https://open-source-silicon.dev logo
Join Slack
Powered by
Both XOR and Magic(DRC, locally) has been running ...
# caravan
c
Both XOR and Magic(DRC, locally) has been running for 4-5hrs now... No meaningful outputs at this point.
m
I think the servers are over loaded..
c
I see... Mine is still running (18hrs), some of my classmates runtime is around 30 mins. I thinks maybe something is wrong with my submitted GDS...
t
@Can Wang: Just pathologically bad for magic's inefficient representation of labels. Did you flatten everything into the wrapper layout? It appears to have a huge number of labels in it.
6.662 million labels, to be exact. I think this may run for a very, very long time. Is there any way you can generate GDS without flattening everything?
c
@Tim Edwards Thanks for checking! I ran into some issue that I think it is not generating the correct GDS so I flatten the whole thing.
t
For the purpose of getting to tapeout, could you please remove the labels before generating GDS? If you flattened in magic, then use "flatten -dotoplabels". That will keep essential ports and things from the top level cell without dropping 6 million text labels into the GDS file.
I do intend to fix magic's handling of labels---I already have some code for that---but it isn't going to happen for this tapeout. You probably need to ask Jeff DiCorpo to kill the precheck process so you can submit with a revised GDS.
c
Sure. I can do that!
Got it!
How do I check for label numbers? I saw <flabel> lines there
Yeah, I found it, nvm
I tried this, it seems that the output mag file has similar numbers of flabel as the previous submitted one. Or should I only include ports? @Tim Edwards
t
Huh. Maybe "-dotoplabels" doesn't work as advertised. Try just "-nolabels". I know that one works.
c
My problem is that I cannot do simple save to generate a hierarchical mag/gds since my design is integrated by several macros
Any command that will descend into subcells and save them as mag file as well?
I will try -nolabels too.
t
I don't think you need to be concerned with generating hierarchical GDS, just get rid of the labels to keep magic from running forever on reading it in for DRC.
c
-nolabels eliminate all the labels including ports
t
That should work for DRC, and for tapeout.
c
I see
For LVS, can I maunally delete the flabel and leave the ports at the top cell level?
I passed XOR, and the magic drc only reports drc error on stcell (decap and dlygate), hope this is okay!
Thanks for the help and support along this quarter! @Tim Edwards
t
For the record, which standard cells report errors? Some of the cells require manual intervention during the open_pdks installation to get them to pass the DRC check by giving Magic additional information on where the mask layers are. I fixed a few of the cells, but the list isn't comprehensive, so apparently you have run into a few additional cells that need the same attention, that I wasn't previously aware of.
c
Copy code
sky130_fd_sc_hs__dlygate4sd3_1.gds
Diffusion contact to gate < 0.055um (licon.11)
sky130_fd_sc_hs__decap_8.gds
Diffusion contact to gate < 0.055um (licon.11)
https://github.com/google/skywater-pdk/issues/261#issuecomment-794802799 @Tim Edwards
40 dlygate cells in total. The rest is decap fill.
t
Ah, the HS library. I haven't gotten around to fixing that one yet.
c
Sorry for adding extra work to you... Let me know if anything I can help! @Tim Edwards
13 Views
Open in Slack
PreviousNext
Powered by