On 07/07/2009 08:21 AM, Peter TB Brett wrote: > On Tue, 07 Jul 2009 07:16:04 -0500, Bill Gatliff<b...@billgatliff.com> > wrote: > >> Anthony Blake wrote: >> >>> I would prefer to implement this sort of functionality with topological >>> directives or constraints, and avoid geometric constraints if possible. >>> >>> >> Kind of like providing the circuit-board equivalent of a geological >> topographical/relief map, so the algorithms know where the "valleys" >> (preferred paths) and "peaks" (mountains, don't climb them unless >> necessary) are? And the lakes, airports, Area 51's, etc.... Except >> that with circuit boards it would be more than 3-D, because some of the >> mountains would be abstract things like excessive trace length (which >> might later trigger the addition of a via and then a retry), DRC rules, >> desire to avoid the extreme edges of the board, and so on... An [n]-D >> "relief map"... >> > Yes, you've just described geometric constraints in great detail. :P As I > understand it, topological constraints are something completely different! > > A little late to the discussion, but here's my $0.02: topological router is the only practical way of finding prospective routes between two points, but the geometric constraints are what really matter (provided a path is found!) to the performance of the circuit. So once a topological solution is provided, optimizing artificial potential or cost functions based on geometric and electromagnetic criteria would be the best way to route within a Voronoi cell, possibly with an edge from the corresponding Delaunay diagram as a starting candidate. (It looks like this may be the way it's done already?) If multiple paths are found, the 'best' path would be chosen by evaluating the cost functions.
Moreover - the cost multiples could be assigned to each branch of a net (or groups of branches), e.g. the distance from a USB connector to the receiver chip would be very high, the difference in the distance between a parallel interface is high, the area between two differential traces, etc. This would provide a *lot* of smarts to the autorouter, and maybe expanded to an auto-place routine, as well. Anthony - the router is looking fantastic! -Ethan _______________________________________________ geda-user mailing list geda-user@moria.seul.org http://www.seul.org/cgi-bin/mailman/listinfo/geda-user