Chris, you won't find much disagreement here on the need to get the Si right first, but there is this problem of controlling the Si vendors, and there is also a certain amount of scale to watch out for. Eventually, we are going to get into a corner where no amount of band-aids will work, and maybe then all the Si vendors will get religion when people just can't use their parts otherwise, but I think we are still far away from that point. Steve. At 07:46 PM 1/8/2004 -0800, Chris Cheng wrote: >Band-aids and hemorrhaging are the correct descriptions here. >I sat on both sides of the table (as a system designer and a semiconductor >vendor) before and being a lazy person, I always find excuses to blame the >other side :-D. >Since I am now in a system house, I think the majority of decoupling work >can/should be done at the Si/packaging level. >I have long make the analogy that core power distribution is like a Japanese >water fountain, you start with the smallest/fastest bucket/cap on chip, move >on to medium size decoupling on package and end up with the slowest bulk on >system. If you don't follow that, you will have more than a single problem >in your hand. i.e. If you see >100MHz EMI radiating from your package, you >will probably have a large >100MHz core power ripple on die. Instead of >building a ground cage around your package or using thin core PCB, you are >better off beating your Si vendor and have them bulk up their on die >decoupling. >Same on I/O power. Instead of multiple power planes for each I/O power or >crazy BC PCB, make sure your I/O has enough on die decoupling and assume the >pull up return current from gnd reference. Just make sure you have enough >gnd via on your PCB and you are done. >If you don't buy what I say, at least give me credit for being consistent. >Read any threads I ever made in this forum dated back to the beginning and >you will realize I sound like a broken record no matter whether the normal >system speed at the time is 50MHz or 5GHz. And I practice what I preach, >honest. No BC and nothing below .01uF on PCB. >Simple and cheap. > >-----Original Message----- >From: steve weir [mailto:weirsp@xxxxxxxxxx] >Sent: Thursday, January 08, 2004 5:18 PM >To: Chris.Cheng@xxxxxxxxxxxx; 'scott@xxxxxxxxxxxxx'; Zhangkun; silist >Subject: Re: [SI-LIST] Re: Power Supply Distribution/Filtering/Decouplin >g Guide > > >Chris, > >I think the devil comes back to what we can control. I believe that in >your work you have been able to influence the IC packaging sufficiently >that best practices on signal / image have been closely followed. I think >that has a huge demonstrated effect on system cost and performance. > >Part of the discussion is on ensuring that at chip-scale those practices >are followed. Part of the discussion has been on quantitative evaluation >of how well a design does before we build it. And part of the discussion >has been on what band-aids can be applied to a poor package design that is >hemorrhaging. In the long-run probably no amount of band-aids will work. > >I think the 100MHz number is low for a lot of cases. Somewhere north of >400MHz, I agree, because of the inductive wall presented by the package >interconnect. However, given the amount of single ended signaling above >100MHz on high density boards, it can get pretty tough to image only one >rail, and there is a lot of energy out there well above 100MHz. > >I also think that we need to be careful about any hard frequency >limits. As we keep moving up the power curve, even low percentages of >power become substantial absolute values. > >Regards, > > >Steve. > >At 01:41 PM 1/8/2004 -0800, Chris Cheng wrote: > >Scott, Steve, Zhangkun and friends, > >I can appreciate the benefit of an optimized via pad for decoupling caps on > >PCB, it certainly helps. I also can appreciate sophisticate power analysis > >CAD tools for power distribution simulation. > >However, I have done enough package and chip power distribution analysis > >(both in Istvan's company and another processor company) that I believe a > >properly designed chip/package does not need any core power decoupling > >higher than 100MHz at the PCB level. I also believe that I/O power > >distribution is a matter of image current/reference plane management rather > >than how low your power plane or PCB cap impedance is. > >So given the above assumptions, does all these via analysis or PCB plane > >simulation tools matters at frequencies below 100MHz ? > >Chris > > >------------------------------------------------------------------ >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 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 > ------------------------------------------------------------------ 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 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