|Thread Links||Date Links|
|Thread Prev||Thread Next||Thread Index||Date Prev||Date Next||Date Index|
I reviewed the na_01_1207 package presentation from IBM. Overall its great work! I have a few comments.
The original KR package models were developed to represent a worst case performance without giving away IP. Most of the channel simulations had 1 or 2 pole low pass poles at 7.5 GHz. I agree these models are not real packages with loss. So I think the package losses IBM presented are reasonable. It can buy back some margin.
There is one more effect that I’ve seen that is much worse that package IL and that is resonance caused by the plated though hole (PTH) via in the package, the bga, and 1 or so pF at the die. In a multilayer BGA package design (page 12) the PTH via does not have to be near the solder ball. Many designs put this via some were on the run between the die bump and the BGA. This has a resonant effect that is much worse than IL. I was wondering if we could get some package models with that effect included.
Package crosstalk data is very useful for a system designer for specific systems. We have found from a multitude of die to die analyses that crosstalk power is directly proportional to received jitter. What was surprising was the correlation between crosstalk power and jitter was relatively independent of how and where the crosstalk occurred in package. It sort of brings up the notion of spec budgeting. I could present this at the up coming meeting.
Richard Mellitz, PE, Intel Corporation
From: John DAmbrosia
Dear Task Force Members,
Proposed models for SFP+ host and PMD IC models with accompanying .s4p files have been submitted by Ali Ghiasi, Broadcom. The data and reference presentation have been uploaded to the task force public/channel data area, and may be found at http://grouper.ieee.org/groups/802/3/ba/public/channel.html