[SI-LIST] Re: One stitching via or more vias is better for 25Gbps
- From: Scott McMorrow <scott@xxxxxxxxxxxxx>
- To: Eric Bogatin <eric@xxxxxxxxxxxxxxx>
- Date: Mon, 8 Sep 2014 15:04:35 -0400
Eric
You are correct in saying "The diff signal mostly overlap in the cavity".
However, you are incorrect in saying "a pure diff signal does a poor job
of exciting cavity resonances." It depends on the signalling rate, the
signal edge rate, and the presence of resonances in the planar cavity.
For vias carrying ideal differential signalling, there are only two angles
within a planar cavity where the sum total of the signal injection is zero.
If we place the vias on the x-axis, symmetrically east and west of the
y-axis center line, and if we define 0 degrees to be north, then the only
two directions where net noise injection into the planes is zero is at 0
and 180 degrees. At 90 and 270 degrees, we lose the differential
cancellation because of the separation distance between the two via dipoles.
If we have a material with a dielectric propagation velocity of 160 ps/in,
and we have two differential vias spaced 40 mil apart, then the electrical
length between the two vias is 160ps/in x 0.040 in = 6.4 ps. Along the axis
of the two vias, the dipole currents are out of phase by 6.4 ps. At 28
Gbps the bit interval is 35.7 ps, so the phase difference is approximately
17% of a bit time. This is not insignificant, especially when you consider
that 28 Gbps drivers typically have edge rates in the 5 to 10 ps range.
As a result, even with theoretically ideal differential systems, spatial
separation between the vias will cause an angular asymmetry in the injected
radiation pattern within planar cavities. That asymmetry in current
injection will excite high-Q resonances, and will also cause crosstalk
between differential via pairs. Those resonances will be formed either by
boundary conditions at the edge of the planes, or at ground stitch vias, or
at other non-signal power vias (in which case some terrible horrible no
good very bad things can happen.)
As an experiment, take a differential via pair and have it transition
between floating plane between two ground planes. Make the floating plane
large enough to have resonances in the x and y directions that are within
your measurement bandwidth Feel free to place the vias dead center in the
plane. Then perform a VNA sweep of the vias. You will see the floating
plane resonances light up. Now place a second set of differential pair
vias through the resonant cavity and measure crosstalk between the two
pairs of vias. You will find that crosstalk peaks at that very same planar
resonances. Now take that very same floating plane, and stitch around all
the edges of the plane to ground. Measure again, and you will see nearly
identical resonances. This time, the phase of the resonances has changed,
because the open circuit boundary has been changed to a short circuit, and
it has increased frequency, since vias take up finite space and shrink the
dimensions of the plane that are resonating.
Don't get me wrong. Real boards have skew and common mode components that
must be dealt with. But even ideal differential pairs can cause excitation
of planes, and the associated crosstalk that occurs. And very bad
excitation at that. Remember that in 28G differential signalling we are
trying to reduce Tx-to-Tx and Rx-to-Rx FEXT to -45 dB and Tx-to-Rx NEXT to
-65 dB. These are very small numbers.
Scott
Scott McMorrow
Teraspeed® Consulting - A Division of Samtec
16 Stormy Brook Rd
Falmouth, ME 04105
(401) 284-1827 Business
http://www.teraspeed.com
On Mon, Sep 8, 2014 at 11:39 AM, Eric Bogatin <eric@xxxxxxxxxxxxxxx> wrote:
> Guys-
>
>
> When a diff signal transitions from one Vss return plan to another, the
> return currents of the diff signal mostly overlap in the cavity and a pure
> diff signal does a poor job of exciting cavity resonances between the two
> Vss planes.
>
>
>
> It is not necessary to have "stitching" vias in a cavity if you are only
> transitioning differential signals.
>
>
>
> However, it is impossible to engineer a real interconnect that is so
> symmetrical as to not have any mode conversion. This means there will
> always
> be some common signal components along with the diff signals. It's the
> common signals which will excite the cavity modes, as Steve has been
> emphasizing.
>
>
>
> When you do a simulation, unless you are clever enough to anticipate a
> possible asymmetry in the interconnect and include it in the simulation,
> you
> will not generate common signals in your simulation and the diff response
> through the cavity may look just fine.
>
>
>
> How big an issue is the common signal in the cavity? .it depends. If you
> are
> really, really lucky, you may have only -30 dB of common signal (3% of your
> diff signal) and the common noise injected in the cavity modes may not be
> enough to couple anywhere and screw up the system.
>
>
>
> But, if you are not so lucky, you may have -15 dB common signal. This is
> still within many specs, not a problem by itself, but will contribute 15%
> of
> the diff signal as common current into the cavity and screw up channel to
> channel cross talk and ground bounce to some other channel or couple out as
> EMI.
>
>
>
> If you want a robust system, use shorting vias in the cavity to suppress
> modal resonances- how many? It's not just about 1 or 2 per diff via. The
> shorting vias are to suppress model resonances. Their spacing is important.
> You will see the first model resonance with a wavelength equivalent to 6 x
> the via to via spacing. For a 1 GHz lowest excited mode, this is 1 inch
> apart.
>
>
>
> If you want to rely on luck in your system, maybe the first few, maybe even
> the first 20 boards will work, but as soon as you have too much mode
> conversion in a channel, your product may fail.
>
>
>
> If you are interested in this topic, I cover it in the Advanced Gigabit
> Channel Design (AGCD) class in the SI Academy, (www.beTheSignal.com) and
> in
> my graduate course at CU. I plan to post my lectures from this course on
> the
> SI Academy by the end of the year.
>
>
>
> --eric
>
>
>
>
>
>
>
>
>
> *******************************************************
> Dr. Eric Bogatin, Dean
>
> <http://www.bethesignal.com/> Teledyne LeCroy Signal Integrity Academy
>
> <http://www.bethesignal.com/> www.beTheSignal.com
>
> Adjunct Prof, ECEE Univ of Colorado, Boulder
>
> 105 S Sunset St, Suite J
>
> Longmont, CO 80501 USA
>
> Twitter @beTheSignal
> e: <mailto:eric@xxxxxxxxxxxxxxx> eric@xxxxxxxxxxxxxxx
>
> cell: 913-424-4333
> ******************************************************
>
> Msg: #6 in digest
>
> From: David Banas <DBanas@xxxxxxxxxx>
>
> Subject: [SI-LIST] Re: One stitching via or more vias is better for 25Gbps
>
> Date: Mon, 8 Sep 2014 14:38:15 +0000
>
>
>
> On 9/7/2014 8:49 PM, leeritchey@xxxxxxxxxxxxx
> <mailto:leeritchey@xxxxxxxxxxxxx%3cmailto:leeritchey@xxxxxxxxxxxxx>
> <mailto:leeritchey@xxxxxxxxxxxxx> wrote:
>
> > All of the things proposed don't seem to have any basis. I, on the
>
>
>
> > other hand, have built test boards to find out. My findings are
>
>
>
> > available in my classes and do not show such vias have any value.
>
>
>
> >
>
>
>
> Hi Lee,
>
>
>
> Our simulation results suggest that, while these stitching vias have no
> discernable effect on signal integrity at the receiver at these speeds,
> they
> do noticeably affect electromagnetic emissions from the PCB edges,
> potentially threatening electromagnetic compliance testing failure, when
> they are omitted. And I was wondering if you have any experimental
> findings,
> in this regard, that you can share.
>
>
>
> Thanks!
>
>
>
> David Banas
>
> Sr. Member of the Technical Staff
>
> Altera Corp.<http://www.altera.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 forum is accessible at:
> http://tech.groups.yahoo.com/group/si-list
>
> List archives are viewable at:
> //www.freelists.org/archives/si-list
>
> 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 forum is accessible at:
http://tech.groups.yahoo.com/group/si-list
List archives are viewable at:
//www.freelists.org/archives/si-list
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts:
