> > Hi, Larry. > > I thought your post is excellent! Our experience is similar to yours > in that it is possible to build "dirt" simple models to achieve great > hardware to model correlations. Some of our results were published in a > recent DesignCon 2004 paper. > > I would add my "two cents" to the already long discussions as in the > following: > > A. Accuracy is an abstract term. > One may refer to different things even using the same term. For > example, is accuracy referring to the worst case peak to peak noise? > Where is this noise measured, at package pin relative local ground or at > transistors relative its local ground or global ground? The accuracy > could also be refer to the ability to predict the voltage and timing > loss of the driver/receiver in the presence of power/ground noise. If we > debate about the accuracy, it would be good if we can be specific > about what we mean in terms of accuracy. > > B. Getting correct current information on the supply is critical. > Depending on how the I/O is designed, different circuits may reside > on different rails. SSN is commonly referring to the output drivers on a > supply which we refer to as VDDIO. However, this is hardly the only > thing that generates noise. The logic circuits such as clock tree that > are before output drivers also generate noise. These circuits typically > are located a separate rail that we call VDD. It is part of the SSN! > There are coupling or interactions among VDD and VDDIO. Without knowing > the precise amount of current going to different supply rails, it would > be hard to get accurate results. Your example is a great case > illustrating this point! > > C. Finding the worst case data pattern is important for SSN predictions > A typical PDN model in frequency domain shows a few resonances. L > di/dt induced noise is typically at high frequency but it may not the > worst case. The most critical case could be the medium frequency range > that is heavily dependent on the data patterns. The ability to excite > the worst case resonance is paramount. > > D. The ability to predict what "circuits" see is very important > Since it is not easy to directly probe on the silicon, one often has > to resort to monitor noise on the package pin and project the noise at > the circuits. Having an accurate passive PDN model would be essential > for this goal. It is also possible to design on chip noise monitor for > hardware to model correlations. Of course, such noise monitor is only > possible for testchips not for products. > > E. Understanding the circuit to noise sensitivity is important. > We typically use the metric "ps/mV" to measure the circuit > sensitivity. Different circuits from different vendors obviously have > different circuit sensitivity. What is acceptable in terms of noise for > one vendor may not be acceptable to others. Knowing this information > would help designer to judge the impact of the noise on system performance. > > The above issues I outlined are somewhat different from what are > already posted. Hopefully, they are useful to those who are interested > in the topic. > > Regards, > > -Chuck Yuan > Rambus Inc. > > Larry SMITH wrote: > > Arpad and all - There has been a lot of good discussion on this thread > > concerning the accuracy of SSN models. Many types of models have been > > discussed including SPICE, IBIS, Behavioral and Macro. In most cases, I > > believe that accuracy refers to the output voltage or impedance vs time. > > > > But for the traditional SSN problem, the issue is really current because > > current is the important variable for L*di/dt simulation. It is > > extremely important that the currents in the Vdd and Ground branches of > > the model are accurate, as well as the output current. I fear that this > > is sometimes overlooked in SSN model discussions. Generally, the > > silicon SSN model will be connected to package inductance matrices, > > power planes and/or transmission lines as well as on-chip, on-package > > and on-pcb decoupling capacitance for the SSN simulation. The most > > important properties of the silicon SSN model are the branch currents, > > which of course are dependent upon the node voltages surrounding the model. > > > > I remember one particular SSN analysis where I was dealing with power > > planes. All current for the driver had to enter and leave through the > > power planes or transmission line traces. Unfortunately, the silicon > > SPICE model that I was using did not conserve current. The sum of the > > current into the VDD, Ground and driver-output nodes was not zero and > > Kirchhoff's current law was violated by the SPICE model! There was no > > zero or ground node reference in the portion of the model that was > > visible, so the leakage was buried down in the process parameters where > > I had no access. > > > > My simulation took some wild voltage excursions as current tried to > > sneak into and out of the driver by some path other than the power > > planes. My solution was to build a dirt simple model from time varying > > resistors, much simpler than a model developed from IBIS data. I got > > good model to hardware correlation even though most people would have > > said that my SSN model was not at all accurate. > > > > The point of all this is that you have to carefully examine the system > > where your SSN models will be used. The system is likely to be very > > sensitive to some model characteristics and not to others. I'm sure my > > SPICE transistor model was very accurate for output voltage vs time and > > maybe even drive strength. But it was not usable in the simulation that > > I was trying to do. > > > > regards, > > Larry Smith > > Sun Microsystems > > > > > > > > Muranyi, Arpad wrote: > > > >>Chris, > >> > >>The same is also true for SPICE! You are putting your faith > >>into those engineer's hands who write your process files. And > >>the average circuit designer usually has no clue for how to > >>correct a bad transistor process file. Many times they don't > >>even know where the limitations of the model are that they are > >>using... > >> > >>Having worked for the same company I work for, you should > >>remember some of the horror stories from those good old days > >>when the IV curve of certain transistor models were on the > >>order of 2-3x away from measurements... And what did we do > >>to get a quick fix? We tweaked the .OPTIONS and did tricks > >>with other simulation parameters because the process files > >>were not going to change any time soon. (Good luck doing > >>this independently for the N and P transistors). But I > >>think I may have already said too much, so I will stop > >>right here. > >> > >>Arpad > >>=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= > >>=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D= > >>=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D > >> > >>-----Original Message----- > >>From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx] = > >>On Behalf Of Chris Cheng > >>Sent: Wednesday, March 16, 2005 3:31 PM > >>Cc: si-list@xxxxxxxxxxxxx > >>Subject: [SI-LIST] Re: package SSN model accuracy requirements > >> > >> > >>... > >>=20 > >>d) Which brings back my fundamental problem with behavioral modelling. I > >>really don't know what the answers in c) should be but none of them = > >>seems > >>straight forward enough to me that the Joe app engineer can crank out in = > >>the > >>very near future. Anytime you have to attempt to deviate from the = > >>original > >>design flow of the Si houses (which I presume is dominated by SPICE), = > >>you > >>are putting faith in the app engineers who, up to now, can not even = > >>abstract > >>out a standard to help me predict a simple SSO problem other than using > >>encrypted HSPICE. > >>=20 > >>What do you think ? > >>------------------------------------------------------------------ > >>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 FAQ wiki page is located at: > >> http://si-list.org/wiki/wiki.pl?Si-List_FAQ > >> > >>List technical documents are available at: > >> http://www.si-list.org > >> > >>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 > >> > >> > > > > > > ------------------------------------------------------------------ > > 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 FAQ wiki page is located at: > > http://si-list.org/wiki/wiki.pl?Si-List_FAQ > > > > List technical documents are available at: > > http://www.si-list.org > > > > 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 > > > > ------------------------------------------------------------------ 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 FAQ wiki page is located at: http://si-list.org/wiki/wiki.pl?Si-List_FAQ List technical documents are available at: http://www.si-list.org 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