I've had some thoughts on the LTI assumption for interconnect modeling and simulation.
Basically, given long serial channels with CDR, I will concede that the LTI assumption is the correct one within the CDR loop bandwidth ... which can be down to a few 10's of kHz. Issues of adjacent crosstalk can be modeled, and time/temperature drift over a period of a few hundred microseconds is negligible.
Beyond the CDR bandwidth, the next time constant that should be considered is the equalizer recalibration time constant. In some cases, the equalizer is set once and never changed. In other systems there is a standard recalibration period. This could be a very long time constant in which the interconnect characteristics may vary. Insertion loss, return loss and group delay can all vary within this time period. In this case, it would be necessary to setup the equalizer calibration using an interconnect initial condition, and then perform multiple simulations across the full range of interconnect impulse responses using the original equalizer settings.
This can be accomplished by having a suite of LTI interconnect models which cover the expected operating range of the system, to account for aging, and "interconnect drift" due to temperature, humidity, or any other factor. Measurement based models could be created by placing a backplane into a temperature and humidity controlled chamber and making various s-parameter measurements across the worst case operational range.
An EMD model for this interconnect would need to be able to point to a suite of models, and provide parameters necessary to select the appropriate model for the simulated condition. If, for example, a system were sitting in a temperature and humidity controlled environment, and the equalizer was initially trained and calibrated at a temperature of 25C and 20% RH, one would select the interconnect model closest to this. Then, once the equalizer has been trained and calibrated, the worst case excursion within the calibration time period could be simulated, by say picking a model for 80C and 100% RH, to simulate the impact of a summer air conditioning failure.
Scott Scott McMorrow Teraspeed Consulting Group LLC 121 North River Drive Narragansett, RI 02882 (401) 284-1827 Business (401) 284-1840 Fax http://www.teraspeed.com Teraspeed® is the registered service mark of Teraspeed Consulting Group LLC Mike Steinberger wrote:
Arpad-Taking either Scott's case of a vibrating vehicle or Richard Ward's even higher frequency case of a vibrating disk drive, how does the highest frequency of vibration compare to the data rate of the channel? I'm having a hard time imagining a significant harmonic component above about 100kHz. Maybe I don't have much imagination, so let's make it 1 MHz..Still, with that many orders of magnitude difference between the mechanical frequencies and the electrical frequencies, any change in electrical wave shape due to mechanical movement is going to be insignificant, so the LTI approximation is still a good one.What I think these examples do suggest is that the designer of a system that has mechanical vibration must analyze the channel performance for a representative sample of the mechanical configurations that will occur over the course of a cycle of vibration. Thankfully, most of us won't have to execute that procedure.I'm reminded of an old saying about user interfaces: "If an interface is so easy to use that any fool could use it, only a fool would." I think a similar statement applies to engineering languages: "A language which is capable of expressing any engineering problem is too complex to be useful for any engineering problem."My 2c. Mike S. Muranyi, Arpad wrote:In our recent discussions on Interconnect-SPICE and our last IBIS-ATM teleconference the question was raised whether we can safely assume interconnects are LTI. Scott's message below (from a different thread) seems to indicate that there are situations when this may not be the case. This makes me nervous about writing a specification that by its definition would disallow those effects to be simulated... Any comments?Arpad=============================================================== ------------------------------------------------------------------------ *From:* Scott McMorrow [mailto:scott@xxxxxxxxxxxxx] *Sent:* Thursday, June 26, 2008 2:03 PM *To:* Muranyi, Arpad*Cc:* twesterh@xxxxxxxxxx; msteinb@xxxxxxxxxx; wkatz@xxxxxxxxxx; huangchunxing@xxxxxxxxxx; IBIS-ATM; guantao@xxxxxxxxxx*Subject:* Re: [ibis-macro] Re: IBIS-AMI ArpadI agree with you, and would add that even the analog network characterization portion of a passive interconnect system is not a well understood topic. Several issues come to mind:* Time variance of the interconnect due to: o Temperature o Humidity o Mechanical Vibration (think twin-ax cables in a vehicle) * Equalizer training in the presence of crosstalk and noise spikes in a hot-plug environmentBut heck, I'd settle for an IBIS-AMI model that correlates with something that can actually be measured.Scott Scott McMorrow Teraspeed Consulting Group LLC 121 North River Drive Narragansett, RI 02882 (401) 284-1827 Business (401) 284-1840 Fax http://www.teraspeed.com Teraspeed® is the registered service mark of Teraspeed Consulting Group LLCMuranyi, Arpad wrote:Sorry to everyone for arguing a little on what has been stated so far. I tend to disagree with the statement(s) that the analog corner modeling is a well understood topic. Just because it is common practice to use high/low supply voltages with low/high temperatures, etc... to achieve best/worst timings, it doesn't mean that this practice actually gives the best/worst timings. Having done large amounts of parameter sweeps while I was working for my previous employer, I saw solution space plots which had failing islands corresponding to "in-between" parameter values, and not at the extremes. The answer then was that we do not know where the system is failing unless we simulate all possible combinations of parameter values. That's when the frequency domain (resonance) analysis, sensitivity analysis, design of experiments (DOE) and similar techniques became popular. I tend to believe that the application of statistical analysis techniques is a natural continuation of this evolution, as kind of an attempt to reduce the amount of time it takes to go through all possible parameter combinations... My $ 0.02 worth... Arpad ===========================================================-----Original Message----- From: ibis-macro-bounce@xxxxxxxxxxxxx [mailto:ibis-macro-bounce@xxxxxxxxxxxxx] On Behalf Of Todd Westerhoff Sent: Thursday, June 26, 2008 1:00 PM To: msteinb@xxxxxxxxxx; scott@xxxxxxxxxxxxx Cc: wkatz@xxxxxxxxxx; huangchunxing@xxxxxxxxxx; 'IBIS-ATM'; guantao@xxxxxxxxxx Subject: [ibis-macro] Re: IBIS-AMI It's worth pointing out an IBIS-AMI model actually has two models - - the analog model in the .ibs file, and - the algorithmic model pointed to by the .ibs fileIBIS-AMI analysis separates analysis of a link into two stages - networkcharacterization and link analysis. The analog model is used for the former, while the algorithmic model is used for thelatter. When we talk about modeling PVT variation, it's worth considering howPVT affects each type of modelseparately. Modeling the effects of PVT on a transmitter's output stageor a receiver's input stage is well understood (it's what IBIS has been doing for years), while modeling the effects of PVT on equalization behavior is [relatively] new ground. I agree this is something we should consider standardizing, and I also agree that we need more experience in this area before it makes sense to try & put such standards in place. Todd. Todd Westerhoff VP, Software Products SiSoft 6 Clock Tower Place, Suite 250 Maynard, MA 01754 (978) 461-0449 x24 twesterh@xxxxxxxxxx www.sisoft.com--------------------------------------------------------------------- IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/ IBIS Macro reflector: //www.freelists.org/list/ibis-macro To unsubscribe send an email: To: ibis-macro-request@xxxxxxxxxxxxx Subject: unsubscribe
--------------------------------------------------------------------- IBIS Macro website : http://www.eda.org/pub/ibis/macromodel_wip/ IBIS Macro reflector: //www.freelists.org/list/ibis-macro To unsubscribe send an email: To: ibis-macro-request@xxxxxxxxxxxxx Subject: unsubscribe