@Tim Edwards General question about LEF generation by `magic`: What is the strategy for the OBS generation? See also https://github.com/The-OpenROAD-Project/OpenLane/issues/1522

For full macros, it is generally expected that you will use

lef write -hide

from magic, which will then put OBS layers covering the entire macro except for cut-outs for the pins, for every layer that has some geometry inside the macro. If you use

lef write

without options, then OBS structures will be placed for any metal in the macro that is not part of a pin.

Sounds reasonable. So macro hardening in OL should then implement the -hide option, this does not seem to be the case now, but I will run some tests so see what is going on.

@Harald Pretl: LEF is a library format, so after reading a LEF file with

lef read

you would need to

load

a cell to view it.

Hi @Tim Edwards I did some testing: The

lef write -hide

in magic is only writing the OBS in a layer, if tracks are there, only PINs are not sufficient. However, it looks like that

pdngen

needs the OBS to function correctly. Question: Does it makes sense to change the

magic

behavior to write OBS in a layer if PIN or a track is there in a given layer?

By "track", you mean any wire that is not part of a pin? That doesn't sound quite right, although it's possible that I just haven't seen that specific situation. I would have expected obstruction on any layer that needs it to keep the router from colliding with something internal to the macro, whether that's something that is connected to a pin or not. Probably I need to see an example of the layout and the resulting LEF file.

See here a DEF that has PDN stripes in

met4

and one connection. When writing a LEF with

lef write -hide

, there is only PINS plus the OBS for the one wire on

met4

, but nothing else in

met4

OBS

. Is this the intended behaviour, or should there be a large OBS stucture in

met4

covering the whole thing?

It's intended behavior, but seems like a good case for implementing some behavior around using cell properties to give magic a hint about how the obstruction layer should be created. Magic aims for "sane" behavior, but that's not always what the designer intends. It's also not clear to me what this has to do with the PDN, though, because the LEF view is at least valid, even though it may not be ideal.