Kim, Just curious, what kind of via and pad structure did you used to connect the bypass caps to the planes (in particular to the 50 mil planes) ? -----Original Message----- From: Kim Flint [mailto:si@xxxxxxxxxxxx] Sent: Wednesday, August 13, 2003 3:59 PM To: si-list@xxxxxxxxxxxxx Subject: [SI-LIST] Re: UltraCAD ESR and Bypass Capacitor Caculator At 08:26 AM 8/11/2003, Lee Ritchey wrote: >If the location of decoupling capacitors matters, perhaps some technical >demonstration would prove that. Short of such a demonstration, this is >speculation and not the sort of thing that should be used to make design >choices. In light of Lee's request for more analysis and less opinion, I thought I might try doing a few simulations to see what effect due to capacitor location could be shown. I mainly looked into layer switching and the effect of cap placement on resulting power fluctuations. I used Sigrity's Speed2000 demo version, so other people can take my files and easily experiment with them as they like even if they don't own a copy. The demo version is free and limited in functionality, but useful for small test cases like this. You can get it from Sigrity's web site, http://www.sigrity.com. I put the simulation files, some explanation about them, and some interesting plots on my website, where you are welcome to grab them: http://www.kimflint.com/capplacement/capplacement.html I think there are at least two aspects to this issue of capacitor placement to consider, and they both appear to be mixed together in the discussion so far. One is the impedance of the PDS with frequency and whether capacitor placement affects it, especially considering I/O behavior on the die. This is important in determining whether voltage margins will be maintained when current is switched by a transistor on our die at a given frequency. We want to analyze whether we get SSO problems and such due to supply collapse and where we might place capacitors usefully to help. That much I think has been covered already. A second aspect might be called "return current control", which I would like to discuss some more. A few people noted that they believe capacitor placement affects EMI, or loop currents. Probably we should include crosstalk in there as well, since a large loop current on an aggressor return path can result in crosstalk to victim nets. Capacitors are often placed in strategic locations on a board to control such problems, by providing a path for the return current to switch layers. For example, we might place caps at locations where traces switch across layers, cross plane splits, or escape packages. So does that work? Or more usefully, when does the cap location matter and when does it not matter? How much effect does it have? In an ideal case, all of our signals would have a perfectly continuous transmission line environment from start to finish. Practically however, this is rarely the case in real board designs. For example, as a signal transitions from the package to the board, it may end up with different reference planes on the PCB than it had on the package. It may be necessary to do this briefly for a BGA escape, or because we simply don't have any information about the package routing of another vendor's BGA and we don't know what reference planes to use. Or, in a dense board we may be forced to switch layers on some signals due to routing congestion. Of course we all try to minimize this, but it's gonna happen. We know that return currents travel on the reference plane following our trace, and wherever we switch layers the return current has to find a way to switch reference planes along with it. If there is no convenient place for the return current to transition to the new reference planes, we may end up with a large current loop. This radiates and causes crosstalk, EMI, noise, etc. We use capacitors to make an ac path across the planes and hopefully limit the loop area. It is really interesting to run the simulations and see the 3d animation of the planes with the wave propagating as the signal switches layers. It looks like a pebble dropping in a pond, with the waves rippling out from the via where the layer switch happens. It gives you an intuitive idea of what happens on your planes when you switch layers, so I encourage you to take my files and try the simulation yourself if you haven't done this before. The simulations I did tend to lend credence to the idea that you shouldn't rely on rules of thumb or absolute statements. You need to consider your application! And simulate! Because from what I interpret from the results, sometimes the cap locations matter and sometimes they don't. It is no more correct to say "location never matters" then it is to say "location always matters." In my simulations I compared closely spaced planes vs. widely spaced planes, different cap positions, and different plane sizes. I used a signal pulse with 100ps edge rates, so the frequency content is fairly high. In cases where the plane spacing is close (3mil) and the interplane capacitance relatively high, the cap location did not matter too much. However, when the plane spacing was larger (50mil) and therefore less natural interplane capacitance, the bypass cap location did make a difference even with 100ps edge rate signals. Clearly though, improving interplane capacitance had a much better result, reducing peak voltage fluctuations on the planes by an order of magnitude in this case. So, in cases where you can't get much interplane capacitance due to stackup limitations, small area power islands, or switching across multiple layers, cap location might make a difference for you even at fairly high frequencies. In other cases where you are switching across closely spaced planes, it probably doesn't matter. Again, I would urge you to try it yourself and use your imagination to come up with your own conditions. Feel free to use my files as a starting point. I'm just one simulation monkey banging on one computer, and there are many variables I didn't try. (different edge rates, different capacitor parasitics and mounting techniques, more caps, etc.) A few more monkeys providing their results could give us all a lot more knowledge. also, comments welcome! If you spot some error in what I did, let me know. thanks, kim ------------------------------------------------------------------ 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 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 ------------------------------------------------------------------ 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 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