PM
To: 'Bob Ross'
Cc: 'IBIS-ATM'
Subject: [ibis-macro] Re: BIRD 158.2
Bob,
Comments in-line, I sent (and attached this time BIRD 158.2) to the
reflector
Walter
From: Bob Ross [mailto:bob@xxxxxxxxxxxxx]
Sent: Friday, May 17, 2013 8:56 PM
To: wkatz@xxxxxxxxxx
Cc: 'IBIS-ATM'
Subject: RE: [ibis-macro] BIRD 158.2
Walter:
I did not see the draft BIRD158.2 attachment, but attached
are some editorial corrections to the content of BIRD158.1.
We can go over them at the ATM meeting.
The port 1 waveform is documented as
__
__| V = Tx_V,0, but should be V = 0, Tx_V
Similarly for
___
|__ should be V = Tx_V, 0
The words in the text are independent of the diagram.
(I cannot edit the diagram.)
My questions/comments are
1. I am not happy that we have to introduce a common mode
offset in contradiction to page 164, Step 4 of the Specification
which states that the input voltages are from -0.5 to 0.5.
WMK> You are comparing apples and oranges. The differential waveform is
Tx_V-0 = Tx_V when high and 0-Tx_V when low. So the differential waveform
is centered around 0.
I think it would be better to introduce Tx_Vp and Tx_Vn
where Tx_Vp defaults to 0.5V and Tx_Vn defaults to -0.5,
and the differential input is from V= Tx_Vn, Tx_Vp on port 1
and V = Tx_Vp, Tx_Vn on port 2. Then the parameters can
be passed in directly into the Converter Parameters of
BIRD160 to specify a differential stimulus that spans
-0.5 to 0.5. The common mode offset is disturbing and
assumes no common-mode to differential conversions.
Alternatively, we could enter in Tx_Vp and Tx_Vn to
match the actual voltage swing limits in the physical
Tx buffer.
Also the Definition of V_Tx is strange: "defines the
rail voltage of the I/O power supply in volts". Are we
really defining a power supply voltage or a voltage
swing limit?
Corners are already in sync, but I would rather put the
burden of syncing up Tx_Vp and Tx_Vn values on the EDA tool
than to create a technically unnecessary offset due to
weaknesses in other parts of the specification.
WMK> This is more realistic. In a normal differential driver both the
positive and negative side swing typically between 0V and PuRef. Thus one
can think of Tx_V as PuRef. When the input to the Tx is an equalized
waveform (e.g. output of Tx GetWave), then in affect the algorithmic model
is modulating PuRef. I think we have heard these exact words from David.
2. Should the Tstonefile name be changed to Tstonefile_s4p
since this if a pre-defined file for S-parameter 4-ports only?
Other predefined Tstonefile configurations are possible now
and in the future (for S, Y, Z parameters or for a
different number of ports). To "Tstonefile" seems
too generic when it applies to a specific configuration.
WMK> I used Tstonefile because this is the name that IBIS-ISS uses, I see
no need to change this.
3. Will the BIRD158 Reserved Parameters work without
BIRD160?
WMK> Yes
Bob
From: ibis-macro-bounce@xxxxxxxxxxxxx
[mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Walter Katz
Sent: Friday, May 17, 2013 3:45 PM
To: Michael Mirmak
Cc: IBIS-ATM
Subject: [ibis-macro] BIRD 158.2
MM,
I am formally submitting BIRD 158.2 to the Open Forum and requesting that
a vote be taken at the next Open Form to approve this BIR for inclusion in
the next release of IBIS.
I have made the change to specifically state that the Touchstone specified
by the reserved AMI parameter Tstonefile does not include the IBIS package
model, and described one method of generating an Impulse Response of the
channel suitable for use as input to the Tx AMI_Init function. I have not
made the change to the graphic that Bob requested because I do not know
how, because I think it is unimportant since the words in the text
describe exactly what to do, and because it can be done as part of the
editorial review. I will re-submit this BIRD if required due to any
editorial changes made during the Tuesday IBIS-ATM meeting.
Walter
Walter Katz
wkatz@xxxxxxxxxx
Phone 303.449-2308
Mobile 303.335-6156
Other related posts:
