Two small details I'd like to add to the discussion: 1) Don't forget that it is sometimes beneficial to have a filter that works both ways. For example: If you have a noisy circuit (I leave it to the imagination to the reader what this could be) you don't want its noise to show up on the supply rail shared with other circuitry. (One filter in the right place may save you from having multiple filters in many places.) 2) I must say I still see designs that forget about return path (Ground) isolation/filtering where different supply voltages may share the same return path. Sometimes this can be very important as well. Especially if digital, analog and RF circuitry are in close proximity on the same board. Best Regards =20 Dipl. Ing. Juergen Flamm, C.I.D. FAE Cadence SPB TFO-NA 818-642-6255 http://www.cdnusers.org -----Original Message----- From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx] On Behalf Of Istvan Novak - Board Design Technology Sent: Tuesday, April 11, 2006 12:18 PM To: leeritchey@xxxxxxxxxxxxx Cc: Joel Brown; Zhangkun; si-list@xxxxxxxxxxxxx Subject: [SI-LIST] Re: Ferrite bead question Lee, To be fair to the people who write those application notes, they have to do it blindly: they have no way of knowing the details of the environment the chip goes into. In those cases, when the use of the ferrite bead is warranted, the purpose of the filter is to provide the necessary feed impedance at the filter's output AND to provide the necessary isolation between the filter's input and output. The necessary feed impedance at the output is something that the person writing the application note should be able to know. The required isolation, however, depends not only on the filter, but also on the noise characteristics at the filter input. This means that the necessary filter does depend on the circuit environment. There are scenarios, as you pointed out, when there is no need for a separate filter; these are the cases when the noise before the 'would-be' filter is low enough so that no additional attenuation is required AND the main rail's impedance is appropriate for directly connecting it to the 'filtered' device. If, however, the main rail has more noise than what the 'filtered' device can live with, we have two choices: we could lower the noise on the main rail, or we can use a filter. Devices, which come with a recommended extra filtering, are usually clock sources, clock buffers, SerDes analog circuitry. The noise sensitivity of these analog pins may require a noise voltage allowance much lower than what a generic logic may need. For instance, PLL supply points may be very sensitive to periodical disturbances around the loop cutoff frequency. Sometimes the noise should not be more than a few mV rms in that frequency range.=20 DC-DC converter output ripple could be more than this limit, and the generic logic may be OK with it. In such cases reducing the output ripple of the converter, though technically might be doable, would probably be more expensive than requiring a dedicated filtering to the lower-current analog point. You are correct in saying that many of these filters are not designed in a proper way. One way of doing it properly would be to get the noise tolerance and feed-impedance requirement for the chip in question, instead of a recommended filter. Based on the noise characteristics of the system that goes around the chip, the system designer should be able to determine whether a filter is needed, and if needed, should be able to design it. Regards, Istvan Lee Ritchey wrote: >Joel, > > >As others have observed, when the solution using a ferrite bead has=20 >been engineered to satisfy the following set of conditions it is=20 >warranted and helpful. > >1. There is a problem that needs fixing. >2. The proposed solution fixes the problem. >3. The proposed solution does not cause new problems. >4. The proposed solution is the best solution to the problem. > >When you apply this set of tests to the usual applications note=20 >recommendation, it often fails test number 1. The proponent has not=20 >acertained that there is a problem and what it actually looks like. > >I put this question to the students in my classes at UC Berkeley,=20 >nearly always practicing engineers. Can you show me an applications=20 >note with analysis supporting the use of ferrite beads? So far, there=20 >have been no such examples. I keep hoping, but know that most=20 >applications notes are not prepared that way. Soon as I find one, I'll >report it in this forum and show it in my classes- and give its author=20 >high praise for doing good engineering. > > More commonly, they are justified because "we have always done it=20 >this way." Further, there has usually not been any exhaustive testing=20 >to insure the solution is valid. That is in no way good engineering. > >I'll accept a recommendation that is accompanied by supporting=20 >engineering analysis. Less, seems to me to be guessing and that's a=20 >good way to get in trouble with modern e lectronic components. > > > > > =20 > >>[Original Message] >>From: Joel Brown <joel@xxxxxxxxxx> >>To: <leeritchey@xxxxxxxxxxxxx> >>Cc: Zhangkun <zhang_kun@xxxxxxxxxx>; <si-list@xxxxxxxxxxxxx> >>Date: 4/10/2006 10:17:06 AM >>Subject: Re: [SI-LIST] Re: Ferrite bead question >> >>Lee, >> >>After reading the responses and giving some more thought to this, I=20 >>agree that in some situations the use of ferrite beads can be detrimental. >>In the design I am currently working on, the ferrite beads are used=20 >>mostly on low speed I/O signals and DC power to ICs that are not=20 >>running at very high speeds. >>The highest speed part is a 100 MHz PCI Express clock buffer, and the=20 >>manufacturer does recommend to use a ferrite bead. >>While I agree your statements have some merit, I find it problematic=20 >>that you seem to be saying ferrite beads only work by luck. >>Certainly there must be situations in which they can predictably be=20 >>used to reduce noise. >> >>Regards - Joel >> >> >> >>Lee Ritchey wrote: >> =20 >> >>>Joel, >>> >>>I agree with all the respondents who have cautioned you on the use of >>>ferrite beads in the power leads of devices. In all of the cases, I >>> =20 >>> >have > =20 > >>>investigated, ferrite beads have been used to treat a symptom rather >>> =20 >>> >than > =20 > >>>the problem. The problem being that the power delivery system had=20 >>>too >>> =20 >>> >much > =20 > >>>noinse or ripple on it. >>> >>>Lately, I have seen many cases where placing a ferrite bead in the=20 >>>power lead of a device, very often the power lead of very high speed=20 >>>serdes, >>> =20 >>> >has > =20 > >>>made the part perform poorly or below spec. >>> >>>Yes, thousands of applications notes insist that ferrite beads be >>> =20 >>> >placed in > =20 > >>>the power leads and the applications engineers will demad you use=20 >>>them >>> =20 >>> >or > =20 > >>>they won't guarantee the circuit will work properly. The other side=20 >>>of that same coin is that when you ask if they will guarantee that=20 >>>the >>> =20 >>> >circuit > =20 > >>>will work properly if you follow the applications note exactly, the >>> =20 >>> >answer > =20 > >>>is still no! The reason is that there has been no analysis to prove >>> =20 >>> >that > =20 > >>>the use of a ferrite bead is a good idea.=20 >>> >>> I have measured wavforms of a 3.12 GB/S serial link where the eye=20 >>>just barely makes the minimum signal amplitude with the ferrite bead=20 >>>in place and has large amounts of margin when the ferrite bead is shorted out. >>>This, even with the manufacturer's recommended capacitors attached.=20 >>> =20 >>> >This > =20 > >>>data was taken from a demo board supplied by the manufacturer who was >>> =20 >>> >not > =20 > >>>aware of the degradation caused by the ferrite bead. >>> >>>The right solution is to design a power delivery system with power >>> =20 >>> >supply > =20 > >>>noise within the limits of the circuits being supplied and this is=20 >>>not >>> =20 >>> >very > =20 > >>>difficult to do. >>> >>>I have done the SI rule generation for three terabit routers with=20 >>>thousands of 3.125 GB/S and 4.8 GB/S serial links in them and used no >>>ferrite beads in them any where and they worked to specification the >>> =20 >>> >first > =20 > >>>time wit hproper margins. So maybe people who want you to use=20 >>>ferrite beads should be challenged with why they want you to add these parts. >>> >>>My fellow SI engineers call ferrite beads "voodoo" beads and "get lucky" >>>beads for good reason. >>> >>>This is not likely to make the ferrite bead salesmen happy, I'm sure. >>> >>> >>> =20 >>> =20 >>> >>>>[Original Message] >>>>From: Zhangkun <zhang_kun@xxxxxxxxxx> >>>>To: <joel@xxxxxxxxxx>; <si-list@xxxxxxxxxxxxx> >>>>Date: 4/10/2006 1:14:21 PM >>>>Subject: [SI-LIST] Re: Ferrite bead question >>>> >>>>Dear Joel >>>> >>>>As to my experience, it is not enough to select bead based on the P/N. >>>> =20 >>>> =20 >>>> >>>The decoupling circuit should be considered very carefully. >>> =20 >>> =20 >>> >>>>In several cases, we use beads and the noise become larger. The >>>> =20 >>>> >following > =20 > >>>> =20 >>>> =20 >>>> >>>element should be considered: >>> =20 >>> =20 >>> >>>> 1.The resonance between bead and capacitor. >>>> 2.The frequency span of the noise source. >>>> >>>>A simulation is suggested. It is very simple. >>>> >>>>Hope this is helpful >>>> >>>>Best Regards >>>> >>>>Zhangkun >>>>2006.4.10 >>>>----- Original Message ----- >>>>From: "Joel Brown" <joel@xxxxxxxxxx> >>>>To: <si-list@xxxxxxxxxxxxx> >>>>Sent: Saturday, April 08, 2006 2:33 AM >>>>Subject: [SI-LIST] Ferrite bead question >>>> >>>> >>>> =20 >>>> =20 >>>> >>>>>I often see recommendations in application notes for ferrite beads=20 >>>>>on >>>>> =20 >>>>> =20 >>>>> >>>DC >>> =20 >>> =20 >>> >>>>>power lines. Sometimes a particular P/N is specified and sometimes >>>>> =20 >>>>> >just > =20 > >>>>>an impedance is given. >>>>>Wouldn't I want to choose a ferrite bead that has the highest >>>>> =20 >>>>> >impedance > =20 > >>>>>possible at the frequencies of interest, and that the current=20 >>>>>rating >>>>> =20 >>>>> =20 >>>>> >>>and >>> =20 >>> =20 >>> >>>>>DC resistance are appropriate? Would the frequencies of interest be >>>>> =20 >>>>> >the > =20 > >>>>>fundamental switching frequency of the part and possibly the 3rd=20 >>>>>and >>>>> =20 >>>>> =20 >>>>> >>>5th >>> =20 >>> =20 >>> >>>>>harmonics? >>>>>Thanks - Joel >>>>> >>>>> >>>>> >>>>> >>>>> =20 >>>>> > =20 > ------------------------------------------------------------------ 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: =20 //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 =20 ------------------------------------------------------------------ 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