Re: [SI-LIST] : the old high-frequency return current model

Tadashi ARAI (arap@jdc.paso.fujitsu.co.jp)
Wed, 06 Oct 1999 18:53:23 +0900

Someone, please tell me.

How can I open an attached file, WINMAIL.DAT.

On Mon, 4 Oct 1999 11:40:21 -0500
Mail from eric@bogent.com described as below:

> This is a multi-part message in MIME format.
>
> ----------------------------------------------------------------
> Brad, and others-
>
> The recent discussion on the impact on differential pairs in crossing a
> split in a plane has gotten me thinking about how we could do a quick
> simulation, in the absence of doing a 3D FDTD simulation of the actual
> currents and voltages. This was a very good suggestion, and I also await to
> see the film.
>
> Here's one approach that I tried and would like folk's opinions on. It has
> certainly opened my eyes a bit at how robust differential pairs are.
>
> Consider a coupled, microstrip differential pair that is 7 inches long. At
> the three inch point, the plane on the back side is etched off for a length
> of 1 inch. This is an extreme case of a differential pair passing over a
> split in the ground plane.
>
> For the special case of lines 5 mils wide, 5 mil spaced, with a 2.9 mil
> thick dielectric to the return plane, the matrix elements of the two lines
> are: Z11 = 52.4 Ohms and Z12 = 5.2 Ohms. I got these from my trusty Ansoft
> Maxwell 2D Extractor, 2D field solver. This gives a diff Z0 of 2 x
> (52.4 -5.2) = 94.6 Ohms.
>
> In the region where the plane has been removed, the single sided Z0 is 160
> Ohms. It is interesting to note that the field pattern of the single ended
> coplanar lines is the same as the differentially driven coupled microstrip.
> The difference, of course, is that the presence of the return plane
> dramatically changes the specific value of the characteristic impedance of
> the differential pair vs the single ended. From
> the differential signal's perspective, it will see an impedance
> discontinuity of 95 Ohms in the microstrip area and then 160 Ohms for the 1
> inch distance, and back to 95 Ohms. The common mode will see an impedance of
> 57 Ohms in the microstrip region and infinite in the coplanar region. I
> think this is how the gap will "block"
> common mode signals, as D.C. was alluding to. Basically, the common mode
> impedance is open in the gap region, but the differential mode impedance is
> only about 80% larger in the gap.
>
> To first order, this will be the main effect the signal sees in crossing the
> gap. We all agree that there will be mixing of the return paths in the plane
> within a few line widths of the traces, at the edge of the gap. This is
> probably on the order of 20 mils. I think this discontinuity will be small,
> compared to the 1,000 mil discontinuity of the 95 Ohms to 160 Ohms
> transition. If this is the case, we can simulate the common and differential
> mode signals with a tool like Hyperlynx to look at the magnitude of the
> signal degradation. A number of folks pointed out that yes, there will be a
> discontinuity, but if you keep the lengths short compared to a rise time, it
> might not affect the noise margin much.
>
> In order to use Hyperlynx to simulate this structure, we have to trick it
> into seeing the coplanar region as a coupled microstrip region, as it
> assumes perfect planes as return paths for all signal lines. We do this by
> using a 2.9
> thick dielectric for the differential sections and using a 50 mil thick
> dielectric for the coplanar region. Using my trusty Ansoft 2D field solver,
> I found that for two coupled microstrips, as long as the dielectric to the
> return plane is thicker than 15 mils, the differential impedance saturates
> at 140 Ohms. The common mode impedance continues to rise, however, as the
> plane is lowered.
>
> This is a bit lower than the 160 Ohms of the single ended line with no
> return plane, and the dielectric 2.9 mils thick, since in the case of just
> thicker dielectric over the plane, there is field in the thick dielectric
> and higher capacitance, and hence, lower impedance. The difference is small,
> and I went ahead and used the two coupled microstrip, with 50 mil thick
> dielectric, as my gap region. The common mode impedance was 320 Ohms- large
> compared to the 57 Ohms common mode impedance of the front of the line.
>
> The Hyperlynx circuit I set up has three diff pairs in series. The first
> section is 3 inches long, 95 Ohm. The next section is 1 inch long, 140 Ohms
> and the third is 95 Ohm, 3 inches long. Each trace is defined by its cross
> section, taking advantage of the built in field solver to calculate the
> matrix elements. I have independently verified the Hyperlynx field solver to
> be within 1-2% of the Ansoft tool. I used a differential TDR as the source,
> with a 100 psec rise time. (The 1 inch gap is about 130 psec long) The end
> of the lines are differentially terminated with 95 ohms.
>
> Slide 1, in the attached file, is the near end response of one channel
> of the TDR and its far end response, TDT, and the differential signal at the
> far end. I compare the response of the differential pair with a 1 inch gap
> and a 0.1 inch gap. You see the impedance discontinuity of the gap in the
> TDR, as you expect, a little distortion in the TDT, and very little effect
> in the differential response. The impact from a 0.1 inch gap, maybe more
> realistic even with 100 psec rise times, is almost non existent. This
> suggests that the differential signals are not impacted much by the gap.
>
> However, as many have pointed out, the real problem with differential lines
> is when there is an asymmetry, as with driver skew. In Slide 2, I compare
> the same traces, with a delay of 50 psec in the second driver, but recorded
> at the far end. The two unmarked scope traces are the + and - signal lines.
> The green is the differential signal. The skew is 1/2 a
> rise time, and might be a common magnitude of delay. Again the differential
> signal is not degraded very much- at most, its rise time is slowed down a
> bit, but the signal quality is good. The big impact is the generation of
> common mode voltages (and currents) on the other side of the gap. I can't
> plot common mode voltage in Hyperlynx, but you can see it clearly in the
> ripples that move in step at the receiver side. Note that without a gap, you
> still get some common mode generated from the driver skew, its just cleaner.
>
> I looked at the near end waveforms, at the source, with and without a skew,
> and these show the generation of large common mode noise with the skew, and
> low values without the skew. I would have appended this figure, but this
> list allows a max posting size of 60k, and I couldn't fit it in this note.
>
> Conclusion: The gap will be an impedance discontinuity to the differential
> signals, and the common mode signals. The differential signals in passing
> over the gap are relatively clean, as they see a short, series,
> discontinuity. Unavoidable asymmetries in the lines and
> drivers will cause larger common mode voltages, since the common mode
> impedance discontinuity is larger. The reflected signals will scale
> with the TD of the gap. If the skew is ever comparable to the rise time,
> series
> termination might be needed to dampen the reflected common mode signals. The
> quickest way to get an intuitive and quantitative feel for the behavior of
> the differential and common mode signals is to look at the size and length
> of the differential and common mode impedance discontinuity a gap or other
> uncontrolled effect presents.
>
> The recent discussion on this topic has been very timely for me, as I am
> working on a paper for DesignCon with Mike Resso of HP and Steve Bird of
> Hadco, on differential impedance analysis with TDR. Based on the last few
> weeks' discussions, I am adding a test structure to our board that has a
> split plane.
>
> Comments are always welcome.
>
> --eric
>
>
>
>
>
> Eric Bogatin
> BOGATIN ENTERPRISES
> Training for Signal Integrity and Interconnect Design
> 26235 W. 110th Terr.
> Olathe, KS 66061
> v: 913-393-1305
> f: 913-393-1306
> pager: 888-775-1138
> e: eric@bogent.com
> web: www.bogatinenterprises.com
>
>
>
>
> **** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****
>
>

//// /// // / / / / / / / / / /
Tadashi Arai//Platform Developing Dept.,Desktop Prd Div. Fujitsu Limited
arap@paso.fujitsu.co.jp TEL:+81-42-370-7617 Inagi-shi, Tokyo, Japan
/ / / / / / / / / / / // ///

**** To unsubscribe from si-list: send e-mail to majordomo@silab.eng.sun.com. In the BODY of message put: UNSUBSCRIBE si-list, for more help, put HELP. si-list archives are accessible at http://www.qsl.net/wb6tpu/si-list ****