Rich,
Yes, Skew profile in time domain is not the same.
You can also plot skew in freq domain – profile is oscillating over mid to high freq.
can you explain what is ref.voltage (%) on right side plot?
-Upen
From: Richard Mellitz <Richard.Mellitz@xxxxxxxxxx>
Date: Thursday, September 14, 2023 at 8:48 AM
To: "Upen Reddy Kareti (ureddy)" <ureddy@xxxxxxxxx>, "STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx" <STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx>
Subject: RE: [802.3_B400G_ELEC] PKG model mode conversion - Follow up on yesterday's question
Hi Upen,
I don’t know if this cut and paste will make it through the reflector. But below is the pn TDT step responses and delay a different thresholds for you Cabled_Host_ball_ball_20p5db channel.
It sort of suggest all skew is not the same.
From: Upen Reddy Kareti (ureddy) <ureddy@xxxxxxxxx>
Sent: Wednesday, September 13, 2023 10:29 PM
To: Richard Mellitz <Richard.Mellitz@xxxxxxxxxx>; STDS-802-3-B400G-ELEC@xxxxxxxxxxxxxxxxx
Subject: Re: [802.3_B400G_ELEC] PKG model mode conversion - Follow up on yesterday's question
|
Caution: This email originated from outside the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
|
Agree Rich,
That is why I shared those channels so that others can study ( in addition to me) in various perspective and come to ways that we can address them.
Impacts like compression of Difference signal and pulse width and increase in levels of common mode signal levels and its pulse width, There can be other impacts coming from Analog path in Silicon ( true and complementary signal level differences
, common mode signal impact on difference signal at differential amplifier and many others to list)
-Upen
I analyzed your channels, and I can say the skew is not as simple as some might think. Skew can even behave like an ISI filter or amplifier at expense of signal. :-)
|
Caution: This email originated from outside the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe.
|
Hi All,
Mode conversion topics are lot more complex as Rich mentioned. We have to collectively decide how to account for them.
In package designs for core Vias - if additional reference vias are placed with in ¼ wavelengths of signal reducing some of this impact.
Mode conversions contribution is not limited to package vias, other segments in package contribute it in spite of physical lengths matched, similarly from pcb and cables.
In concentric construction structures like cables -The tightly coupled cables behave differently than loosely coupled cables
In planar structures like PCB and packages – there is limit how tightly you can couple due to manufacturing constraints/limits. In addition in PCB fabric weave alignments plays a role.
Observing skew impacts and SCMR limits considered for .ck is not correlating well on their impact on Serdes in my studies.
There are many false pass and false fail cases when I build channels and analyzed with a specific ref. serdes architecture ( COM is not used here as it needs to be updated to fully account mode conversions and CM impact difference signal
in Analog front end etc)
we can start with channel models I distributed with skew ( cabled host channels only have skew, pcb_host channels I did not include skew)
-Upen
Hello Rich/Liav,
I respectively 0.8 dB and 1.0 dB mode conversion for package A and B BGA transition to PCB, Mike Li presentation show mode conversion for 1 mm and 0.5 mm package.
Lets wait an see to hear back on Liav how small is mode conversion for package B, and if we need to add some penalty for it in COM. At this point in Ghiasi_3dj_02a these are place holder that I expected will be reduced with more data and
that will help address C2M application with lower bump-bump loss than suggested in contribution.
Based on what Liav says below package B mode conversion was reduced to something smaller than 1 dB I assumed in my bottom up analysis
In response to Rich comment on the DM to CM or CM to DM conversion these are captured by the s-parameters and would be reflected in the loss, in addition we also have limit on CM generation at various test point to prevent excessive CM
voltage on a receiver that may need to operated with few mV of signal..
Thanks,
Ali Ghiasi
Ghiasi Quantum LLC
I think mode conversion is a complicated mater. Many effects are covered or included in the transmitter and receiver specification. For example, p-n skew and imbalance directly affect p_max/V_f. There are other
effects which may not be included such at DM to CM and back to CM to DM conversion in the package model. We do have SCMR limit. Presently we do not use a SCMR derived parameter in COM. Instead, perhaps a bit of over budgeting is not a bad thing as Liav indicated.
The COM package model strategy and COM strategy are similar. That is it is performance you must meet and not the performance you get. (Paraphrased from conversation with John Calvin). I think the package model is much like the 3dB COM limit and should be
a like a global budget for specific design choices.
|
Caution: This email originated from outside the organization. Do not click links or open attachments unless you
recognize the sender and know the content is safe.
|
Yesterday Ali referred a question to me regarding the package model mode conversion value.
I wanted to provide a better answer than “negligible”.
After going back to check on the exact value, I found the mode conversion of the models supplied to be better than -30dB all the way up to above 60GHz (actually in some of the models I was seeing also better than -36) – Further tuning of
de-skew, etc. may take the value even lower.
All that said, one needs to keep in mind that these models are synthetic, and lack much of the routing complexity of actual package routing.
Such complexity may force higher conversion results, which we actually may want to take into account in the budgeting of the spec (and I guess that was what Ali was referring to).
Exact impact of mode conversion in dB loss (of actual case packages), yet to be exactly analyzed, but…
Once we analyze it, IMHO, we will need to use this “a bit over valuated budget” to account for other parts of the same table which were undervalued.
To unsubscribe from the STDS-802-3-B400G-ELEC list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-ELEC&A=1
To unsubscribe from the STDS-802-3-B400G-ELEC list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-ELEC&A=1
To unsubscribe from the STDS-802-3-B400G-ELEC list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-ELEC&A=1
To unsubscribe from the STDS-802-3-B400G-ELEC list, click the following link:
https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-ELEC&A=1
To unsubscribe from the STDS-802-3-B400G-ELEC list, click the following link: https://listserv.ieee.org/cgi-bin/wa?SUBED1=STDS-802-3-B400G-ELEC&A=1
|