[SI-LIST] Re: Do you ever measure any of the things you model?
- From: "Lance Wang" <lwang@xxxxxxxxxx>
- To: "'Alfred P. Neves'" <al.neves@xxxxxxxxxxx>, "'Lynne D. Green'" <lgreen22@xxxxxxxxxxxxxx>, "'Max Artusy'" <max5461@xxxxxxxxxxxxx>, <si-list@xxxxxxxxxxxxx>
- Date: Fri, 28 Mar 2008 23:54:18 -0400
Hi, Max,
This is an interesting point to SI. In the real world, the most of accuracy
issues are from models quality not simulator self. Some good guys (like Tom,
Alfred, etc.) are doing the excellent jobs to improve them. However, it is
sad to say a lot of people don't know the real facts for SI accuracy issues.
Here are two popular arguments about SI accuracy:
"It seems we have to do SI for our designs, But SI couldn't help us a lot
since it is not accurate compared with measurements."
"We don't do SI even for the GHz applications. SI is not accurate enough.
Design Guideline (Rules) is the easier and more accurate way to get our jobs
done."
Not only buffer model accuracy is an issue but also interconnect model's.
It is time to enforce the model quality now. SI is dealing with ps and mv
now.
Correctly processing your data is another part to improve your SI analysis
accuracy.
Best regards,
Lance Wang
IO Methodology Inc.
978-764-2298
FREE!!!
SignalMethTM - A Powerful Waveform Post-Processing Tool
Goto www.iometh.com <http://www.iometh.com/> for details
Some time back, couple of years ago, we (Teraspeed Consulting LLC) =
designed
an equalized twin-ax cable system using fine gauge twin-ax, equalized
Quallan RX, and drove the entire signal path with a BERT. Note that =
the
BERT was also modeled using Gaussian filters in ADS.
Tom Dagostino (Teraspeed Labs) developed a measure-based IBIS model of =
the
Quallan equalizer using a simple, but very high signal integrity board, =
and
we used Agilent ADS which made the behavioral modeling of the equalizer
characteristics, S-parameter model of the cable, and IBIS representation =
of
the input of the RX easy to perform.
We were able to achieve excellent correspondence between system =
simulation
and measurements (a few psec correspondence for a 10Meter length signal
path, see page 36) made with our DSO. The correspondence was very much =
due
to the quality of the model in that it was generated with very carefully
acquired measure-base data using a high signal integrity fixture, fast =
TDR
head, etc., =20
I checked the IBIS model separately and it looked exactly like measured =
data
in terms of not just rise/fall time in an eye diagram but also the =
pedestal
and features of the eye were captured extremely well. I have found IBIS
model accuracy to be critical, and unfortunately most of the vendor =
based
models are a bit disappointing in their representation.
Check it out:
http://www.home.agilent.com/upload/cmc_upload/All/SI3_Teraspeed_Equalizat=
ion
06.pdf.
Alfred P. Neves=A0=A0=A0=A0=A0 <*)))))><{=20
735 South East 16th Avenue
Hillsboro, OR 97123
(503) 718 7172 Office
(503) 679 2429 Mobile
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx] =
On
Behalf Of Lynne D. Green
Sent: Friday, March 28, 2008 5:17 PM
To: 'Max Artusy'; si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Re: Do you ever measure any of the things you model?
Hello, Max,
Yes, most designers (model users as opposed to model makers) know how to
test hardware. It's figuring out where and why a mismatch occurs that
drives us nuts (SPICE model? IBIS model built from SPICE? simulator?
hardware build? testing?).
IBIS models are known to be an issue.
http://www.vhdl.org/pub/ibis/summits/jan02/bell1.pdf lists some common =
IBIS
model problems. Eventually, model users learn which companies put out
useless IBIS models, and simply cross them off their list of suppliers.
In a good methodology (used by many (all) of the best IBIS providers), =
the
models and hardware are checked. Some use dedicated test boards, others
wait for prototype boards, some use both. When a problem does occur, it =
is
easily debugged.
One problem is companies that assume IBIS model creation can be handed =
off
to a newbie along with s2ibis3 or another script. These models ship =
without
running even ibischk4. The only other phase of this "methodology" =
involves
Customer Support.
Another problem is "fabless" companies where models are created from =
SPICE,
but the company does not make chips or build boards (that's for their
customers to do...). Unfortunately, model/hardware mismatches become =
more
difficult to resolve when company boundaries are crossed, although the =
IBIS
Quality Checklist has made communication of issues easier.
http://www.vhdl.org/pub/ibis/quality_wip/
Finally, some things are non-trivial to model or test, but can have
significant impact on system performance: pwr/gnd current densities near
chips having tens or hundreds of pwr/gnd pins; self-heating of an I/O =
due to
switching (data pattern); power droop and crosstalk at an I/O due to =
shared
pwr/gnd pins, crosstalk between vias in a board or package, overclocked =
I/O
switching, etc.
So we muddle along, produce our designs, and hope for better models.
Lynne
Next IBIS seminar May 1-2, Redmond WA.
"IBIS training when you need it, where you need it."
Dr. Lynne Green
Green Streak Programs
http://www.greenstreakprograms.com
425-788-0412
lgreen22@xxxxxxxxxxxxxx
-----Original Message-----
From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx] =
On
Behalf Of Max Artusy
Sent: Friday, March 28, 2008 4:32 PM
To: si-list@xxxxxxxxxxxxx
Subject: [SI-LIST] Do you ever measure any of the things you model?
I notice that you all are most concerned with your model.
Now by the scientific method, you have a hypothetical solution, produced =
by
your model.
The next step is to EVALUATE the result.
That is done with a very good scope.
No one ever seems to comment on this, leading me to suspect you only =
model.
You know they did that with the Hubble space telescope.
That technique produced one of the most expensive service calls in =
history.
Once modeled, they should have TESTED the thing.
Testing goes hand in hand with modeling, on the assumption that the =
model is
incomplete.
No measurement, saves lots of time. Any naive manager just loves to save
test time.
"Our models are perfect, you know" says he. ( He missed point of =
Scientific
method, entirely)
So what kind of scope and what kind of probing technique do you use?
How do you ground your probe setup to the PCB, as a place to start?=20
How much noise on the ground??
What is the exact construction of the probe interface to the PCB?
This all counts, in high performance probing.
Max Artusy
------------------------------------------------------------------
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 technical documents are available at:
http://www.si-list.net
List archives are viewable at: =20
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
=20
------------------------------------------------------------------
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 technical documents are available at:
http://www.si-list.net
List archives are viewable at: =20
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
=20
------------------------------------------------------------------
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 technical documents are available at:
http://www.si-list.net
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
-- Binary/unsupported file stripped by Ecartis --
-- Type: application/ms-tnef
-- File: winmail.dat
------------------------------------------------------------------
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 technical documents are available at:
http://www.si-list.net
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts:
