[SI-LIST] Re: New blog posting: Four things that drove me nuts about IBIS 5.1

  • From: <colin_warwick@xxxxxxxxxxx>
  • To: <twesterh@xxxxxxxxxx>
  • Date: Sat, 20 Oct 2012 09:39:57 -0600

Pressed "send" too soon...  Sorry...
The thing I forgot to add is that I believe a lot of the confusion around this 
topic is the word "mode" itself in the term "dual-mode model."

Here's the reason. We often use the word "mode" in connection with simulator 
mode (e.g. bit-by-bit mode or statistical mode). The word "mode" in "dual-mode 
model" sparks the incorrect conclusion that "a dual mode model is necessary to 
take advantage of both modes of the simulator" in the mind of the uninitiated. 
(And I included myself in that category, before Fangyi patiently explained to 
me that an impulse-only model can run perfectly well in both modes.)

A clearer term might be "dual-function model" or  something...

Then we'd have a narrative like:

1) A GetWave-function-only model can run only in bit-by-bit mode.

2) An impulse-function-only model can run in both modes, and the impulse 
function is used in both modes.

3) A dual-function model can also run in both modes, but the GetWave function 
is used in bit-by-bit mode, and the impulse function is used in statistical mode

-----Original Message-----
From: WARWICK,COLIN (A-Americas,ex1) 
Sent: Saturday, October 20, 2012 10:46 AM
To: 'Todd Westerhoff'
Cc: 'si-list@xxxxxxxxxxxxx'
Subject: RE: [SI-LIST] Re: New blog posting: Four things that drove me nuts 
about IBIS 5.1

Hi Todd,

Thanks for you feedback. I agree with you 100%.

I said "buyer beware" because I've come across several model consumers (and 
even a few model builders) who were unaware of the trade-offs you mention, so I 
wanted to highlight them in the app note.

The fact is that if the model builder provides GetWave it implies to me that 
the IC has non-linearity and/or time-varying (adaptive) behavior, because, if 
the IC were LTI, the model builder could have captured the behavior completely 
with the impulse response only. Therefore, if the model builder provides both a 
GetWave and an impulse response, in my mind the model consumer should be aware 
that that impulse response is probably some kind of approximation. For example, 
an adaptive peaking filter of the type you mention may change dynamically in 
the GetWave code used during a bit-by-bit simulation, but its response will be 
static in the output of the impulse response code used in a given statistical 
simulation.

In summary, Agilent supports such models in both our model building and our 
model consuming EDA tools. When used with the engineering judgment you mention, 
we believe they have great value.

Best regards,

-- Colin Warwick

Product Manager for High Speed Digital, Agilent EEsof EDA

...feeds blog @ http://Signal-Integrity.TM.Agilent.com/feed/

...tweets @signalintegrity  

-----Original Message-----
From: Todd Westerhoff [mailto:twesterh@xxxxxxxxxx] 
Sent: Saturday, October 20, 2012 12:22 AM
To: WARWICK,COLIN (A-Americas,ex1)
Cc: si-list@xxxxxxxxxxxxx
Subject: Re: [SI-LIST] Re: New blog posting: Four things that drove me nuts 
about IBIS 5.1

Colin,

In your table 1, you imply that an AMI model that processes both impulse 
responses (Init) and waveform data (Getwave) is risky, with a "Buyer Beware" 
note.

You've correctly identified that generating a good LTI approximation of NLTV 
behavior requires skill on the part of the model developer - well beyond a 
straightforward adaptation of internal Matlab code a semiconductor company may 
already have on hand. What you haven't done, is identify the benefit (why would 
anyone want to do this in the first place?) and whether people have already 
done it.

Let's start with the benefit.  Statistical simulations run 30-300x faster than 
their time-domain (bit by bit) counterparts AND predict very low probability 
events with much more precision. In the lead-in to your paper, you mention the 
need to perform hundreds of simulations to explore a design space. In our 
experience, that's actually a low number - it's more like thousands to tens of 
thousands of simulations. One of our customers estimated the time needed to run 
design space explorations in time-domain (bit-by-bit) mode as 6 months on a 64 
processor server farm.  That's just not tenable.

A good dual-mode model (your "Buyer Beware" case) allows Design Space 
Exploration to be run in Statistical Mode.  Thousands of cases can be explored 
quickly in Statistical Mode to see which cases are worth investing the time to 
run in Time-Domain Mode.  In the webinar we presented with Xilinx in 2009 (or 
2010, I don't have access to my records at the moment), we showed how 500 
Statistical Simulations could be run in about 15 minutes to narrow down a 
design space.  Those simulations used a dual-mode AMI RX with both DFE and an 
adaptive peaking filter. The simulations were run on a single desktop computer 
with 4 non-hyperthreaded cores. The equivalent bit by bit analysis (as your 
paper suggests) would likely have taken 8 hours or more.

It's true that coding a dual-mode AMI model for a device with nonlinear and 
adaptive behavior takes talent ... still, there are plenty of such models out 
there.  We've been delivering dual-mode AMI models and AMI model development 
tools since 2008. We correlate the results of Statistical and Time-Domain 
simulations on every model we create.

It's also true that users need to understand the limits of the two simulation 
approaches, and to use good engineering judgement when reviewing simulation 
results.  When you consider the impact that reducing simulation time by 30-300x 
can have on a design cycle, we believe the extra effort required to produce a 
good dual-mode model is well worth it.

Todd.

Todd Westerhoff 
VP, Software Products 


Signal Integrity Software Inc. â www.sisoft.com 
6 Clock Tower Place â Suite 250 â Maynard, MA 01754 
(978) 461-0449 x24 â twesterh@xxxxxxxxxx 

â I want to live like that â 

-Sidewalk Prophets 




----- Original Message -----
From: "colin warwick" <colin_warwick@xxxxxxxxxxx>
To: si-list@xxxxxxxxxxxxx
Sent: Friday, October 19, 2012 2:48:23 PM
Subject: [SI-LIST] Re: New blog posting: Four things that drove me nuts about 
IBIS 5.1

Hi si-listers,

A couple of people asked for a clarification about what combinations 
(statistical versus bit-by-bit mode, with the four true/false model flags) were 
valid, and what happened in each of the resulting 32 cases. I tried to capture 
all this in a table a placed it in the updated posting here:

http://signal-integrity.tm.agilent.com/2012/four-things-that-drove-me-nuts-about-ibis-5-1/

Feedback/corrections welcome!

Best regards,

-- Colin Warwick

Product Manager for High Speed Digital, Agilent EEsof EDA

...feeds blog @ http://Signal-Integrity.TM.Agilent.com/feed/

...tweets @signalintegrity

-----Original Message-----
From: WARWICK,COLIN (A-Americas,ex1) 
Sent: Tuesday, September 25, 2012 3:06 PM
To: 'si-list@xxxxxxxxxxxxx'
Subject: New blog posting: Four things that drove me nuts about IBIS 5.1

Hi si-listers,

My new blog posting has the pseudo-provocative title "Four things that drove me 
nuts about IBIS 5.1." 

http://signal-integrity.tm.agilent.com/2012/four-things-that-drove-me-nuts-about-ibis-5-1/

It's about my struggle to understand IBIS AMI in general and the so-called 
"flow BIRD" (BIRD 120) in particular.

Hope it helps!

-- Colin Warwick
Product Manager for High Speed Digital, Agilent EEsof EDA ...feeds blog @ 
http://Signal-Integrity.TM.Agilent.com/feed/
...tweets @signalintegrity

------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field

or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field


List forum  is accessible at:
               http://tech.groups.yahoo.com/group/si-list

List archives are viewable at:     
                //www.freelists.org/archives/si-list
 
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu
  
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field

or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list

For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field


List forum  is accessible at:
               http://tech.groups.yahoo.com/group/si-list

List archives are viewable at:     
                //www.freelists.org/archives/si-list
 
Old (prior to June 6, 2001) list archives are viewable at:
                http://www.qsl.net/wb6tpu
  

Other related posts: