[SI-LIST] Re: Ferrite bead question
- From: Istvan Novak <istvan.novak@xxxxxxx>
- To: scott@xxxxxxxxxxxxx
- Date: Mon, 10 Apr 2006 22:15:05 -0400
Scott,
Further to your excellent points, we should keep in mind that the 78
milliohms at 1.5GHz
in your example represents (or rather requires) a less than 10pH
inductance in a lumped
model. To make sure that horizontal inductance does not limit us, this
would require
a power/ground plane pair with a fraction of a milliinch separation. In
real circuits the
inductance is usually much higher, but the fact that we phisically have
to spread out the
noise sources (signal terminations in your example) will help to
compensate for this.
Regards,
Istvan
Scott McMorrow wrote:
>Hmmm ...
>Lets take an example of a 533 MT/s DDR bus, with a 266 MHz frequency.
>
>One bit time = 1.875 ns
>One period = 3.75 ns
>Let risetime = 400 ps (10%-90%)
>Fknee = .35/400ps = 875 MHz
>Assume that driver current is 20 mA/driver
>Assume that 64 drivers switch simultaneously
>Total current transient = 1.28 Amps in 400 ps.
>Assume total voltage swing = 2V pp for SSTL-2.5
>Assume a 100 mV power noise allowance (5%)
>Power system Ztarget = 100mV/20 mA = 78 mOhms
>
>Computing the Fourier Coefficients for a trapezoid, using a cute little
>calculator I found here:
>http://www.eecircle.com/applets/001/001.html, with 400 ps rise time
>we get:
>
>Harmonic component
>DC - 0.500
>1st - 0.312 (266 MHz)
>3rd - 0.089 (800 MHz)
>5th - 0.037 (1.33 GHz)
>7th - 0.013 (1.87 GHz)
>9th - 0.001 (2.4 GHz)
>
>For this particular driver, 62% of the AC energy is contained at the
>fundamental frequency, 17.8% is contained at the 3rd harmonic, 7.4% is
>contained at the 5th harmonic, and 2.6% is contained at the 7th
>harmonic. If we use our one pole filter bandwidth formula of BW =
>0.35/risetime, then we are covered up to the 3rd harmonic of the signal
>fundamental. But, 20% of the signal energy is still carried in harmonics
>above the 3rd. Our total power system noise in the time domain is
>approximately:
>
>Vnoise = 1.28 Amps x ( Z@266 MHz x 0.62 + Z@800MHz x 0.178 + Z@xxxx GHz
>x 0.074 + Z@xxxx GHz x 0.026 + Z@xxx GHz x 0.02)
>
>It is not unusual for power system impedance at high frequency
>resonances to exhibit a 100:1 increase over the Ztarget at lower
>frequencies. What if our power system had flat impedance of 78 mOhms
>out to 1.5 GHz and then climbed to a resonance peak of 7.8 ohms at 1.87 GHz.
>
>Ignoring the 9th harmonic, the noise would be:
>
>Vnoise = 1.28 Amps x ( 0.078 x 0.62 + 0.078 x 0.178 + 0.078 x 0.074 +
>7.8 x 0.026)
>Vnoise = 1.28 Amps x ( 0.078 Ohms x .872 = 7.8 x 0.026)
>Vnoise = 1.28 Amps x ( 0.068 + .2028)
>Vnoise = 87 mV below 1.5 GHz + 260 mV @ 1.87 GHz
>
>With a reactive network with a resonance such as this, the resonance
>peak would generally be out of phase with the lower frequency noise, so
>the result would not be generally additive. However, that 1.87 GHz peak
>would dominate, and show up as ringing in the power system. I suspect
>it would also show up as a healthy signature in the EMI profile,
>although I am not expert in these matters. Which is why those silly EMI
>engineers are often concerned about harmonics of the fundamental like
>the 7th, 9th and even the 11th.
>
>This particular example is contrived and not quite exact. But, it is
>not unusual at all in designs with planes fragmented like a map of the
>Balkans, as Dr Johnson would say. Or in large planes, where planar
>resonances are right in the middle of all the switching harmonics.
>We've measured power systems where there are some pretty interesting
>high frequency resonance peaks, which show up quite well in both the
>power system and time domain signal measurements, and are above Fknee.
>Remember, according to Dr Johnson, amplitude at the knee frequency is
>6.8 dB below the 20 dB/decade rolloff seen in a signal's spectral
>energy. At the 9th harmonic (which can be approximated as one decade
>down) we have a total spectral attenuation of 26.8dB, or 21.9 times
>lower the the fundamental. It does not take much resonance "gain" over
>the power system target impedance to dwarf noise at the fundamental with
>noise at a higher harmonic ... even one beyond the knee. What's quite
>interesting about this, is that this sort of power system problem would
>show up as high frequency ringing on a DDR memory bus signal. But, it
>would not have been diagnosed with standard signal integrity tools, and
>would only show up when a complete simulation of simultaneous switching
>of the bus was performed that included modeling of the device, package,
>power planes, bypass capacitors, mounting pads and vias.
>
>This goes back to what I term the "Wack-a-Mole" phenomenon in power
>distributions systems. If you do not fix the fundamental problem of
>power system resonances, then all you're doing is moving energy around
>from one place to another, just to see it pop up out of another hole.
>
>
>regards,
>
>scott
>
>
>Scott McMorrow
>Teraspeed Consulting Group LLC
>121 North River Drive
>Narragansett, RI 02882
>(401) 284-1827 Business
>(401) 284-1840 Fax
>
>http://www.teraspeed.com
>
>Teraspeed® is the registered service mark of
>Teraspeed Consulting Group LLC
>
>
>
>Eric Goodill wrote:
>
>
>>Ken,
>>Yeah, that's pretty much exactly what I meant by "pick a probable
>>maximum frequency," but I didn't say that in my post. Sorry.
>>
>>-Eric
>>
>>kfrobinson@xxxxxxxxxx wrote:
>>
>>
>>
>>>Eric,
>>>The frequency components in a digital system can be estimated from DC to
>>>.35/(rise time). The IC manufactures do give 10-90 rise time. Pick the
>>>fastest rise time IC in your system.
>>>Ken
>>>
>>>-----Original Message-----
>>>From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx]
>>>On Behalf Of Eric Goodill
>>>Sent: Monday, April 10, 2006 12:16 PM
>>>To: si-list@xxxxxxxxxxxxx
>>>Subject: [SI-LIST] Re: Ferrite bead question
>>>
>>>Howdy,
>>>One thing I've never understood about designing the frequency response
>>>of a power delivery system is what are the frequency requirements? That
>>>is, where does the vendor give me the spectrum of the current demand for
>>>the part? They don't (I'm skipping over they it'd be tough). So the only
>>>solution I can see is to pick a probable maximum frequency and say from
>>>DC to that frequency we need less than X ohms impedance based on some
>>>assumptions about the peak current needs (related to lack of current
>>>spectrum). This is probably over designing, but I don't have any other
>>>data to use. Do others feel the same way?
>>>
>>>-Eric
>>>
>>>Scott McMorrow wrote:
>>>
>>>
>>>
>>>
>>>>Joel,
>>>>I think Lee is being a little bit black and white, but not without
>>>>good reason.. Whether or not ferrites "work" in a power filtering
>>>>design is a matter of whether the engineering was performed. What I
>>>>think Lee often sees are systems where no engineering has gone into
>>>>the design of ferrite, and they have been thrown into a circuit
>>>>without thought. Often because "that's the way we've always done it"
>>>>or "that's that way the competitors do it" or because "that's the only
>>>>
>>>>
>>>>
>>>
>>>
>>>
>>>>ferrite we have in our parts system." The inductance of a ferrite can
>>>>
>>>>
>>>>
>>>
>>>
>>>
>>>>interact with the power system and capacitors to form a pretty nasty
>>>>resonance. This resonance often sits in the low frequency region,
>>>>around the VRM output transition region, and has a strong tendency to
>>>>cause peaking in the 100 kHz to 1 MHz range. This peaking can cause
>>>>noise modulation of the Serdes, which Lee has often observed. If the
>>>>PCB power delivery network has a lower impedance in the frequencies
>>>>that affect the Serdes, than they ferrite filter does, then shorting
>>>>the ferrite will help. But if a ferrite filter network is correctly
>>>>designed, it is well matched and does not cause peaking.
>>>>
>>>>Whether or not noise injected into a Serdes affects it's output is a
>>>>function of the internal PLL/DLL design. This is hardly ever
>>>>specified, but can be measured.
>>>>
>>>>Scott
>>>>
>>>>Scott McMorrow
>>>>Teraspeed Consulting Group LLC
>>>>121 North River Drive
>>>>Narragansett, RI 02882
>>>>(401) 284-1827 Business
>>>>(401) 284-1840 Fax
>>>>
>>>>http://www.teraspeed.com
>>>>
>>>>Teraspeed(r) is the registered service mark of Teraspeed Consulting
>>>>Group LLC
>>>>
>>>>
>>>>
>>>[snipped off the reset of the thread]
>>>
>>>
>>>-- Binary/unsupported file stripped by Ecartis --
>>>-- Type: text/x-vcard
>>>-- File: ericg.vcf
>>>
>>>
>>>
>>>
------------------------------------------------------------------
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
Other related posts:
