1) Inter-plane capacitance is the first source of ac return current.
2) "The shorter the rise time, the closer the via or de-cap." This suggests to
me that the return current distribution vs. frequency is really a continuum from
using the whole plane at dc to using a small swath under the trace at microwave
frequencies. If this adage is true, then there still must be a significant
amount of return current at one rise time away from a signal via at most of the
frequencies we're interested in as signal integrity engineers.
3) 2D field solvers don't assume anything about return current. I guess if I
had thought long enough, this should have been clear. A 2D field solver
essentially solves an electrostatic problem for capacitance, often assuming an
infinite plane, and then computes inductance using the telegraphers equations.
The magnetostatic problem is never solved. Nevertheless, they still provide
accurate solutions from a return current perspective.
Greg Edlund
Advisory Engineer, Critical Net Analysis
IBM
3650 Hwy. 52 N, Dept. HDC
Rochester, MN 55901
gedlund@us.ibm.com
---------------------- Forwarded by Gregory R Edlund/Rochester/IBM on 10/04/99
07:53 AM ---------------------------
"Ingraham, Andrew" <Andrew.Ingraham@compaq.com> on 09/30/99 02:02:40 PM
To: "'gedlund@us.ibm.com '"
<"IMCEAMAILTO-gedlund+40us+2Eibm+2Ecom"@compaq.com>
cc:
Subject: RE: [SI-LIST] : the old high-frequency return current model
Greg,
Maybe the answer is that, even though the current distribution looks a lot
different when the nearest decoupling cap is more than a few trace widths
away, its effect on the impedance is just not that much to worry about?
Kind of like the argument about using chamfered corners on trace bends.
Yes, in theory you get an impedance discontinuity if you don't, but I think
Ed Sayre says you'll never see it unless you operate well above 1GHz.
If the cap is half an inch away from the via, that's around 80ps away. So
unless the risetimes are of that order of magnitude or faster, then the
discontinuity might be insignificant.
But then at those kinds of speeds, our discrete capacitors are pretty much
ineffective anyway. By that point we are relying on the intrinsic
capacitance between layers. The higher you go in frequency, the better that
intrinsic capacitance looks (and works).
Regards,
Andy Ingraham
-----Original Message-----
From: gedlund@us.ibm.com [mailto:gedlund@us.ibm.com]
Sent: Thursday, 30 September, 1999 14:12
To: si-list@silab.eng.sun.com
Subject: [SI-LIST] : the old high-frequency return current model
Shoot! I was out of town and missed one of the most interesting discussions
of
the year! (Plane-jumping return currents) So at the risk of re-opening this
thread, filling all your mailboxes again, and being branded an outcast, here
goes. (Remember, that delete button is only a few inches away...)
You're all familiar with this picture of high-frequency return current
bunching
up under the signal trace, right? According to the picture, it dies off
pretty
quickly as you move along the x-axis away from the trace. Well, I've been
considering rules for the area density of ground vias and decoupling
capacitors,
and it occurs to me that if this picture were true, then the only place for
a
ground via or capacitor is within 2-3 trace widths of the signal via in
question. (Which is, for most of our applications, absurd.) Otherwise I'd
be
forcing the return current out of that very tight loop, increasing the
inductance, adding a discontinuity, generating plane noise, emissions, and
all
those nasty things. Now, I know that boards work quite well up to a few
hundred
MHz with considerably less than 100 de-caps per square inch! So where's the
discrepancy? Is there a hole in my fairly simplistic, qualitative analysis?
Or
is this just like everything else: knowing how some parameter varies
between
the end cases is much harder than analyzing the end cases?
On another tangent, I believe 2-D field solvers make the assumption that the
return current is evenly distributed across the surface of a plane when you
ask
them to compute C, L and Z for a given cross-section. Doesn't this also
conflict with the high-frequency current distribution picture?
Eagerly awaiting your answers and hoping I have time to read them,
Greg Edlund
Advisory Engineer, Critical Net Analysis
IBM
3650 Hwy. 52 N, Dept. HDC
Rochester, MN 55901
gedlund@us.ibm.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 ****
**** 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 ****