Just a minor correction regarding:
· Package model description in IBIS
o Allows finer grain description of detailed models
§ This can be multiple lumped RLC models or multiple Tlines or any combination.
The [Package Model] / [Define Package Model] keywords in
IBIS allow a single lumped RLC matrix, or a multi-segment
T-line modeling approach.
Arpad
===========================================================
From: owner-ibis-users@eda.org [mailto:owner-ibis-users@eda.org] On Behalf Of Doug Burns
Sent: Thursday, June 02, 2011 6:54 AM
To: bradb@sigrity.com; 'Scott McMorrow'
Cc: 'Ray Anderson'; 'ibis-users'
Subject: RE: [IBIS-Users] Requesting for some help / information regarding the use of IBIS in SI simulation.
Brad,
I don't think we really disagree, but let us simplify the statements
Given the model you have, what would you agree with:
· Single RLC Model ala Simple IBIS
o This does not model the resonances
o This provides a large discontinuity to a high speed edge
· Simple Tline model using RLC Characteristics
o This does not model resonance
o Better match to actual (ie less ringing than Lumped model)
§ Better does not mean accurate or correct, just closer to the desired than a lumped model
· Package model description in IBIS
o Allows finer grain description of detailed models
§ This can be multiple lumped RLC models or multiple Tlines or any combination.
§ In the limit, a Tline segment can be modeled as multiple RLC pieces to represent a distributed nature
o Will model resonances you speak about at least at first order
I assume that you will agree with all of the above.
Therefore, what I am saying is that interconnect, by its very physical nature represents a distributed RLC structure. Modeling any segment that is long compared to the edge rate of the signal is better modeled as a Tline than a lumped model such that the distributed nature of the segment can be captured.
For the example you cite, a simple RLC does not model the behavior and there is no rational to believe just making the model into a tline model would fix that issues. However, for a multi-segment model using lumped values for each segment, my experience is that using tline models for all segments that have length greater than the 1/3 to 1/2 the rise time (Almost always the case with high speed signals), that a tline description is better than the RLC description. This is true even in the multi-segmented implementation
Doug
-----Original Message-----
From: Brad Brim [mailto:bradb@sigrity.com]
Sent: Thursday, June 02, 2011 5:55 AM
To: 'Doug Burns'; 'Scott McMorrow'
Cc: 'Ray Anderson'; 'ibis-users'
Subject: RE: [IBIS-Users] Requesting for some help / information regarding the use of IBIS in SI simulation.
hi guys,
Guess we'll agree to disagree.
I have experience with any number of simple packages that violate your
"better" supposition for a uniformly distributed TL model. Why would this
occur, because I completely agree with you that it is counter intuitive.
From a package I recently characterized in some detail for system SI/PI
analysis; let's say the net behaves as static RLC to 2Ghz, then at about
4GHz there is some form of resonance in the signals or the planes. There's a
suckout in the response and then at higher frequencies it behaves more like
the TL you guys propose. If you assume the TL, then you completely miss this
behavior, and I do mean "completely". For this DDR bus you probably don't
care about the spectrum above 8GHz where it is again more TL like. But you
certainly DO care about the spectrum between 2Ghz and 6Ghz where it looks
nothing like a TL. If I choose properly a TEE or PI circuit to model the
behavior in this mid frequency range (since either works to 2Ghz) then you
get a much better transient system-level simulation than using your proposed
TL model. This is true even though the TEE/PI response at higher frequencies
is much worse than the uniform TL model. There is energy in the mid spectrum
but not in the high spectrum for this application. If I use a multi-stage
RLC model then select the element values to fit the broadband response then
I can get match at both mid and high frequencies to well 10Ghz and make all
of us happier (but maybe not to 40GHz).
This is NOT a corner case. I see it in many of the package I characterize.
If I was trying to do a best fit to 40Ghz for this same package because my
signals had energy spread out that broadband, then it would probably be
better to use the uniform TL model. However, not all applications call for
it and the system is much better characterized with a model more correct
where the signals actually have energy.
cheers,
-Brad
> -----Original Message-----
> From: Doug Burns [mailto:dburns@sisoft.com]
> Sent: Wednesday, June 01, 2011 6:55 AM
> To: 'Scott McMorrow'; bradb@sigrity.com
> Cc: 'Ray Anderson'; 'ibis-users'
> Subject: RE: [IBIS-Users] Requesting for some help /
> information regarding the use of IBIS in SI simulation.
>
> All,
>
> I have to agree with Scott on this. For the type of packages
> as outlined by Scott, if one were to take the detailed model
> from a simulator and decompose it to a RLC format, the best
> correlation between the detailed model and the simplified
> model will occur when the simplified RLC is treated as a
> distributed structure such as a transmission line. Like
> Scott, we have proven this multiple time on a wide variety of
> package styles (leadframe through 24+ layer packages). There
> is no disagreement that a detailed model will be better,
> especially for high frequencies, but we must work within the
> limits of what IBIS can support. The mode detailed that
> package description is in IBIS, the better estimation we can
> make, but using lumped data for high speed signals is just
> causing an issue where one really will not exist. EDA tools
> need to make the best estimation of the available data to
> predict performance.
>
> Doug
>
> Douglas Burns
> Signal Integrity Software, Inc
> 6 Clock Tower Place
> Maynard, MA 01754
> 978-461-0449 x14
>
>
> -----Original Message-----
> From: owner-ibis-users@eda.org
> [mailto:owner-ibis-users@eda.org] On Behalf Of Scott McMorrow
> Sent: Wednesday, June 01, 2011 9:33 AM
> To: bradb@sigrity.com
> Cc: 'Ray Anderson'; 'ibis-users'
> Subject: Re: [IBIS-Users] Requesting for some help /
> information regarding the use of IBIS in SI simulation.
>
> Brad
>
> It is certainly your prerogative to disagree with me. In the
> context of the original question, we are not talking about
> randomly distributed RLC, but rather package connections from
> board to silicon, which is the case in most IBIS models.
> Given that, we know the general topology.
> Whether the package uses lead frame, 2 layer FC packaging,
> 2/4 layer organic wire bond packaging, or multilayer buildup
> packaging, a single path from die to ball looks remarkably
> similar to cascaded transmission lines. Even bond wires look
> like transmission lines.
>
> If I'm given a single RLC for an IBIS model of a
> conventionally packaged silicon device , then I can guarantee
> you that the simulation will be more faithful to measurements
> when the transmission line equivalent is used, rather than
> lumped elements. Been there, done the measurements many,
> many times. When multiple sections are specified in the IBIS
> .pkg format, then there is no ambiguity, a non-zero RLC
> section length defines a distributed element, and a zero
> length section defines a lumped element. The simulator
> should model it as such.
>
> Having done correlation of package modeling to measurements
> from DC to 40 GHz on a 12-layer build-up with near zero
> error, I'm well aware of the issues involved with using the
> wrong modeling method, and poor material modeling. I agree
> with you that a single RLC is an extremely poor approximation
> of any complex package interconnect, or any interconnect for
> that matter. However, given that a user is presented with an
> RLC that was extracted with a quasi-static solver, as is the
> case with most IBIS models, I maintain that the result will
> be closer to reality when transformed into the distributed
> transmission line equivalent. I've seen way too many cases
> where simulators used a lumped RLC directly, grossly over
> predicting ringing at the die and in the channel, and
> confirmed by physical measurements.
>
> When we get into full-wave and hybrid full-wave AC modeling
> of packages there are a host of other issues that we can
> discuss, such as accurate dielectric modeling, accurate
> surface roughness modeling, as-designed vs. as-built package
> geometries, +/- 15% impedance variation due to manufacturing
> process and electromagnetic non-locality. But that is for
> another time.
>
>
> best regards,
>
> Scott
>
>
> Scott McMorrow
> Teraspeed Consulting Group LLC
> 121 North River Drive
> Narragansett, RI 02882
> (401) 284-1827 Business
> (401) 284-1840 Fax
>
>
> TeraspeedR is the registered service mark of Teraspeed
> Consulting Group LLC
>
>
> On 6/1/2011 1:53 AM, Brad Brim wrote:
> > hello Scott,
> >
> > I don't agree with you. Such simulators (or model extractors which
> > write
> out
> > such "postprocessed" model data) are no more assured accurate than
> simply
> > applying or writing out the base RLC data.
> >
> > Most package extractors, especially ones you may refer to as
> > applicable "from the beginning", are based on static RLC
> simulation in
> > which a
> single L
> > and C are extracted at exactly DC from two distinct simulations;
> > magnetostatic and electrostatic respectively. One has NO CLUE how to
> form
> > the equivalent circuit given only the static lumped L and C
> computed
> > independently. Of course, as asked and discussed by Ray, if
> the user
> > has insight from the physical structure then a more broadband model
> > MIGHT be generated, but there is no guarantee!
> > One person assuming a PI circuit (with the C split equally
> in half -
> > or
> even
> > unequally - and the L in the middle) and another person
> assuming a TEE
> > circuit (with the L split and the shunt C in the middle) are no more
> correct
> > than the other. One has NO IDEA of the high frequency
> behavior of the
> > circuit (either just above the quasistatic limit where the likely
> spectral
> > content is relevant or higher where it is less likely relevant)
> > without
> some
> > form of AC high-frequency simulation or measurement. The
> behavior at
> > "infinite" frequency in meaningless and to argue that a uniformly
> > distributed multi-stage ladder network is valid has no more
> justification
> > than another person arguing it should me a lowpass network with all
> > the inductance on the die side because of wirebond inductance. This
> > argument
> has
> > blindly been accepted and argued for many years. It has been
> > unjustified
> and
> > wrong, as you would say, "from the beginning".
> >
> > Want a valid model at higher frequencies, then extract AC
> information
> from
> > which to generate it. One's guess or intuitive argument without AC
> > information to back it up is only a guess, as Andy stated. The
> > classical uniformly distributed ladder network is a
> deceptively simple
> > and
> intuitively
> > pleasing assumption that is easy to understand and unfortunately
> accepted by
> > way too many people in the industry. This supposed more
> accurate guess
> has
> > failed in too many cases and can even be worse than a
> simple totally
> > unbalanced low pass filter assumption over the frequency
> spectrum of
> > relevance.
> >
> > Many commercially available RLC extractors continue to make this
> assumption
> > today. Users are placed at significant risk because they falsely
> > believe
> the
> > model is in some sense "broadband" or "wideband". It is a
> guess with
> > no guarantee to be any better than any other assumed more
> distributed
> > equivalent circuit for which there is no theoretical or physical
> > justification.
> >
> > If one starts with AC data, assumes an equivalent circuit and *then*
> selects
> > element values in such to fit the data then there is reason to argue
> this is
> > a more valid model at higher frequencies. One cannot do
> this from only
> DC
> > terminal-based data. Not that it's difficult, it is impossible.
> >
> > best regards,
> > -Brad
> >
> >
> >> -----Original Message-----
> >> From: owner-ibis-users@eda.org
> >> [mailto:owner-ibis-users@eda.org] On Behalf Of Scott McMorrow
> >> Sent: Tuesday, May 31, 2011 5:04 PM
> >> To: Ray Anderson
> >> Cc: ibis-users; Ray Anderson
> >> Subject: Re: [IBIS-Users] Requesting for some help / information
> >> regarding the use of IBIS in SI simulation.
> >>
> >> Since the beginning of IBIS, the most accurate simulators
> would turn
> >> the RLC package parasitics into a transmission line
> equivalent. In
> >> lieu of any additional information it is the most
> reasonable thing to
> >> do, since packages can be approximated as transmission lines.
> >>
> >>
> >> Scott McMorrow
> >> Teraspeed Consulting Group LLC
> >> 121 North River Drive
> >> Narragansett, RI 02882
> >> (401) 284-1827 Business
> >> (401) 284-1840 Fax
> >>
> >>
> >> TeraspeedR is the registered service mark of Teraspeed Consulting
> >> Group LLC
> >>
> >>
> >> On 5/31/2011 7:59 PM, Ray Anderson wrote:
> >>> If you are using extracted RLC data to manually set up
> the package
> >>> model you can certainly make intelligent decisions on the
> >> topology of
> >>> the model and the distribution of the RLC parasitics.
> >> However if you
> >>> are depending on a simulator to read the data from a IBIS
> >> .pkg format
> >>> file (or the [PIN] data) and set up a model for you then
> >> your results
> >>> may differ from simulator to simulator.
> >>>
> >>> -Ray
> >>> Xilinx Inc.
> >>>
> >>>
> >>> -----Original Message-----
> >>> From: owner-ibis-users@eda.org
> [mailto:owner-ibis-users@eda.org] On
> >>> Behalf Of Andrew Ingraham
> >>> Sent: Tuesday, May 31, 2011 4:48 PM
> >>> To: ibis-users
> >>> Subject: Re: [IBIS-Users] Requesting for some help / information
> >>> regarding the use of IBIS in SI simulation.
> >>>
> >>> Yes.
> >>>
> >>> I am usually tempted to start with an evenly distributed
> >> RLC model ...
> >>> unless I have reason to think that the distribution is
> >> otherwise. For
> >>> example, wirebonds might have most of the inductance and
> >> resistance,
> >>> whereas the leadframe may have more capacitance, which
> >> might influence
> >>> one to imbalance those factors unequally between die end
> >> and pin end.
> >>> But without hard data to back it up, that is just guessing.
> >>>
> >>> Andy
> >>>
> >> --
> >> This message has been scanned for viruses and dangerous content by
> >> MailScanner, and is believed to be clean.
> >>
> >>
> --------------------------------------------------------------------
> >> |For help or to subscribe/unsubscribe, e-mail
> majordomo@eda-stds.org
> >> |with the appropriate command message(s) in the body:
> >> |
> >> | help
> >> | subscribe ibis<optional e-mail address, if different>
> >> | subscribe ibis-users<optional e-mail address, if different>
> >> | unsubscribe ibis<optional e-mail address, if different>
> >> | unsubscribe ibis-users<optional e-mail address, if different>
> >> |
> >> |or e-mail a request to ibis-request@eda-stds.org.
> >> |
> >> |IBIS reflector archives exist under:
> >> |
> >> | http://www.eda-stds.org/pub/ibis/email_archive/ Recent
> >> | http://www.eda-stds.org/pub/ibis/users_archive/ Recent
> >> | http://www.eda-stds.org/pub/ibis/email/ E-mail
> since 1993
> >>
>
> --
> This message has been scanned for viruses and dangerous
> content by MailScanner, and is believed to be clean.
>
> --------------------------------------------------------------------
> |For help or to subscribe/unsubscribe, e-mail majordomo@eda-stds.org
> |with the appropriate command message(s) in the body:
> |
> | help
> | subscribe ibis <optional e-mail address, if different>
> | subscribe ibis-users <optional e-mail address, if different>
> | unsubscribe ibis <optional e-mail address, if different>
> | unsubscribe ibis-users <optional e-mail address, if different>
> |
> |or e-mail a request to ibis-request@eda-stds.org.
> |
> |IBIS reflector archives exist under:
> |
> | http://www.eda-stds.org/pub/ibis/email_archive/ Recent
> | http://www.eda-stds.org/pub/ibis/users_archive/ Recent
> | http://www.eda-stds.org/pub/ibis/email/ E-mail since 1993
>
-- This message has been scanned for viruses and dangerous content by MailScanner <http://www.mailscanner.info/> , and is believed to be clean. -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean. -------------------------------------------------------------------- |For help or to subscribe/unsubscribe, e-mail majordomo@eda-stds.org |with the appropriate command message(s) in the body: | | help | subscribe ibis <optional e-mail address, if different> | subscribe ibis-users <optional e-mail address, if different> | unsubscribe ibis <optional e-mail address, if different> | unsubscribe ibis-users <optional e-mail address, if different> | |or e-mail a request to ibis-request@eda-stds.org. | |IBIS reflector archives exist under: | | http://www.eda-stds.org/pub/ibis/email_archive/ Recent | http://www.eda-stds.org/pub/ibis/users_archive/ Recent | http://www.eda-stds.org/pub/ibis/email/ E-mail since 1993Received on Thu Jun 2 06:48:52 2011
This archive was generated by hypermail 2.1.8 : Thu Jun 02 2011 - 06:49:50 PDT