RE: [SI-LIST] : PCI bus impedance

About this list Date view Thread view Subject view Author view

From: Willis, Ken (
Date: Thu May 17 2001 - 04:43:12 PDT

There are a couple of other factors that tend to drive you towards 50 ohms
as well. If you use fine pitch (1mm) BGAs, you end up needing 4 or 5 mil
lines to route between the pin escape vias. Then since you need to use
a fair number of signal layers to route into these monsters, you end up
using pretty thin dielectrics so the board doesn't get too thick. If the
gets too thick you need to use bigger vias to keep the aspect ratio down,
which ends up costing too much in real estate. A vicious cycle.
The other factor is that if you have any LVDS or CML interfaces, they tend
to have internal 100 ohm differential terminations. To minimize impedance
mismatch, that also tends to drive you towards 50 ohm single-ended

-----Original Message-----
From: Ron Miller []
Sent: Wednesday, May 16, 2001 1:41 PM
To: ''; Wang Xiao-yun
Cc: Patel, Bhavesh;
Subject: RE: [SI-LIST] : PCI bus impedance


There is an overriding need to keep at the old standard of 50 ohms.

Virtually all test equipment is 50 ohms and coaxal cables and adapters
are predominantly 50 ohms.

As the drive for miniturization and denser connectivity continues, to
move in 1/2 or 1/4 increments of the 50 ohm makes a lot of sense since
transformers and coaxes are already available. Also, we need to stick with
a tightly restricted number of impedances for interoperability, so 12.5 ohms

(or 15 ohms) and 25 ohms makes a lot of sense.

Ron Miller

-----Original Message-----
From: [
<> ]
Sent: Wednesday, May 16, 2001 12:43 AM
To: Wang Xiao-yun
Cc: Patel, Bhavesh;
Subject: Re: [SI-LIST] : PCI bus impedance

Hello Xiao !

Principally you are right; impedance "bumps" occur between 10-20 ohm at the
and 50 ohm on the transmission line.
But that is only really true for very short pulses (ex: TDR / 30 psec
risetime ).
Longer risetimes cause waves to "ignore" such a "bumpy road"; or lets say
become diffused as risetimes (resp. wavelenghts )
become longer.

A comparision to water waves may be very imaginative: assume a wave
along a river. Short wide gaps of river width will not affect the
propagation of
the wave, but longer ones (at least 1/20 of the wavelength) will have an
Same you can do with plane wave effects (power planes / INoise); when
stones in a pond with a certain frequency, you will get a wave interference
pattern, which may be similiar to what happens between Vcc-Gnd-planes when
by current injections.

At frequencies we regard, those capacitive bumps can be regarded as integral
of the transmisssion line ( not valid for higher frequencies, where the
length of
the capacitive impact is in the range of a wavelength; here you have to
3D-field solvers to get the real effects, which also include inductive and
resistive stray elements).

Looking upon a via as an integral capacitive part of a transmission line is
a rule
of thumb for engineers to make live easier and get faster results.
Not everybody can afford 2D- or 3D-field solvers @ 20.000 $ and up for 1
Just look upon this as a "cooking receipt" to get an instant result instead
doing a biochemical analysis of the whole meal.

By the way: it will not matter, whether you choose a 40 ohm transmission
line or
65 ohm transmission line; both will work as the "impedance"
of the line goes down to 25 / 12.5 ohm with the effects of the drivers
50 ohm is chosen by many people as this has been done since the last 50
years and
only a few put this into question.
Long ago, scientists found, that at 33 ohm the energy throughput for a
transmission line is a maximum, but attenuation is best at 75 ohm.
As the transmitted energy and received energy had to be guided on the same
line, a
good compromise was to choose 50 ohm as the "middle".
So thats an old rule, which may need a new interpretation for todays needs,
as it
came into existence when copper pipes were used as transmission lines
and design tools were slegde hammers and pipe wrenches.

Look at Futurebus physical design: terminations are defined 33 ohm at both
which will be the effective impedance of a transmission line with vias and
cap. icluded with this slotpitch and connector type.
From my point of view I would choose 40 ohm, as manufacturing tolerances are

decreasing their effects at this level. Production yield will go up,
thickness of
pcb will go down, PCB will become cheaper.

Hope, this explains a little bit the background of bussed lines.



Wang Xiao-yun wrote:

> Hello, George:
> I'm a bit confused with your explanation and here is my question.
> When a via is put on a transmission line of nominal 50 ohm, the
> at that point and only that point is changed and therefore the signal will
> an impedance mismatch at exactly the place where the via is. The rest of
> segments of the line still has 50 ohm impedance.
> I agree the manufacturing tolerance of the impedance will range around
> +/- 10% at least, but it shouldn't be the reason to increase the impedance

> because it does no help. The only thing I can do is to route the tracks
> 50 ohm target and simulate them with manufacturing tolerance in mind.
> I also agree with you that lower impedance will induced less crosstalk,
> as a result I will be much more happy with a design with unique impedance
> 50 ohm wherever possible. If I were asked to have some traces of 65 ohm, I

> would like to find out wether it's really necessary. Sometimes you may
> have to do some extra work to accomplish it, for example, change your
> stackup a bit.
> Comments are appreciated.

**** To unsubscribe from si-list or si-list-digest: send e-mail to In the BODY of message put: UNSUBSCRIBE
si-list or UNSUBSCRIBE si-list-digest, for more help, put HELP.
si-list archives are accessible at

**** To unsubscribe from si-list or si-list-digest: send e-mail to In the BODY of message put: UNSUBSCRIBE
si-list or UNSUBSCRIBE si-list-digest, for more help, put HELP.
si-list archives are accessible at

About this list Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Thu Jun 21 2001 - 10:12:00 PDT