At 07:43 PM 4/7/2004, Ian Wilson wrote:
On 08:51 AM 8/04/2004, Bagotronix Tech Support said:
But seriously, folks, I like the 99SE autorouter OK.  It's fast.
[...] As far as unrouting (or rerouting) locked tracks, the trick
is that your preroute must be 100% routed AND locked.  If it's only
partially routed, 99SE will unroute it.  If you go right up to 1 mil away
from the pad and then stop your track, that counts as a partial route to the
autorouter, and it will get ripped up.

It will also rip up any track that contains arcs (at least that was my experience when dumping a PCI card with meandered clock line at the router - after hand routing all the PCI signals).

All of this is related. It 99SE router is fairly dumb. That makes it fast. You want smart, expect to wait. Speed comparisons between the 99SE router and Situs are ... quite unfair. Situs could and should be better and faster, no doubt, but it follows rules far more completely than the 99SE router. To respect the rules takes processing power.

Perhaps it is possible (or could be made possible) for Situs to be set up to be dumber, like 99SE.... It would presumably run faster.

Arcs are not recognised by the 99SE router at all; the math is quite a bit more complex, i.e., time consuming. So it was not implemented. This applied to board outlines as well as to tracks. Since arcs aren't recognised, tracks containing arcs are treated the same as tracks with missing segments, i.e., as unrouted, even when locked. Rather than trying to deal with partial routes -- a whole other level of complexity -- the 99SE router simply insisted that they be complete, and "complete" meant a continuous connection from pad center to pad center. Once again, easy to program and fast to execute.

A quick fix for preroutes or board outline containing arcs was to reduce the arcs to line segments. If there were only a few, it could be done manually with just a few segments. If there were many, the layer could be photoplotted (using "software arcs," a bit of ambiguous language to the uninitiated) and all the draws reimported -- after deleting all the tracks and arcs on the layer. (Protel should have had a plot control for the number of segments per unit arc, the value used is far too high, the only argument for which is cosmetic.)

It will also rip up fills that are part of footprints - such as on SOT-89 devices. The Altium supplied library SOT-89 can be used to demonstrate this - though I would not suggest using this in a design as I think fill was missing the all important mask and paste fills. It is a neat way of making more routing channel space - just remove some component copper.

This kind of behavior, however, was a serious deficiency, really, simply incomplete programming. The program also should not have proceeded to rip up incomplete preroutes; rather, at worst, it should have notified the user that they existed, at the beginning of the route process, and should have refused to proceed if the route routines really could not handle locked copper primitives not a part of a completely routed net. "Locked" should mean "locked."

But Altium had been planning for a long time to replace the Neuroroute-cum-99SE router, so the router did not get a lot of attention.

One of the 99SE router's poor features was weak cleanup. When dealing with off-grid pads, the router would route close to the pad on grid, then, sometimes, put in a whole series of meandering track segments in order to finally reach pad center. It would then leave this garbage. Somehow it could not recognise that the shortest distance from point A to point B was a straight line, even if that line was not orthagonal or semiorthagonal. This should not have been difficult to fix. But it apparently had to wait until Situs.

The requirement for complete routing of a net for preroutes to be respected was really a serious limitation. You can see this if you take a routed design, like the Photohead pcb in the Examples, and delete a couple of line segments. A half-way smart router should be able to restore those segments in seconds. Instead, it rips up the whole net and then tries to find routes according to its passes. Not surprisingly, it often fails. It would not have been rocket science to have an optional reroute pass that used existing partial routes, run separately or before any other routing passes. This would have solved, almost completely, the incomplete preroute lock problem. I have not experimented with Situs.

My judgement about the 99SE router? A useful tool, once one realized its limitations. A tool does not need to be able to meet all needs in order to be useful. In some situations, the 99SE router is more useful than Situs, it appears, though it is possible that Situs might be controlled through setup to make it as useful or more useful than the 99SE router. But in general, from the reports, it seems that Situs is a better router, already, and there is reason to expect that it will continue to improve. The topological routing theory is good (great article in the manual), now, let's see what Altium can do with it!

The P99SE does not clean up well. The DXP/P2004 Situs router has some clean-up passes that work pretty well I think. It is a shame that the routing passes seem difficult to get working and it is so slow (at least when it is getting close to completion)- at least that is my experience, which is not a great deal as I tend to drop autorouters fairly quickly when they don't work for me. Maybe I expect too much or my designs are not good autorouter candidates or I am not good at placement for autorouting.

One of the big complaints about autorouters in general is that, in attempting to run to completion, they fill all available channels. If there are only a few nets left, it is usually not difficult to finish the routing by hand, still, per net, it is often much more laborious than simply routing by hand would have been. So, before a certain level of completion, one would be better off manually routing the whole thing. I've been predicting for years that routers will keep getting better, until they reach the point that they are better than the routing of even a skilled manual designer. But I don't think they are there yet.

At this point, speed is quite desireable. With a fast router, one can test placements quickly. 99SE was pretty good for this. As should be underscored repeatedly, in routing, placement is king. If you've got plenty of room and slow risetimes, you can route even a poor placement. But as the board becomes more dense and signal speeds rise, placement becomes everything. Rat's nest displays, especially those that recalculate when the parts are moved, are very, very useful for seeing how a placement looks before trying to route it. A manhattan distance report can be useful for examining preroutes, though such a report can be misleading. Consider the routability of nets that (in rat's nest) cross vs. those that do not. A manhattan distance report does not consider net crossing, yet crossed nets are clearly more complex to route.

I'm going to have to spend some time with Situs....

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * To post a message: mailto:[EMAIL PROTECTED] * * To leave this list visit: * * * Contact the list manager: * mailto:[EMAIL PROTECTED] * * Forum Guidelines Rules: * * * Browse or Search previous postings: *[EMAIL PROTECTED] * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

Reply via email to