Hi Leonard: > I am doing a PCB Layout using Xilinx's Rocket I/O chips. I=20 > have read a lot about signal integrity from the Xilinx's web=20 > site and even got DR Johnson's DVD on rocket IO which is a=20 > great DVD but, not being an SI expert, I have a few questions=20 > that perhaps someone could explain or point me into the right=20 > direction to an article I could read? I appreciate any help=20 > you could provide. There's a new RocketIO design kit for HyperLynx GHz on http://www.mentor.com/products/pcb/pads/design_kits/hl-rocketio_design_k it.cfm. I believe it would help you experiment around enough to answer most of the questions you've listed below and more. A couple of caveats: a) The kits are currently based around pre-layout schematics; and some of yoru questions would require pulling in an actual foard file. (Easy to do.) b) To simulate an actual design, you'll need the released software (v7.2) or the v7.5 beta. (Contact me offline if you want to apply as a beta site. We're interested in getting a few more RocketIO customers on the beta for 7.5.) General comments below. (Simulation or test boards are advised to get "real" answers here ... To augment the generalized advice you're likely to get.) > 1. I have seen on DR Johnson's DVD how the eye pattern looks=20 > on one side of the signal and then 26 inches later how much=20 > smaller it looks and the receiver still works. My question is=20 > that FR4 material has a delay about 1.8 NS per foot. At 26=20 > inches across two cards and a backplane, doesn't the 4=20 > nanoseconds of delay in the fr4 material exceed the signal=20 > speed in the Gigahertz range (1 nanosecond) or is it these=20 > drivers and receivers work differently then lets say standard=20 > TTL signals? With SERDES designs, delay, which is a big factor for synchronous signals, takes a back seat to loss, both dielectric loss and (resistive) skin effects. The closure of the eye being the big issue. > 2. In a perfect design signals would go from the BGA at 100=20 > ohms to perfect vias to the bottom of the board to perfect=20 > 100-ohm traces to a connector. I have a few questions about=20 > what to do in a non-perfect design. >=20 > a. On connections to the BGA pins I can use the standard dog=20 > bone pattern and have the vias 40 mils from the bga pins but=20 > the traces are not going to be next to each other to form 100=20 > ohm differential in this short space. Is it acceptable to do=20 > this since or should I route the traces from the bga pads to=20 > vias further away from the BGA where I can route the traces=20 > as differential pairs? This is a common problem when fanning out from a BGA. As long as overall path lengths are matched, and crosstalk with adjacent aggressors is considered (using common sense and simulating). > b. I have not seen anything that talks about if there are=20 > any requirements for the spacing between the two vias on a=20 > differential pair or how to calculate the impedance of a via.=20 > I know having a larger antipad on the plane layers helps but=20 > what about spacing between the vias? Or the fact that the gap=20 > between the traces changes going into the two vias? Keep in mind that vias - generally speaking - do bad things to super-high-speed signals, so the fewer the better. HyperLynx GHz simulates the impact of vias on signal/eye quality, and v7.5 includes a "Via Visualizer" capability that you may find handy, as far as looking at via parasitics (L, C, Z0, Td, etc.).=20 > c. I have set up a 30-mil edge-to-edge gap between all rocket=20 > IO differential pairs to reduce the crosstalk but at the=20 > connector area I can get many more traces routed on a bottom=20 > layer if I violate the 30-mil spacing for about 500 mils to=20 > about 5 mil Gap instead of the 30 mil. Would this cause any=20 > crosstalk problems since its such a short trace? It will cause some crosstalk ... The question is whether the crosstalk will be significant enough to cause a problem at the receiver. Again, this is where simulation of your actual layout would be necessary, rather than relying on overly-simplified design rules. > d. In order to match the lengths of the differential pair I=20 > usually extend one of the traces at either the connector end=20 > or the last via and wrap the trace around the pin or via to=20 > get additional length. Are there any problems with this=20 > method? Is there another method that is better ? If you wrap in the same direction, self-coupling can be the problem here. The severity of the potential problem depends on how clean your eye is in the first place, and the parallel length over which the self-coupling occurs. To mitigate this effect, it's better to take a non-parallel routing path to avoid the self-coupling problem. Bill Hargin Product Manager - HyperLynx Mentor Graphics Corp. 425-497-5079 - Direct 425-301-4425 - Mobile ------------------------------------------------------------------ To unsubscribe from si-list: si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field or to administer your membership from a web page, go to: //www.freelists.org/webpage/si-list For help: si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field List FAQ wiki page is located at: http://si-list.org/wiki/wiki.pl?Si-List_FAQ List technical documents are available at: http://www.si-list.org List archives are viewable at: //www.freelists.org/archives/si-list or at our remote archives: http://groups.yahoo.com/group/si-list/messages Old (prior to June 6, 2001) list archives are viewable at: http://www.qsl.net/wb6tpu