OS: Linux/64bit Mint 18.3
KiCAD: the current download provided by synaptic: 4.0.2+dfsg1-4
Autorouter: Any and all…whatever it takes (for this high-tracecount, but large area board)… but initially I tried the built-in autorouter.
Ahhhh… no, not at the beginning, yes for final touch-ups. I’d be here for a year…
Of course, that is currently unavailable… unless someone can point me to where it might be downloaded
Can you please specify their names, and where they can be downloaded from? Although I think I know how to run it, it would be nice to see some docs.
BTW, my board is 54 square inches (the size is needed for large connectors, not due to high IC density), with tons of space … only about 2 square inches are covered by several QFN64 ICs, so its not like this is an impossible task to route.
I was provided with “pcbroute” (one fork of freerouting.net) which “seems” to “sort-of” work, but it can’t really be, as even when pcbnew is configured with 32 layers, “WHEN FIRST STARTED” it fails on the first 3 traces… come on, its got a blank canvas of 31 completely free layers, how can it possibly fail? The first 31 traces have their own blank layer !!! So, something is definitely wrong.
After the first pass (8-12 hours) it routes about 1/2 of the traces, then after several more passes (taking a few days), it gets stuck at just completing 2/3 of the traces, leaving about 700 uncompleted.
OTH, I actually think the built-in router for KiCAD works better, but from what I think is happening, it only routes 2 layers at a time (or one layer, if for example you pick top layer & top layer as the pair of layers to route). By working better, it seems to me that KiCAD does almost as well with 2 layers, as “pcbroute” does with 32 layers!
Transitioning KiCADs routing to pcbroute actually allows pcbroute to progress farther than using pcbroute from scratch, but still comes nowhere close to finishing, even with 32 layers.
Hence the topic of this thread: Do I have to pick every combination of layers and re-run the routing? Example for 4 layers: 1&1, 1&2, 1&3, 1&4, 2&2, 2&3, 2&4, 3&3, 3&4, and finally 4&4 ? You can see how this is possible but impractical with 4, 6, 8 or more layers. Further, if you can pick only 2 layers, then a 3+ layer trace is never going to be possible!
That is still inferior to a different algorithm: trying to rout on initial layer first, then picking the best option after first trying all layers to complete…
----------------
Regarding new algorithms, I came across the old autorouter code from Mike at PikeAero.com (et al)
https://sourceforge.net/p/qautorouter/wiki/Home/ 
and it nicely and graphically loads up the Spectra .dsn export from newpcb
But, thats all it does, as that is as far as those developers had time for. Yet, it gives an AMAZING head start to anyone who wants to write a fast router algorithm… freerouting.net and clones being speed-crippled by having been written in Java (not C or Julia).
Therefore, I raise the alert to those (especially in the University crowd) who might want to experiment with routing algorithms, to start from this point… BTW with blessings from Mike !
