Re:A new algorithm for graph crossings, hiding in plain sigh...
by not2b on Wednesday May 26, 2021 @05:59PM
(#27296631)
I don't work on placement problems, though part of my work provides input for the placement problem (produce an optimized logic netlist that then has to be placed and routed). But if I can suggest something in the same ballpark: imagine that you have a large graph and you want to partition it into a small number of partitions (say, 4 to 10) of similar size. You want each partition to be a planar graph, and you want the number of interconnects between these graphs to be minimal. The graphs would correspond to metal layers on an ASIC. There's a lot of fuzziness in the way I stated it; to make it more rigorous a suitable cost function would need to be defined.
[ Parent
| on HN ]
Re:A new algorithm for graph crossings, hiding in plain sigh...
by thomasahle on Thursday May 27, 2021 @12:14PM
(#27304753)
That sounds like a nice problem: Partition the vertices into planar sub-graphs, minimizing the number of edges between parts.
It's not really a dynamic problem anymore. It also becomes very specific to ASIC. Do you know if it has already been studied?
Re:A new algorithm for graph crossings, hiding in plain sigh...
by orange3xchicken on Thursday May 27, 2021 @02:39PM
(#27306515)
Someone correct me, but I think hMetis is one popular software/algorithm for multi-level graph/hypergraph partitioning. There is also KaHyPar which is a bit more academic.