[SI-LIST] Matching within 1 mil is just plain sillyness, IMO
- From: Scott McMorrow <scott@xxxxxxxxxxxxx>
- To: wdowsley@xxxxxxxxx
- Date: Fri, 01 Jun 2007 13:14:18 -0400
Several points I'd like to make. Even though the software says that
there is a 1 mil delta, that does not mean there is really a 1 mil
delta. In every PCB layout tool that I am aware of, measurements are
made from the origin point of one pad to the origin point of the other
pad, through the centerline of the trace. If all parts are centered on
the pads, this can be reasonably accurate. However, many surface mount
devices with rectangular pads will contact the pads at various location
off-origin. This results in significant error. An example of this
might be TSOP memory devices.
Traces are measured from the center of the track. However, the true
path delay can be shorter (or longer) going around a bend, depending on
whether the bend is predominantly inductive or capacitive. If the
number of bends are not equal, and the direction of the bends between
two traces are not equal, their delays will not be equal and
significantly larger than 1 mil ( 170 fs in FR4).
Another example might be traces crossing over or near antipads, where
delay is altered.
Another example is with serpentine traces. Crosscoupling between trace
loopbacks can introduce significant delta in end-to-end delay.
Ultimate, look at the numbers. 1 mil of stripline trace in FR4 has a
delay of 170 femtoseconds. Does it make sense in any design to match
something to accuracy that is beyond the propagation velocity uniformity
of the underlying materials? Ultimately your FR4 has a variation of
somewhere around 3 to 10 ps/inch. For my money, if traces are matched
to within 50% of the underlying material uncertainty, I feel pretty
good. That amounts to about a 50 mil trace matching requirement at the
smallest (8.5 ps). If you're worried about EMI, don't be. Even if this
was the maximum skew in a differential pair the 8.5 ps common mode
pulse width would have maximum energy around 85 GHz. I'm pretty sure
that most dielectrics will absorb that.
Go ask your silicon designers just how well matched the outputs of their
drivers are, and what the skew in the internal clock trees are. Ask the
IC package designers how closely they matched the trace delays in their
package, and how accurately they report those numbers back to you. And
ask your material vendors what the lower bound of material Er variation is.
Then justify the pain and suffering that a layout designer has to go
through to match to within 1 mil or even 10 mils of trace length. I'm
quite tired of lazy engineers not running the numbers and placing the
burden on someone else. In most cases, if your design cannot withstand
even 100 mils of trace length mismatch (17 ps of delay mismatch), you
have a real serious problem with your design margins elsewhere, which
you are not going to eliminate with ultra silly trace length matching.
If you were a mechanical designer, designing components with a material
that had the dimensional stability of rubber, and ask the machinists to
mil it to within 1 mil tolerance, you'd be laughed out of a shop.
respectfully,
Scott
Scott McMorrow
Teraspeed Consulting Group LLC
121 North River Drive
Narragansett, RI 02882
(401) 284-1827 Business
(401) 284-1840 Fax
http://www.teraspeed.com
Teraspeed® is the registered service mark of
Teraspeed Consulting Group LLC
Bill Owsley wrote:
> Jeff,
> You said it much better than I was inclined to.
> Thanks.
>
> As for leading edge designs using an auto-router, the guys spend nearly as
> much time working the constraint table as it would take to lay it out by hand
> and then it can be done over and over. But they still get those annoying
> leftover nets that auto-router chokes on, reducing them to a miniumum is
> their objective.
>
> "Loyer, Jeff" <jeff.loyer@xxxxxxxxx> wrote:
> I think folks are missing Bill's point. I don't think he's saying that
> the design is going to break at 2 mils routing difference, only that he
> can easily attain a maximum 1 mil delta, so he uses that as his
> constraint. If his CAD folks balked at that number, he'd probably
> adjust accordingly and, if it got close to having significant impact,
> would do a more detailed analysis.
>
> This is very similar to what we've done for common-clock FSB signals:
> tell our CAD folks to match them within 10 mils (or some number), even
> though we know we can absorb more. But, 10 mils is easily attainable,
> and guarantees essentially no skew due to length differences. Granted,
> it's a little more work on CAD's part, but the tradeoff is I don't have
> to do a grand analysis to figure out exactly how close they HAVE to be
> matched, and then worry about whether my analysis was correct (and
> require subsequent validation in the lab, etc.).=20
>
> This is about compromises, and where you put your energy into in-depth
> analysis. If you can easily do 1 mil spacing, no problem. It's a
> little tighter than I'd generally specify, but it's not absurd, either.
> If you're using an auto-router (that's surprising to me, I didn't think
> leading-edge designs could use them), you probably have different
> constraints.
>
> Jeff Loyer
> ------------------------------------------------------------------
> 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 technical documents are available at:
> http://www.si-list.net
>
> 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
>
>
>
>
>
> ---------------------------------
> Need a vacation? Get great deals to amazing places on Yahoo! Travel.
>
> ------------------------------------------------------------------
> 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 technical documents are available at:
> http://www.si-list.net
>
> 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
>
>
>
------------------------------------------------------------------
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 technical documents are available at:
http://www.si-list.net
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:
- » [SI-LIST] Matching within 1 mil is just plain sillyness, IMO
