Mail Archives: geda-user/2015/07/27/06:49:23
On Mon, Jul 27, 2015 at 10:35:24AM +0200, gedau AT igor2 DOT repo DOT hu wrote:
>
>
> On Mon, 27 Jul 2015, Gabriel Paubert (paubert AT iram DOT es) [via geda-user AT delorie DOT com] wrote:
>
> >On Sun, Jul 26, 2015 at 04:30:04AM +0200, gedau AT igor2 DOT repo DOT hu wrote:
> >>
> >>
> >>On Sat, 25 Jul 2015, DJ Delorie wrote:
> >>
> >>>If that includes specifying coordinates, then sweet :-)
> >>>
> >>>>I've just installed 20140306 (was that the last snapshot?) to test it.
> >>>>With the gtk hid, 'R' doesn't do anything. Report doesn't print length
> >>>>info either.
> >>>
> >>>Might be lesstif-only. :Report(NetLength) for gtk.
> >>
> >>Cool, thanks! Then I modify this point to add only the differences
> >>in some details I had in mind:
> >>
> >>- handle junctions (I already have most of graph buliding/handling
> >>the code because of mincut)
> >>
> >>- because of that, more integration with the GUI: lengths printed on
> >>the traces between junctions
> >>
> >>- I am not sure yet to what extent, but try to handle some of the
> >>corner cases: don't count overlapping traces twice, take
> >>poly/via/pin as a junction
> >
> >As soon as you have a polygon in a net, defining its length becomes
> >dubious, to put it mildly.
>
> Yup. High freq will follow paths that I won't be able to predict. My
> plan is to say the whole poly is a big junction and I calculate
> trace lengths from junction to junction.
>
> Practical example: you have a trace between your MCU and your two
> sdram chips and the return path is on a ground plane. You can see
> the trace lengths going from the MCU to the ram chips, seeing all
> arms of the T junction. Since return path is a ground plane, you
> can't use this feature to tell anything about the "trace length" of
> the return signal.
>
> I think it's not common to mix traces and polygons along a high
> speed trace where you want to tune the length - but I'm real novice
> in high speed design.
Correct, there are essentially two methods:
- single ended microstrip/stripline in which the length is defined
by the net segments of the signal and return current goes through the
ground plane(s) more or less following the path defined by the signal
(except that actual current paths will be significantly wider, resulting
in lower effective resistance/inductance). Grounded coplanar waveguide,
very popular with Hittite's (now Analog's) evaluation boards, is
similar. I've never used nor seen pure coplanar used so I won't
comment on it.
- differential signaling, in which two parallel lines carry current in
opposite direction (a least the AC part, differential ECL signals
sometimes have a bias current on the signal tracks, depending on where
the loads are located). Both lines ideally have to have the same
length, this is not always possible because of turns and/or location of
the pads (sometimes the two pins of a differential pair on a FPGA in a
BGA package are not neighbours).
In both cases:
- these are point to point signals and stubs are essentially prohibited,
- line length is easy to define, it should be the distance from source
to load, ignoring the (always short) stubs, and for differential signal,
the length difference between both nets should be minimized.
Regards,
Gabriel
- Raw text -