| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
|
Joel, all Great! We are having conversation and hopefully making progress.
Let's first look at the proposed model - As Adam has pointed out, this is our contract with .3. It is the case that we have to demonstrate operation at. In looking at the line, I interpret its implementation as loss dominated with little or no via effect, i.e. stub removed via counterboring or other technique, or bottom layer connection. From a loss perspective I think Steve's model is too optimistic, and that Joel's new model looks better. This is my opinion, based on my data with my own and other backplanes, which I am preparing as we speak.
Upper layer connections will exhibit the dreaded via effect if the stub is not removed. I believe the case that I was talking with Brian about probably wasn't even an upper most layer connection - I am buried in s4p data courtesy of UNH as we speak. I, like others, have seen stubs that cause us headaches in the 3.5 to 6 GHz range. My interpretation of the channel modeling in this group is that removal of stubs has been assumed and is an implementation issue. But as others have pointed out there are specifications out there which were designed to support XAUI speeds, and don't necessarily call out or imply stub removal techniques.
I do not agree with the comment that whether the curb moves up or down is dependent 90% on counterboring. There are different degrees of slopes, which have been shown multiple times. If you have a stub you will wind up on a roller coaster of various degrees. The starting point of your ripples will be driven by trace width, length, and materials.
The question of stubs however, will greatly affect the shape of the model. If we decide to include this I anticipate there will be a lot more people who are greyer, balder, or with bigger black bags under their eyes. I think in the interest of our schedule, we should stick to what Joel has proposed, but that the signaling adhoc should not limit its scope to channels that only are above the model. I believe we should put some stub effects in there, and see if any of the signaling schemes are capable of handling them. This will ultimately be in the best interest of the market.
My two cents.....
John
-----Original
Message-----
Joel, all: I think that people focussed on the very high frequency region of my proposed SDD21 magnitude. This isn't what I intended. I was focussing mainly on the region from 50 MHz to 5 GHz. However, when I saw that the equation coefficients I picked created a curve that was below most of the measurements we had, I stopped and didn't go any further. John D'Ambrosia pointed out to us that this doesn't allow for an upper-layer ATCA QuadRoute channel, which happens to have a null in the region of 10 GHz. (My guess is that it probably doesn't allow for any upper-layer channel that doesn't use backdrilling.) We (Xilinx) think that most signaling methods are probably not very dependent on what happens at 10 GHz. So, in the interest of resolving this, Brian Seemann and I are re-evaluating Xilinx' proposed curve. One question I would ask: If the curve is to be formulated primarily on the basis of whether or not backdrilling is done, then why have we paid so much attention to Dk,Df? Whether the curve moves up or down by a dB is probably 90% dependent on backdrilling and 10% on material. Regards, Steve A. Joel Goergen wrote:
|