It might help a tiny bit if your design is just on the edge in terms of metal density, but decap_8 is still pretty dense. The filler_placement routine will try to fill the empty spaces in your design with the largest filler cell that fits (combined from DECAP_CELL and FILL_CELL). If you remove decap_12 from the list, most of your area will be filled with decap_8, so your only advantage will come from the small areas where adjacent decap cells meet which is not covered by metal. If your li1_ca_density in the tapout was, say, 0.39 instead of the 0.4 required, it may help. Otherwise the improvement won't be noticable.

@User It is 0.391376… Is it going to help?

I'd guess it won't be enough, but you can try it. (I know it takes time and there is 21 hours to the deadline.)

I am trying to use

sky130_fd_sc_hd__decap_12

instead of s`ky130_ef_sc_hd__decap_12` and keep the other cells as @User told. Let's see what happens

@User Is it going to help if I remove DECAP8?

why?

If you remove decap_8, most of your design will be filled with fill_8 which will avoid the li1_ca_density problem, but you'll get a different one with fom_pattern_density being too low. If you remove both decap_8 and fill_8 (from FILL_CELL), it's again a slight improvement from just removing decap_12, but it might not be enough. Also your design will be filled with more smaller cells so everything will take longer.

@User OMG!

The real solution would be to use

sky130_ef_sc_hd__decap_12

, but there seems to be some problem with that cell which falls outside my area of expertise. It's something that @User could help with if it wasn't the middle of the night in PST.

@User Tim is working on the issue but it wont be ready till the deadline! It needs a kinda PDK patch.

@User I think there is no fill_3. There is decap_12, decap_8, decap_6, decap_4, decap_3 and also fill_8, fill_4, fill_2 & fill_1. Anyway, if there is a DECAP_CELL with the same width as a FILL_CELL, DECAP_CELL takes priority as it appears earlier in the filler_placement argument list. But you can try to set DECAP_CELL to "sky130_fd_sc_hd__decap_3" and FILL_CELL to "sky130_fd_sc_hd__fill_1 sky130_fd_sc_hd__fill_2". It's not the real solution but it might help in the interim.

@User Thanks a lot. Really appreciate it.

@User Thanks a lot! We will also try it

I also have an experimental patch that fills the empty space with a mix of decap & fill cells. If you are comfortable building your own openroad & openlane, you can find it here: https://github.com/htfab/treepram/tree/main/patch. Then you can set DECAP_PERCENT in your config.tcl to something around 75. It's tricky because a too high value will cause a li1_ca_density error and a too low value will cause a fom_pattern_density error. If your project builds & checks fast enough to experiment, you could try finding the sweet spot.

@User Is there a way to find out the density without going through the tapeout phase?

@User how can I install it?

@User I don't have a really good way to do that, so if you learn about it please also tell me. My heuristic so far is to run a local precheck (or just the

klayout_met_min_ca_density

step) and look for

li1_ca_density

in

precheck_results/*/logs/klayout_met_min_ca_density_check.log

. Now that's the li1 complement density for just your part of the chip, not the full integrated one, so it will have a pretty different value, but as a rule of thumb, if you decrease the metal density in your part of the chip then the total density will also decrease to a smaller extent and vica versa.