One problem with the "multiple pass" approach is that Model_Specific
adaptation, if turned on, gets in the way.
Bob
On Tue, Jun 20, 2017 at 11:00 AM, Bob Miller (APD) <bob.miller@xxxxxxxxxxxx>
wrote:
In the past, there has been proposal to make AMI_Init re-entrant by
specifying that if the AMI memory_handle passed in is NULL, AMI_Init starts
from scratch like today. If not NULL, AMI_Init assumes it is valid and
skips creating/initializing the memory structure.
Such an AMI_Init works already in legacy simulations with multiple EDA
platforms (many pass a NULL memory_handle in as SOP).
We did not get agreement on that, though, and when we quit pursuing BCI in
Init it went away.
Bob
On Tue, Jun 20, 2017 at 10:49 AM, Bob Miller (APD) <
bob.miller@xxxxxxxxxxxx> wrote:
Does Fangyi's proposal solve the adaptation problem (rx2 init needs the
complete upstream info in it's impulse_in)? I don't think so unless I've
forgotten something therein...
Bob
On Tue, Jun 20, 2017 at 10:32 AM, Muranyi, Arpad <
Arpad_Muranyi@xxxxxxxxxx> wrote:
Good question. Let’s discuss it in ATM today…
But there were a few “implementation options” to Fangyi’s proposed
solution too.
For example, instead of adding more IR inputs and outputs to the Init
function, we
could execute the Init functions multiple times to generate the
different IRs that
are needed for the “correct flow”. This might work without changing the
models,
as far as the IR input/output is concerned. However, we would need to
figure out
how to prevent the Init functions to initialize/optimize twice, though…
I didn’t think
all this through in detail yet. This is why we need all the experts
come together and
find a good way to do this…
The fundamental concept in Fangyi’s proposal that we need to aim to
implement is
that we need to find a way to have the Init functions generate “self
IR-s” and
“cumulative IR-s” in each channel. Without having both of these, I
don’t think we
can solve the problems we are discussing…
Thanks,
Arpad
=================================================================
*From:* Walter Katz [mailto:wkatz@xxxxxxxxxx]
*Sent:* Tuesday, June 20, 2017 10:59 AM
*To:* Muranyi, Arpad <Arpad_Muranyi@xxxxxxxxxx>;
ibis-macro@xxxxxxxxxxxxx
*Subject:* RE: [ibis-macro] Re: Response to BIRD-190
Arpad,
If we approve Fangyi’s BIRD today, what do we do in the Redriver flow
for legacy models that do not return the new IR from the AMI_Init function.
Walter
Walter Katz
wkatz@xxxxxxxxxx
Phone 303.449-2308 <(303)%20449-2308>
Mobile 303.335-6156 <(303)%20335-6156>
*From:* ibis-macro-bounce@xxxxxxxxxxxxx [mailto:ibis-macro-bounce@free
lists.org <ibis-macro-bounce@xxxxxxxxxxxxx>] *On Behalf Of *Muranyi,
Arpad
*Sent:* Tuesday, June 20, 2017 11:44 AM
*To:* ibis-macro@xxxxxxxxxxxxx
*Subject:* [ibis-macro] Re: Response to BIRD-190
Walter,
I don’t think any of us is trying to block innovation here. But there
are reasons to have
a specification and have consistency in it. We can twist the rules and
intent of a spec
only so far before chaos takes over. But in the name of serving the
user and innovation,
let’s approve and implement Fangyi’s proposal right now! This is the
only proposal at
this time that I am aware of that solves the problem as it should be
solved.
You keep saying that BIRD166 does this and that while BIRD190 doesn’t,
but you seem
to forget that it has been demonstrated that while BIRD166 may fix the
statistical flow,
it introduces problems in the time domain flow which weren’t there
before. Also, you
haven’t responded to some of the questions related to your latest
version regarding
the “8b” paragraph, which basically says that if any of the GetWave
functions are missing,
the analysis should be done with all of the Init functions. It is not
guaranteed that all
models will have Init_Returns_Impulse=true. If any of the models has a
false for that,
and any of the other models have no GetWave function, you won’t be able
to simulate
at all. BIRD166 needs some work. It simply can’t be approved as it
stands today.
Thanks,
Arpad
=====================================================================
*From:* Walter Katz [mailto:wkatz@xxxxxxxxxx ;<wkatz@xxxxxxxxxx>]
*Sent:* Tuesday, June 20, 2017 7:20 AM
*To:* Muranyi, Arpad <Arpad_Muranyi@xxxxxxxxxx>;
ibis-macro@xxxxxxxxxxxxx
*Subject:* RE: [ibis-macro] Re: Response to BIRD-190
Arpad,
As you said, I was trying to answer your question about original intent.
IBIS is a triangle with three corners:
1. EDA Tool
2. IC Vendor
3. User
We can dissect original intent, and history of a standard that is 22
years old and has gone through 11 factors of two (according to Moore’s
law). We have gone from 25 Megahertz to 56 Gigahertz. It is amazing that
2^11 =56G/25M, and now designing 116 Gigahertz for two years from now.
Ultimately, we are here to serve the User, and his requirements. It is
our duty as EDA vendors and IC Vendors to deliver models and tools that
serve Users well. This is not accomplished by dissecting original intent,
or is DFE LTI, and how LTI is it. We do not serve our Users well if we try
to block innovation that is demonstrably and successfully being used today
to design systems that are driving technology such as autonomous driving,
WebEx, Facebook, and Google.
IBIS-AMI was specifically designed to support innovation: InOut
Parameters and Model Specific Parameters. Innovation has let IBIS-AMI meet
the needs of the SerDes (and now DDR5) market with relatively few
substantive changes over a 10 year life span.
Ultimately, IBIS-AMI Users require a solution that “Predicts the
operations of a channel with sufficient accuracy to design robust products”.
We should be making sure that the IBIS standard allows IC Vendors to
produce “Compliant” models that allow EDA vendors to deliver a software
product that Users can use to address this requirement. I believe BIRD 166
does that, I believe BIRD 190 does not. I also support the extensions that
Keysight has proposed does additional things in line with this fundamental
User requirement.
Walter
Walter Katz
wkatz@xxxxxxxxxx
Phone 303.449-2308 <(303)%20449-2308>
Mobile 303.335-6156 <(303)%20335-6156>
