Fred, Peter, You are correct, as always, it depends. For people who might be familiar from their experience mostly with only one or the other extreme scenario, let me summarize briefly the situation. Though it was not specifically mentioned, everyone on this thread assumed (or I think so) that the bypass capacitors are connected to planes (as opposed to the smallest possible power patch, when we cant speak about distance, because there is just an active device with a few capacitors around it on the patch). It is usually agreed that at low frequencies, the location of capacitor does not matter much. So bulk capacitors could be excluded from the discussion, though it is easy to show that we can always find an exception if we want to: imagine the bulk capacitors that have their series resonance frequencies in the 100kHz-1MHz range. The wavelength is very long compared to any practical PCB size, and still, it may be a good idea to put the bulk capacitors close to the output of the DC-DC converter, otherwise the switching ripple current (in the other of a few hundred kHz) will have to travel all over the board and would spread the switching ripple. At higher frequencies, the location of capacitors MAY matter. It depends on the impedance ratio of planes and capacitors. Similar to a matching resistor for a trace, when the impedance values agree or are close, the setup becomes increasingly insensitive to the geometry and relative locations. If we terminate the end of a trace properly, the input impedance on a low-loss trace is always the characteristic impedance, regardless where we are along the trace. Same applies in two dimensions to planes: if the impedance of the planes is close to our required PDN impedance, component locations matter very little as long as we provide a matched termination to the planes. The capacitor location increasingly matters as soon as we cannot or do not want to use matched planes: either the impedance requirement might be so low that it cannot be easily provided by (matched) planes, or we do not have the proper components to provide the plane matching. As the ratio of plane impedance and capacitor ESR increase, there is an increasing sensitivity to component location. The typical scenario is that the plane impedance is high and the ESR is low. This is similar to having a shorted trace: the input impedance is a strong function of distance even at very low frequencies. If we keep the proper distance from the active device, our impedance target still can be met. Capacitor-plane resonance will build up, but as long as we dont excite them, it is OK. Dependent on system considerations (determining our particular impedance target), available components, cost and manufacturing constraints, we may have one or the other extreme, or anything in between. All of the comments leaning one way or the other on this thread are correct within their validity range. We have designs embracing the full range explained above, we built and measured them. This thread started out with the posting about the UltraCAD tool. It is a matter of preference to use expression solvers (like the UltraCAD tool, or a spreadsheet, or a MATHCAD file) or a simulator (SPICE or alike). I use several different options, Berkeley SPICE, PSPICE, HSPICE, spreadsheets and scripts based on analytical expressions. All have their particular sweet spot of usage. SPICE can run fast, but it is easier to use an expression solver to simulate a full impedance surface, which can be set to step through a range of frequencies and display the changing impedance surface of a plane with a bypass capacitor at an arbitrarily specified location. Such a spreadsheet will be available for free download in the near future: send me an off-line mail if you are interested. Notification is sent out only to those, who ask for the e-mail notification. Best regards, Istvan Novak SUN Microsystems ----- Original Message ----- From: "Fred Balistreri" <fred@xxxxxxxxxxxxx> To: <si-list@xxxxxxxxxxxxx> Sent: Monday, August 11, 2003 6:21 PM Subject: [SI-LIST] Re: UltraCAD ESR and Bypass Capacitor Caculator > I believe the condradictions start when we generalize. Every design > is different and the solutions for one do not necessarily apply to > another. But we should be able to agree on the basic concepts and > come to understand what the assumptions are when statements are made. > > Best Regards, > > Peter Arnold wrote: > > > > All, > > > > I am not sure there should be so much controversy over this placement thing. > > Among the conclusion in the UMR paper that I have a copy of [1] is this > > assertion: "...all decoupling capacitors are shared in the frequency range > > where they are effective (typically below 200-300MHz), and the location of a > > decoupling capcitor on the board is relatively unimportant." > > > > The "unimportant placement" claim is made dependent on the condition that > > the caps, as mounted, are ineffective above 200MHz. At this frequency an > > 8"x9" test board is electrically small with respect to 200MHz wavelength in > > FR4 (30") and the paper reports placement not to be crucial in determining > > impedance. > > > > I think this electrical-smallness, and taking capacitor mounting inductance > > at 2-10nH to be >> plane inductance between parts, is what prompts them to > > simplify the power delivery model to a network connected between two spice > > nodes as if all components were physically at the same place. Analyzing this > > the authors identify a frequency "fa" lying between the last zero and the > > last pole of the impedance plot above which the decouplers are ineffective. > > If this frequency is low as in the examples given, then cap placement might > > well not matter. > > > > But designs today do not necessarily support the simplification given above. > > Modern techniques like via-in-pad, multiple fanouts, "broadside" case > > connections etc. reduce mounted inductance to 1nH or less per part. This > > should allow mounted capacitors to be effective at higher frequencies where > > wavelengths are only a few inches and the board is no longer electrically > > small. At this point the everything-between-two-nodes and > > no-inductance-in-series-with-plane-capacitance simplification may not hold > > and a bedspring-type model would be needed. Decoupler placement may then > > become important (as experience appears to confirm!) This does not > > contradict the UMR authors' statement, which relates to a > > high-mounted-inductance regime and frequencies below 200MHz on a small > > board. > > > > Can anyone comment? > > > > [1] Hubing et al., "Power Bus decoupling on Multilayer Printed Circuit > > Boards," in IEEE trans. EMC vol 37, no. 2, May 1995. > > > > Regards, > > peter arnold. > > > > -----Original Message----- > > From: si-list-bounce@xxxxxxxxxxxxx > > [mailto:si-list-bounce@xxxxxxxxxxxxx]On Behalf Of Ray Anderson > > Sent: Monday, August 11, 2003 9:36 AM > > To: si-list@xxxxxxxxxxxxx > > Subject: [SI-LIST] Re: UltraCAD ESR and Bypass Capacitor Caculator > > > > 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. > > > > > > It's time to do some good engineering on this subject and do away with > > > opinion. The UMR paper is good engineering. Anyone who chooses to > > > disagree with it has the burden of showing where it is wrong by using some > > > good science. > > > > > > Lee > > > > > > Lee Ritchey > > > leeritchey@xxxxxxxxxxxxx > > > Why Wait? Move to EarthLink. > > > > Or conversely, show that it doesn't matter. > > > > Take a large system board (say about 24" square) put a bunch of high current > > processors and ASICS on the left side of the board. Put all your decaps > > on the right side of the board (since you maintain position doesn't matter). > > > > In this hypothetical case I can just about guarantee the board won't > > function > > properly or pass EMI. Admittedly, it is a contrived case, but I think it > > illustrates the point. > > > > As far as the technical demonstration goes, we have indeed demonstrated to > > ourselves that position does matter. > > > > I'm not attempting to further an argument, but I do feel those who have > > responded with viewpoints other than yours shouldn't be chastised and > > accused > > of spreading unfounded opinion. Don't discount the chance that an opinion > > that > > differs from yours just might be correct. > > > > -Ray Anderson > > Sun Microsystems Inc. > > > > ------------------------------------------------------------------ > > 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 > > > > -- > Fred Balistreri > fred@xxxxxxxxxxxxx > > http://www.apsimtech.com > ------------------------------------------------------------------ > 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