[SI-LIST] Re: Plane breaks - Diff pair crossing
- From: "Hassan O. Ali" <hassan@xxxxxxxx>
- To: "Bert Simonovich" <lambert@xxxxxxxxxxxxxxxxxx>
- Date: Mon, 24 Jan 2005 13:58:42 -0500 (EST)
Bert,
Thanks for sharing this insightful info. I wish to add just a few points -
probably to
help visualize the plane-split scenario better.
The differential impedance (for a differential-pair) is generally calculated as:
Zdiff = 2*Zodd = 2*sqrt(Ls-Lm)/sqrt(Cs+2Cm)
(please see eqn(2): http://www.tdasystems.com/library/appnotes/DIFF-1099-02.pdf)
In other words, Zdiff is also a function of the self per-unit inductance,
capacitance
(and for lossy lines, self per unit resistance and conductance) values. These
per unit
self parameters change drastically at the split giving rise to impedance
discontinuity.
A more intuitive formula (although not easily understood with physical
parameters) is
that provided in http://www.ultracad.com/diff_z.pdf
Zdiff = 2*(Z11-Z12)
where Z11 is the self impedance of each trace, and Z12 is some impedance
related to the
coupling of the two traces. Now, assume, before the plane split, Z11 is close
to 25ohm
(as we normally make it), and Zdiff is close to 100ohm (as we normally make
it).
Before the split, the plane under the trace contributes its inductance and
resistance to
Z11. Right at the split, there is no plane - just a small split capacitance and
a
potentially large split inductance. The split capacitance and inductance give
rise to a
large Z11 value especially at lower frequencies which results in a large Zdiff
value.
So, in SI terms, a differential signal at the slot sees a large impedance
discontinuity
and reflects badly. In EMC terms, signals going through a large loop (even in
differential mode) may cause EMI problems. And if there are even small
common-mode
currents, the slot will be excited and may also cause EMI problems.
Regards.
Hassan.
On Jan 24, "Bert Simonovich" <lambert@xxxxxxxxxxxxxxxxxx> wrote:
>
> In Dr. Howard Johnson's new book "High Speed Signal Propagation", he
> explains quite nicely the principle of differential pairs crossing a split
> plane as it pertains to return current behaviour. He refers to this as a
> "Differential U-Turn".
> In that chapter, he goes on to say there are significant high frequency
> return currents which flows on any reference plane beneath the differential
> traces. The differential pairs have mutual capacitance(Cm)from each trace to
> the reference plane as well as between traces. These are represented by the
> e-fields as shown on common 2D field solver tools. This Cm induces the
> return currents on the other trace as well as the reference plane. However,
> most of the return current will still return on the solid plane instead of
> the opposite trace because the trace widths are much narrower than the
> plane.
>
> When a differential pair crosses a split in the reference plane, the return
> current flows under each trace until it reaches the split. Since it cannot
> traverse the split in the plane, the return current in each direction does a
> U-turn as it moves along each edge of the split in opposite directions and
> returns under the opposite trace back to sources. For the width of the
> split, the current returns on the opposite trace as it normally would in the
> case where there is no ref plane.
>
> This whole action creates an inductive loop at the U-turn zone bounded by
> the width of the split in the plane, and the space between pairs. This
> inductive loop creates magnetic fields inducing crosstalk and EMI which will
> vary in proportion to the size of the U-turn zone. Crosstalk and EMI can be
> minimized by reducing the width of the split and spacing between diff pair
> traces.
>
> Regards,
>
> Bert Simonovich,
> Signal Integrity & Backplane Architecture
> NORTEL Networks
> Phone: (613) 763-7654
> e-mail: lambert@xxxxxxxxxxxxxxxxxx
>
>
> -----Original Message-----
> From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx] On
> Behalf Of steve weir
> Sent: Monday, January 24, 2005 10:51 AM
> To: ron@xxxxxxxxxxx; Grasso, Charles
> Cc: 'Chris.Cheng@xxxxxxxxxxxx'; 'Si-List'
> Subject: [SI-LIST] Re: Plane breaks - Presentation download
>
>
> Ron, I have to disagree a bit there too. A single differential pair will
> definitely fare an order of magnitude better crossing a slot, but any
> common mode energy from the net pair still excites the slot for EMC
> purposes and coupling shift from the 10% range up to the 30% range still
> presents an impedance discontinuity. Here, shrinking the gap to be a small
> fraction of Tr would seem to cause the least impairment. Regards,
>
>
> Steve
> At 07:13 AM 1/24/2005 -0800, Ron Miller wrote:
> >For a diff pair it is not so much of a problem because there is
> >coupling
> >from both
> >legs and they can be very tightly coupled, so the coupling into the slot
> >is balanced
> >and cancels.
> >
> >The problem is primarily with single traces.
> >
> >Ron
> >
> >Grasso, Charles wrote:
> >>
> >>Steve/Ron - Just to calibrate myself ... Now you DO
> >>mean a trace crossing a split orthogonally right?
> >>If so I fail to see how changing the width of the
> >>split makes things better except maybe in one condition.
> >>i.e. where a diff pair cross a split. The return current
> >>for one trace will be carried by its pair and the split will be almost
> >>invisible.
> >>
> >>The effect of crossing a split on a diff pair
> >>can be seen in a presentation by Ansoft Corp
> >>available for download from the RMCEMC website.
> >>Go to <<a href='http://www.ieee.org/rmcemc><a
href='http://www.ieee.org/rmcemc'>http://www.ieee.org/rmcemc</a>'>http://www.ieee.org/rmc
emc><a href='http://www.ieee.org/rmcemc'>http://www.ieee.org/rmcemc</a></a> the
link
> >>is on the front page. You'll need to go in about 12 pages or so...
> >>
> >>Best Regards
> >>Charles Grasso
> >>Senior Compliance Engineer
> >>Echostar Communications Corp.
> >>Tel: 303-706-5467
> >>Fax: 303-799-6222
> >>Cell: 303-204-2974
> >>Pager/Short Message: <mailto:3032042974@xxxxxxxx>3032042974@xxxxxxxx
> >>Email:
> >><mailto:charles.grasso@xxxxxxxxxxxx>charles.grasso@xxxxxxxxxxxx;
> >>Email Alternate: <mailto:chasgrasso@xxxxxxxx>chasgrasso@xxxxxxxx
> >>
> >>
> >>
> >>-----Original Message-----
> >>From: steve weir [<mailto:weirsp@xxxxxxxxxx>mailto:weirsp@xxxxxxxxxx]
> >>Sent: Saturday, January 22, 2005 4:19 PM
> >>To: <mailto:ron@xxxxxxxxxxx>ron@xxxxxxxxxxx;
> >><mailto:Chris.Cheng@xxxxxxxxxxxx>Chris.Cheng@xxxxxxxxxxxx
> >>Cc: Si-List
> >>Subject: [SI-LIST] Re: risetime effects of plane breaks
> >>
> >>
> >>Ron,
> >>
> >>Do you really mean to imply that the cross-talk falls substantially if
> >>the slot gap is increased to 2H or more? That is a new and very
> >>counterintuitive notion to me. I would be very interested in seeing
> >>any A/B model that could demonstrate such a phenomena.
> >>
> >>Regards,
> >>
> >>
> >>Steve.
> >>
> >>At 10:13 PM 1/21/2005 -0800, <mailto:ron@xxxxxxxxxxx>ron@xxxxxxxxxxx
> >>wrote:
> >>
> >>>
> >>>A few years ago Intel discovered that when a trace crosses a split it
> >>>can excite a transmission line mode into the slot between the panes
> >>>called "slot line" strangely enough. If the gap is small it works
> >>>quite well and all the traces crossing it become cross-talk for one
> >>>another.
> >>>
> >>>To avoid it make the gap at least 2 or 3 times the thickness of the
> >>>dielectric.
> >>>
> >>>ADS (Agilent) has a model for slot lines with the other transmission
> >>>lines.
> >>>
> >>>Ron
> >>>
> >>>Chris Cheng wrote:
> >>>
> >>>
> >>>>
> >>>>Scott,
> >>>>Excellent summary. That was my concern on striplines crossing with a
> >>>>bus rather than individual signals. In a way, it is like wire bond
> >>>>signal leads without the ground leads mixed among them. The signals
> >>>>start referencing each other instead. Or you can see it as a
> >>>>trade-off between adding shielding layers or spreading the bus
> >>>>spacing (decreasing routing channels) in a high density/performance
> >>>>design. My own rule of thumb is space them at least equal or larger
> >>>>than the gap itself when crossing. That's is at least a 3x decrease
> >>>>in routing channels so it is quite costly and has to be weight
> >>>>against adding shielding layers. Sometimes its worth it, sometimes
> >>>>its not. As for EMI, if you dig back some discussion I had with
> >>>>Steve, I always prefer solid ground planes referencing microstrips
> >>>>on top and bottom of PCB and then stitch the edges with ground vias.
> >>>>Hopefully any of those excited noise on the cut power planes will be
> >>>>trapped inside.
> >>>>
> >>>>-----Original Message-----
> >>>>From: Scott McMorrow
> >>>>[<mailto:scott@xxxxxxxxxxxxx>mailto:scott@xxxxxxxxxxxxx]
> >>>>Sent: Thursday, January 20, 2005 2:39 PM
> >>>>Cc: Si-List
> >>>>Subject: [SI-LIST] Re: risetime effects of plane breaks
> >>>>
> >>>>
> >>>>When this thread started I was on vacation. However, I found this
> >>>>interesting enough to resurrect some previous simulations I'd
> >>>>performed in CST Microwave Studio. After much playing, twiddling
> >>>>and generally having fun I can say several things:
> >>>>1) It's pretty easy to confirm Doug's results using 3D fullwave
> >>>>simulation. In fact, in about 30 minutes I can replicate his case
> >>>>and create a design that can be easily modified for many other
> >>>>possibilites. The microstrip split plane crossing is a no-brainer.
> >>>>Just don't do it and expect anything approaching an EMI "clean"
> >>>>system.
> >>>>
> >>>>2) Chris and Steve ... and eventually myself, wanted to know more
> >>>>about the various different stripline plane crossing configurations,
> >>>>so I setup a simulation with a VDD island not unlike what might be
> >>>>found in a memory system, and performed multiple simulations with
> >>>>dual asymmeteric stripline crossing the plane twice on it's way to
> >>>>the memory module. Not surprisingly the following is true:
> >>>>
> >>>> It is best not to cross a split plane ... even with stripline.
> >>>> If you do, it is better to cross a split that is adjacent to a
> >>>> ground plane
> >>>> It is even better if you cross a split adjacent to a ground plane on
> >>>> the stripline layer furthest away from the split plane (i.e. next to
> >>>> a ground plane)
> >>>> It is worst to cross a split plane that has no adjacent ground.
> >>>> The width of the gap in the plane makes very little difference until
> >>>> it becomes really small or really big.
> >>>> Crosstalk scales almost linearly with the number of aggressors
> >>>> crossing the split. (i.e. - it can get really bad!)
> >>>> Bypass of the split power island helps for frequencies below 500
> >>>> MHz, provides no help for frequencies higher than 500 MHz, and as
> >>>> such has no benefit to most of the noise and crosstalk created by
> >>>> high speed signals crossing onto and off of the island.
> >>>>
> >>>>The energy released into the power/ground plane cavities by high
> >>>>speed signal split plane crossings is huge and essentially cannot be
> >>>>suppressed with bypass capacitors. Any attempt at supprerssion with
> >>>>capacitors exhibits what I call a "Whack-A-Mole" property. You can
> >>>>never get rid of those pesky little moles. All you can do is to move
> >>>>them around by thumping them. Given that all this energy is rattling
> >>>>around the PCB power planes from split plane crossings, it will
> >>>>eventually go somewhere. Since it's really easy to develop all
> >>>>sorts of resonant power island cavities that have primary resonant
> >>>>frequencies in the 500 MHz to several GHz range, it is not at all
> >>>>unlikely that any split plane crossing has an extremely strong
> >>>>potential to excite a resonance in a frequency range that will cause
> >>>>most systems to fail EMC compliance testing About all you can do is
> >>>>to shield the cavity patches using ground layers. This should
> >>>>reduce the radiated energy significantly, but will not totally
> >>>>eliminate it, since eventually it will find it's way to all those
> >>>>pesky device and package leads.
> >>>>
> >>>>
> >>>>best regards,
> >>>>
> >>>>Scott
> >>>>
> >>>>
> >>>>
> >>>>
> >>>
> >>>
> >>>--
> >>>Ronald Miller
> >>>Ghz Data, Signal Integrity Consulting
> >>>7721 Sunset Ave.
> >>>Newark CA 94560
> >>>tel 510-793-4744
> >>>cell 510-377-9380
> >>>fax 510-742-6686
> >>><<a
href='http://www.ghzdata.com>www.ghzdata.com'>http://www.ghzdata.com>www.ghzdata.com</a>
> >>>
> >>>
> >--
> >Ronald B. Miller, Microwave/SI Engineer RAIL = NOISE + REFLECTIONS
> >Gigahertz Data Div of MI Corp. \\ // \\ //
> >7180 Thornton Avenue \\// \\//
> >Newark, CA 94560 ->JITTER<- EYE ->JITTER<-
> >tel 510-793-4744, //\\ //\\
> >fax 510-742-6686 // \\ // \\
> ><<a
href='http://www.ghzdata.com>www.ghzdata.com'>http://www.ghzdata.com>www.ghzdata.com</a>
RAIL =
> >NOISE + REFLECTIONS
> >
> >
> >
>
>
------------------------------------------------------------------
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:
