StanfordAHA/garnet

Tapeout: False(?) DRC violations on 45-degree wires - 90-degree turns

Closed this issue · 1 comments

Special Wire of Net VDD
Type:	Non-sufficient Metal Overlap 
bbox = (4272.335, 2483.208) (4273.235, 2484.108)

45-degree routing happily uses 45-degree wires to round off manhattan turns, as I suppose it should. However, if the 45-degree wire is too short, the two manhattan wires overlap and cause a DRC error (see figures).

I believe this is a false error, please let me know if you disagree. Also, let me know if you have a good way of preventing these false errors in the first place e.g. I suppose it wouldn't be too hard to manually delete the 45-degree wire and extend the other two wires so it lookl like a normal manhattan turn.

My plan is to flag these as false violations for now, so we can keep the existing flow. Later maybe Caliber will tell whether this is a real violation or not.

This bend caused no violation:
bendgood

This bend caused trouble:
bendbad

I'm going to close this issue in favor of #389

For now we are doing manhattan routes and avoiding the problem.