Hi steve, I forgot to indicated that, the signal on Top layer is also 2GHz PECL signal. Sorry about that. Do I have to add more layers in such circumstances, or any expedient? Best Regards, George On 4/18/06, steve weir <weirsi@xxxxxxxxxx> wrote: > > George, I would definitely alter the stack-up of this system. > > I recommend looking carefully at a stack up more like this: > > -1 Top > -2 S1 > -3 GND1 > -4 S2 > -5 L_In1 > -6 GND2 > -7 VCC1 > -8 VCC2 > -9 GND3 > 10 L_In2 > 11 S3 > 12 GND4 > 13 S4 > 14 Bottom > > You will still have to figure out how to adequately couple the PECL Vcc to > GND on each of the daughter cards. At 2GHz that should not be too hard to > manage. > > Steve. > > > sss > > At 12:35 AM 4/18/2006, Kidman Ma wrote: > > Hi Steve, > > Yes, You're right. The stack up i posted before is part of the following > 14 layer stackup. > --------Top > --------GND1 > --------L_In1 (Low speed signal <50MHz) > --------L_In2 (Low speed signal <50MHz) > --------VCC1 (continuous +5v plane) > --------S1 [2.125G PECL, 8mil to VCC1, 6mil to VCC2] > --------VCC2 (continuous +12v plane) > --------S2 [2.125G PECL] > --------GND2 > --------S3 [2.125G PECL] > --------GND3 > --------S4 [2.125G PECL] > --------GND4 > --------Bottem > > This board is one midplane of system, no active data path component on it. > No PECL VCC, either. All driver and receiver is routed on other "daughter > card". No layer switching (no vias) over S1, except the connectors to other > active cards. > > Do you think it's better if I switch +5v plane to VCC2? That means, let S1 > reference to +5v power plane through 6mils gap. > > Regards, > George > On 4/17/06, *steve weir* <weirsi@xxxxxxxxxx> wrote: > George, > > This is primarily a common mode issue. I assume that the five layers > shown represent only part of your stack-up. Is VCC2 your PECL > Vcc? If not, how about moving the PECL Vcc to layer 3and adjusting > your dielectric spacing: > > Vxxx or Gnd > big gap > S1 > little gap > PECL VCC > little gap > S2 > big gap > Vxxx or Gnd > > First, there is nothing magic about "Gnd". To quote Dr. > Archambeault, ground is a place for potatoes and carrots. We are > concerned about wave guides, which means the return image path and > common mode reference voltage. If the same Vcc is used at both the > transmitter and receiver for PECL outputs and inputs, that is the > best but not only choice for the image return plane(s). > > The hierarchy of return paths from most desirable to least is: > > 1) Single surface that matches the source and destination reference > voltage. > > 2) Two surfaces of a single sheet that match the source and > destination reference voltage > > 3) Multiple surfaces tied together by stitch vias that match the > source and destination reference voltage. > > 4) Single surface that does not match the source and destination > reference voltage > > 5) Two surfaces of a single sheet that does not match the source and > destination reference voltage > > 6) Multiple surfaces tied together by stitch vias that do not match > the source and destination reference voltage. > > 7) Surfaces coupled together by interplane and bypass capacitors > > The reason for the hierarchy is uncertainty. 1 is the ideal. 2 is a > very close second. 3 can be a little or a lot worse. 4-7 all rely > on capacitive coupling through cavity dielectric and bypass > capacitors. At 3GHz, a bypass capacitor with 1nH mounted inductance > looks like 20 Ohms. To the 100ps rising edge of a 3GHz serdes, > anything further than about 0.3" away from any given signal > transition through the PCB cavity does not reflect back in time to > impact an incident edge. That means a bypass cap 0.3" away is all > but pointless for the incident edge. Suppressing resonances, is a > different issue. That radius increases inversely with the spectral > components. For the 2.4nS long worst case runs in 8b10b that radius > of action likely includes most of your board. Plane geometries, > stitch via patterns, and to a smaller extent bypass capacitor loading > will determine the location and magnitude of resonances. What you > pass through the cavities particularly in terms of single ended > busses determines excitation. > > If you make layer 3 your PECL Vcc it looks like you have #2 in the > bag. If you want to go another route, you have some homework to > do. Essentially, you need to determine what the impedance profile > looks like plane to plane in each of those cavities and the signal > spectra of what you will be injecting into the cavity(s) in > question. If you can tolerate both the losses that you will get for > the diff pairs going through, and the increased EMI and crosstalk > that will result from ALL of the signals you pass through those > cavities, then all is well. If you have few or no single ended > signals traversing the cavities, then there won't be that much to > mess with your diff signal common mode. Similarly, if you can get > the coupling impedance low enough between planes across your signal > spectra to tolerate whatever it is you do inject, then even 7) still > works. But, you need to figure out where you are with your signals > and those cavities first. Lee attests that he has made 7) work on > hundreds of boards. In the end, it all comes down to coefficients. > > Doug Smith's most recent "Technical Tidbit" is quite germane, and > should open your eyes to some of the differences between 1) and > 7). Just appreciate that on Doug's test board, there are fewer > layers than your board, no ground stitch vias and no bypass > capacitors, all of which affect the results. Also appreciate that > what he tested is susceptibility / radiation. A different test on a > different configuration will yield different results. > > Additional resources include a study done by Scott McMorrow on our > website www.teraspeed.com, and a number of papers by Dr. Bruce > Archambeault. > > Steve. > At 01:51 AM 4/17/2006, geor_dai wrote: > >Dear All SI experts, > > > >One question regarding to return path for signals and signal quality > >concerns. > > > >If we have stripline configuration as below, a signal trace S1 between > >two power planes. Because limited routing space in the board, not all > >high speed signals can be on S2 layer, which is better for signal > >quality. So some of 2.125Gbps signals running on S1. > > > >--------VCC1 > >--------S1 [2.125G PECL] > >--------VCC2 > >--------S2 [2.125G PECL] > >--------GND > > > >My question is, > >When the signal S1 is transient from high to low or low to high, which > >plane would be the return path, VCC or GND? EMC/EMI impact? Any > >compensation suggestion? > >How much risk if we go to PCB fabrication like this stack up? > > > >Thanks very much! > >Best Regards, > >George > > ------------------------------------------------------------------ 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