Larry, When I said the operating frequency in the previous e-mail, I really meant the maximum frequecy below which the package can supply enough current to the core/IO circuit. Sorry about the confusions I made. Let's still looked at your example 1. Ztarget=0.001Ohm Imax=50A Vnoise_max=0.05V Z(Lpackage)@7Mhz=+j0.001Ohm(inductive) If I adjusted the Zpcb@14Mhz=-j0.001Ohm(capacitive), it gives: Z(node2 @14Mhz)=Zpcb+Z(Lpackage)=-j0.001+j0.002=j0.001Ohm It still meets the Ztarget requirement at 14Mhz, but the voltage drop cross the Lpackage is 0.1V (with Imax=50A) which exceeds the Vnoise_max. Does this big voltage drop cross package really matter? I don't think so. The point I am trying to make is that the power delivery system should be treated as a whole system. When the decoupling caps are placed on the PCB, the chip's package models and on-die decouplings should be considered. You can not design a universal PCB decoupling for all packages. Thanks, Zhiping Larry Smith wrote: > Zhiping - the frequencies mentioned in previous examples have nothing > to do with the operating frequency of the core or I/O circuits. They > are the corner frequencies where the inductance impedance becomes higher > than the target impedance. > > In your diagram below, the voltage source (VRM) is on the left and the > load is on the right. The chip load wants to look out and see a > voltage source with the target impedance (or less) from high frequency > all the way down to DC. If we drop 5% or more voltage across lpackage, > the circuits may malfunction, loose data, fail to meet timing, etc. > You are correct that the only voltage that really matters is > node2-Gnd2. But it is going to be really hard to maintain the voltage > at node2-Gnd2 if excessive voltage is dropped across Lpackage. > > The on-chip capacitance delivers the current from high frequency down > to some frequency where the on-chip capacitor impedance becomes higher > than the target impedance. Below that frequency, current must come in > through the inductor. A good way to look at this is a power filter > with lots of series inductors and parallel capacitors, all the way from > the VRM to the chip. A better way to look at it is a transmission line > that has an impedance the same as the target impedance. Close to the > chip, the lumped elements of the transmission line must be small enough > to support the frequency that the chip demands, possibly GHz. As you > move away from the chip and closer to the VRM, the transmission line > does not have to support as high of frequency at each lumped element. > The L's and C's have longer time constants. But the impedance of the > transmission line (sqrt(L/C)) must still meet the target impedance or > else voltage droop will occur along the way. > > regards, > Larry Smith > Sun Microsystems > > > Date: Wed, 08 May 2002 14:05:40 -0700 > > From: Zhiping Yang <zhiping@xxxxxxxxx> > > X-Accept-Language: en > > MIME-Version: 1.0 > > To: Larry.Smith@xxxxxxx > > CC: si-list@xxxxxxxxxxxxx, sghsu55@xxxxxxxxxxxx > > Subject: Re: [SI-LIST] Re: Frequency criterion in power plane_Power > > integrity > > Content-Transfer-Encoding: 7bit > > > > Larry, > > > > I can not fully agreed with you on the conclusions from your example. > > Here is the plot: > > > > ______node 1_____Lpackage___ node2 > > | > > Zpcb > > | _____Gnd1__________________GND2 > > > > As a designer, you may only care about the voltage noise at the die side > > node2 (Node2-Gnd2). In your example, the 5% voltage drop you > > mentioned is across the Lpackage (Node1-node2), which may not > > be important. If you correctly select the Zpcb( for example, let it be > > capacitive), you may be able to get higher operating frequency than > > the frequency you mentioned in the examples for the same package. > > > > Thanks. > > > > Zhiping > > > > Larry Smith wrote: > > > > > Sogo - the current transition time has a lot to do with inductance and > > > current magnitude. The governing equation is V = L*di/dt. Usually, we > > > do not want to see more than 5% voltage drop on our power rail and that > > > sets V in the above equation. L is a property of the packaging. The > > > core power path for a microprocessor might have 50pH loop inductance. > > > The current transient might be 50 amps from a 1 V supply. Now that we > > > have defined L, di and V, we can calculate the transition time, dt = > > > L*di/V = 50pH*50A/.05V=50nSec. > > > > > > In other words, if we try to draw 50 amps out of a 50pH inductance in > > > less than 50 nSec, the voltage is going to droop more than 5% of 1V. > > > The GHz frequency associated with a 50nSec rise time is > > > 0.35/50nSec=7MHz. This example has been for a microprocessor whose > > > core demands lots of current at a low voltage. The fast 50 amp current > > > transient simply is not going to make it out of the 50pH package > > > inductance. > > > > > > Let's take another example of a 3.3V memory Dimm that draws 1 amp of > > > transient current. Perhaps the equivalent loop inductance for the Dimm > > > power supply is 1 nH. By the same calculation, dt = > > > 1nH*1A/(3.3*.05)=6nSec. The frequency associated with 6nSec is > > > .35/6nSec=57MHz. In other words, the Dimm can draw power from the > > > mother board from DC up to 57 MHz. Above that frequency, the power > > > will have to come from onboard the Dimm. > > > > > > Please note that these calculations assume that all current is in the > > > Vdd/Gnd loop. If signal return current gets involved, the problem is > > > much more complicated. A simple target impedance between Vdd and Gnd > > > is a good starting point, but SSN analysis involving Vdd, Gnd and > > > signals should be done after that. > > > > > > regards, > > > Larry Smith > > > Sun Micorsystems > > > > > > > From: "sogo" <sghsu55@xxxxxxxxxxxx> > > > > To: "'Larry Smith'" <ldsmith@xxxxxxxxxxxxxxxxxx>, > > > > <si-list@xxxxxxxxxxxxx> > > > > Subject: RE: [SI-LIST] Frequency criterion in power plane_Power > > > > integrity > > > > Date: Wed, 8 May 2002 05:36:44 +0800 > > > > MIME-Version: 1.0 > > > > Content-Transfer-Encoding: 7bit > > > > X-Priority: 3 (Normal) > > > > X-MSMail-Priority: Normal > > > > X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2600.0000 > > > > Importance: Normal > > > > > > > > Hi Larry, > > > > I'm pleasure to get your message and appreciate with lots of paper you > > > > wrote in this topic. Thank for your detail answer. I'm wonder the > > > > frequency band in PC board is so low, only 35MHz. In my project, I would > > > > like to simulate the power impedance of PDN in MB. However, the PDN of > > > > MB is linked to DDR through DIMM connector. I'm curious that why you > > > > define the rise time of PC board only 10ns. If the transient current is > > > > from northbridge, I believe the transient time is a couple of ns > > > > recently. That is said, the frequency band is about 350MHz. Is it right? > > > > In this frequency range, single node modeling of power plane is not > > > > sufficient. I have done some simulations, the results indicated the > > > > accurate model of power plane is necessary in modern high speed digital > > > > system design. > > > > Thank for your assistances in advance. > > > > Best regards > > > > > > > > -----Original Message----- > > > > From: Larry Smith [mailto:ldsmith@xxxxxxxxxxxxxxxxxx] > > > > Sent: Wednesday, May 08, 2002 12:10 AM > > > > To: si-list@xxxxxxxxxxxxx; sghsu55@xxxxxxxxxxxx > > > > Subject: Re: [SI-LIST] Frequency criterion in power plane_Power > > > > integrity > > > > > > > > Sogo - This is a very important question. By using the target > > > > impedance concept in the frequency domain, we can guarantee that a > > > > certain voltage tolerance is met in the time domain. But, at how high > > > > of frequency must we meet the target impedance? Obviously, we don't > > > > have to meet the target impedance at 1 teraHertz. > > > > > > > > The answer depends upon the rise and fall times of the current > > > > transient. The rise time is different at various positions of the > > > > circuit. The current transition times at several circuit positions > > > > might be as follows: > > > > > > > > circuit position tRise freq > > > > ----------------- ------- ---- > > > > power connector 1 uSec 350 kHz > > > > pc board 10 nSec 35 MHz > > > > silicon gates .1 nSec 3.5 GHz > > > > > > > > The frequency is found by the formula 0.35(GHz)/tRise(nSec). Most of > > > > the energy associated with the rising or falling edges is below this > > > > frequency. If your power distribution system meets target impedance at > > > > the calculated frequency and all the way down to DC at the various > > > > points of the circuit, you are guaranteed to have a supply that stays > > > > within the voltage tolerance used in the target impedance calculation. > > > > > > > > regards, > > > > Larry Smith > > > > Sun Microsystems > > > > > > > > > Delivered-To: si-list@xxxxxxxxxxxxx > > > > > X-eGroups-Return: sghsu55@xxxxxxxxxxxx > > > > > Date: Mon, 06 May 2002 04:12:53 -0000 > > > > > From: "sogo_hsu" <sghsu55@xxxxxxxxxxxx> > > > > > To: si-list@xxxxxxxxxxxxx > > > > > Subject: [SI-LIST] Frequency criterion in power plane_Power integrity > > > > > User-Agent: eGroups-EW/0.82 > > > > > MIME-Version: 1.0 > > > > > X-Originating-IP: 202.39.18.4 > > > > > Content-Transfer-Encoding: 8bit > > > > > X-Approved-By: Raymond.Anderson@xxxxxxx > > > > > X-ecartis-version: Ecartis v1.0.0 > > > > > X-original-sender: sghsu55@xxxxxxxxxxxx > > > > > X-list: si-list > > > > > > > > > > > > > > > Hi all, > > > > > In high-speed digital system, power integrity becomes an important > > > > > issue right now. Some one evaluated power integrity in the quanity of > > > > > frequency dependent impedance of power plane, such as target > > > > > impedance. As we know, the impedance shall be maintained over a wide > > > > > band. But, what's the criterion? Knee frequency ? bandwidth of clock > > > > > pulse? or harmonics of clock? > > > > > Thanks in advance. > > > > > Best regards, > > > > > Sogo > > > > > > > > > > ------------------------------------------------------------------ > > > > > 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 > > > > > ------------------------------------------------------------------ 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