Your results are just what I would expect for crosstalk analysis. There is a KB article (https://support.mentor.com/knowledge-base/MG55023 ) on Support Center with more details. A section of this article says,
"Quick Batch Crosstalk:
The typical way batch mode is used for crosstalk analysis is to first run a Quick Analysis of all the nets on the board using any real models that may already be assigned, and default IC characteristics for those nets without models assigned. This type of simulation will highlight the potential crosstalk problem areas on the board that should be further analyzed using batch mode Detailed Analysis or manual Interactive Analysis. The Quick analysis batch report will identify potential victim nets and aggressor nets based on approximations."
The interactive and batch simulations are often slightly different because of different configurations of model and simulator options. The quick analysis is an estimate based on a formula, not simulation. The formula can't include all effects of crosstalk, so it might have a rather large error. The quick estimate is never the final answer. It is useful in determining if some nets really need detailed simulation. For example, if I know that I can allow 100mV of crosstalk on nets within a certain bus, and the quick estimate says that a few nets have less than 50mV then I don't need to spend the time simulating crosstalk on those nets.
In the early days of HyperLynx when computers ran slower, this method could save significant time in a simulation. With modern computers, the time savings might not be so significant (depending on the layout that you need to analyze) so it might be just as efficient to go ahead and run the crosstalk simulations in a batch analysis.
Dear Mr Weston,
In the Quick analysis mode, the rise/fall time and switching range of IC model effect much to crosstalk level.
I have run two cases simulation below in quick analysis mode:
1/. Use the default IC model with 100ps rise/fall time and 740mV switching range. The crosstalk level of DDR2_D26 net is 118mV.
2/. Use my IBIS DDR4 controller. The crosstalk level of DDR2_D26 net is 139mV.
I have checked my IBIS with default ocd34 output driver model below:
I see that, with fast-strong corner, the rise time is around 100ps and switching range is arround 740mV.
From (1) and (2), I guess that the quick analysis use the rise/fall time and switching range to calculate the crosstalk level. That is the reason why the crosstalk level is too high when compare with detail analysis or manual interactive mode.
Can you show me the formula which is used to calculate the crosstalk level in quick analysis mode ?
Your investigation confirms that having a model assigned to pins on a net affects the quick estimate of crosstalk. The formula uses some data from models that are assigned. If no models are assigned then the estimate uses the default IC parameters. The driver rise and fall times, and voltage swing are taken from the Ramp keyword instead of discovered from the waveform tables. It's just easier that way. The estimate also uses C_comp from assigned pin models. The other factors in the formula are the trace coupling in the PCB layout. I don't have the exact equation.
A major difference between the formula and a simulation is the relative placement of the coupled traces with other traces. You can run some experiments in LineSim to see that the same coupled section of transmission lines creates different crosstalk when the other traces on the coupled nets are changed.