Hi Jack -

1) We don't have a network analyser, so I cannot quantify a difference if there is one. In most of our builds the controlled impedance lines end up on prepreg. We have not had any problem with that SI-wise as measured with scope or judging by packet loss. I am not too hung up in the accuracy of controlled impedance lines, I think other things are more likely to cause reflection (also not quantified)

2) Quarter wavelenght stitching is good for avoiding resonance and waves propagated between layers, but do you usually know what your highest frequency to be concernad about is? I don't. There usually is not space enough to maintain a grid like this everywhere, so in reality I want a GND via everywhere one can fit. But this does not apply directly to what we are discussing here. It is more important to me that there are GND vias close to all other vias. The closer the better, and vias are cheap. If every signal has a good close return, the likelyhood of igniting any resonant structures is much reduced.

MEd vennlig hilsen/Best regards
________________________________________________
Haldor Husby, Senior Development Engineer
Data Respons Norge AS
Kongsberg Næringspark
P.O. Box 1022
N-3601 Kongsberg, Norway

Tel: +47 32 29 94 00 	Fax: +47 32 29 94 40
Dir: +47 32 29 94 18 	Mob: +47 48 04 83 68
[log in to unmask] 



-----Original Message-----
From: Jack Olson [mailto:[log in to unmask]]
Sent: Fri 1/5/2007 3:16 AM
To: Haldor Husby
Cc: [log in to unmask]
Subject: Re: Straddling the Return Path
 
very interesting....

Just out of curiosity, you inspired two related questions:
1) Do you ever find that in your 3-layer configurations, one
side is tightly controlled dielectric because it is a core, but
the other side being pre-preg fluctuates? Have you ever had
a problem in that regard?

2) By saying you stitch as tightly as possible, are you
going above and beyond a quarter wavelength of your highest
frequency? Do you even calculate that, or do you just shotgun
every available space with GND vias?

Jack
(aka "the blasted on SI-LIST for mentioning stitching" guy)
.



On 1/4/07, Haldor Husby <[log in to unmask]> wrote:
>
> I think you had just the right amount of coffee today. Your observation is
> good, if the return planes for vertical and horizontal routing are not the
> same net, you essentially loose control of the return path every time you
> make transition between the routing layers. Since transitions often happen
> close to IC's ther is usually a lot of decoupling around, but somtimes
> there
> are clusters of transitions away from any circuitry, and much crosstalk
> can
> happen there.
>
> For several years now I have used board stackups consisting of closely
> spaced substacks of 3 layers with the return plane in the centre as far as
> possible. The substacks are then separated with layers that are more than
> 2x
> the distance between layer internal to the substack. This approach
> resembles
> what you are suggesting, and it has worked very well for me in terms of
> noise control and signal integrity. Applications span wireless networks,
> FM
> radio, sensitive analog measurement circuits and GBethernet over copper.
> (also switching power supplies creating large noise currents at 100 MHz
> close a couple of cm away from a FM receiver with no loss of sensitivity).
>
> I also use a single GND if at all possible, and only use GND as return.
> The
> GND layers are stitched as tightly as space will allow, and I always use
> GND
> stitching around the edge of the board.
>
> Several of our customers have remarked that boards we have re-developed
> for
> them are remarkably more quiet than earlier versions and that it has had a
> real impact on perfromance.
>
> I know everyone cannot afford such extravagant use of copper, but I warmly
> recomend the approach you suggest, and it can be realized in other ways
> than
> the one I have suggested here.
>
>
> Med vennlig hilsen/Best regards
> ________________________________________________
> Haldor Husby, Senior Development Engineer
> Data Respons Norge AS
> Kongsberg Næringspark
> P.O. Box 1022
> N-3601 Kongsberg, Norway
>
> Tel: +47 32 29 94 00    Fax: +47 32 29 94 40
> Dir: +47 32 29 94 18    Mob: +47 48 04 83 68
> [log in to unmask]
>
>
>


---------------------------------------------------------------------------------
DesignerCouncil Mail List provided as a free service by IPC using LISTSERV 1.8d
To unsubscribe, send a message to [log in to unmask] with following text in
the BODY (NOT the subject field): SIGNOFF DesignerCouncil.
To temporarily stop/(restart) delivery of DesignerCouncil send: SET DesignerCouncil NOMAIL/(MAIL)
Search previous postings at: www.ipc.org > On-Line Resources & Databases > E-mail Archives
Please visit IPC web site http://www.ipc.org/contentpage.asp?Pageid=4.3.16 for additional information, or contact Keach Sasamori at [log in to unmask] or 847-615-7100 ext.2815
---------------------------------------------------------------------------------