StanfordAHA/garnet

Tapeout: 45-degree RDL routes? Or manhattan?

Closed this issue · 1 comments

Our current flow routes redistribution layer (RDL/metal-10/AP) bumps to IO pads with 45-degree i.e. non-manhattan wires.

My first question: is it supposed to do this? It seems to happen in spite of the fact that the gen_power procedure (in gen_floorplan.tcl) appears to specifically request manhattan-style routing:

  set_db flip_chip_route_style manhattan 
  route_flip_chip -incremental -target connect_bump_to_pad -verbose \
      -route_engine global_detail -selected_bumps \
      -bottom_layer AP -top_layer AP -route_width 3.6 -double_bend_route

My follow-up questions have to do with what looks like false DRC violations arising from the 45-degree paths. I have filed these as separate issues
90-degree turns - #387
runlength spacing - #388

Talked to @mahorowitz . He recommended that we go with manhattan instead of 45-degree routing, which seems to have been the original intent anyway, and with any luck this will forestall the other issues that arise later (#387 #388). I like this path and will go forward with it until/unless someone convinces me otherwise...