Roi - thanks for your question. I am going to move it to the HyperLynx specific community where it will have more visibility with subject matter experts. The Members Resources area is for general community questions and suggestions.
Probing at the die will account for Rx equalization including CTLE and DFE. To see the effect of DFE alone, you can turn off Tx FFE and Rx CTLE and keep on only DFE. The DFE signature on the signal is unique, and includes some abrupt voltage shifts in the eye density plots.
The result you see from the IBIS-AMI analysis in HyperLynx is the output of the RX. Actually, there is no option to explicitly view the eye diagram at the pins. You could effectively get the view at the pins by setting the RX options to do nothing. As Cristian alluded to, you can verify that the eye diagram is from the output of the RX by noticing the steps in the waveforms that look something like this.
Those steps are created by the DFE function in the RX.
Here is the eye density plot from our IBIS-AMI Channel Analyzer using Mentor_Tx_Channel_Mentor_Rx.ffs LineSim project and our Generic SerDes IBIS-AMI models, with FFE disabled, CTLE disabled and DFE enabled (15 taps):
In addition to that, in the last page of the wizard you can choose the "Extended" option from the dropdown list available for the AMI model debug info log. This will generate the impulse responses before and after Rx Init, that you can load in EZwave:
You will see the effect of the DFE taps in the impulse response after Rx Init.
Moreover, in the Reporter's window the values of the normalized DFE taps are displayed:
Hope this answers the question.
your answers were very helpful.
If you really want to probe at the pin, you can add a dummy differential symbol (no model assigned to it) and select it in the dropdown list of the receiver probe:
You will note that in the "Configure AMI Models" page of the wizard the "Configure Rx AMI..." button will be grayed out because there is no model assigned to this symbol.
Here are the results without DFE:
You can now compare the two eye density plots and notice the difference.
I tried this dummy symbol method but did not get a good eye diagram.
With the probe set normally at the RX+ symbol, I get the eye diagram with DFE:
I can configure the receiver model to use its low power equalization instead of DFE and get this:
So far, so good. But when I set the probe to the dummy symbol, I get this:
It's so smooth it looks like it's not recovering a clock at all. There's a check box called "Ignore rx AMI model clock ticks", which seemed like a good idea since there's no AMI model for the receiver. But that gave me the same result:
Are there other unusual settings you used to get your result without DFE?
Also, is the dummy symbol method valid if I have an s-parameter model for the receiver die? The IBIS file for the FPGA has no curves in it; it just calls out the AMI files. They have s-parameters to model the die. It seems that any nonlinear loading effects of the receiver must be modeled in the Waveform call of the receiver DLL, because s-parameters are only linear. Thus, if there is no receiver AMI / DLL assigned, is the physical load of the receiver missing from the simulation? Or do they do it that way because any nonlinear effects are too small to matter, and the receiver model only handles the DSP?
Without having any information about your models it is hard to precisely say what is going on with your experiments, but I am guessing that your Rx model has other signal conditioning blocks such as CTLE, AGC and (less) possibly FFE. Although you turn off the DFE the eye is still open because of those additional blocks. I would suggest contacting our support if you need better assistance.
For your first question about the view; it might make sense. If the equalization is really good, it could open an eye that appears to be non-existent. Try changing the color map range by unselecting the auto range and move the sliders.
On the other hand, your second question indicates that there might not be an eye opening to work with. If the receiver analog effects like termination are not in the IBIS model but have to be included through the DLL then the signal at the component pins is not complete. In this situation you would need to get the unequalized eye from the Rx model with the equalization features all disabled.
The model I'm working with is the Xilinx Ultrascale Plus GTH, which you may have, but I didn't want to post it here because I got it under NDA. I'm using their example simulation which they show in the user guide. The example channel is pretty lossy. I'll see what details I can get from them about the models, now that I know a bit more about what I'm seeing and how IBIS AMI is supposed to work...