Hello, for some reason I am getting a DRC error with Caravel when trying to do a tape out on efabless. Can somebody help? This is the error: STDERR: [ WARN ] Poly Density Check Result: 'caravel_24114807.oas' has 1132 DRC violations.

Yes, there is a new poly density rule. A new version of the PDK should be out before the next chipignite that has decap/fill cells that don't violate the new rule.

Is there a way that I can get over that so that the tape-out is successful on efabless? I am doing it for the UNIC-CASS program, and this is a mock-tape out

You'd have to use some script (like with gdstk or such) to manually edit your GDS and substiture

decap

cells for

fill

cells until you meed the 38% poly density rule.

@Nikola Babić There is an openlane parameter for the types of cells that can be used for decap. I think decap_3 is about the right density. If you remove the other decap cells from the list, you might be able to pass the density check.

Thank you, I will try it

In the

openlane/<cell>/runs

directories, I believe there’s a

config.tcl

file that contains all the implicit and explicit setting. Look for

decap

there and set that variable. Here’s and example of the

.tcl

setting.

set ::env(DECAP_CELL) {sky130_ef_sc_hd__decap_12 sky130_fd_sc_hd__decap_8 sky130_fd_sc_hd__decap_6 sky130_fd_sc_hd__decap_4 sky130_fd_sc_hd__decap_3}

You could modify that for a json setting only using

decap_3

So I just delete all except decap_3?

From what I know, that’s all you need to do. Please post your results (whether it worked or not).

I will. Thank you. I couldn't find the directory that you pointed to on the efabless repo so I have tried another which seems to have the same, or similar structure. It's

signoff/<cell>/config.tcl

. I'll see whether that works or not.

Don’t modify that file. Just use it as a reference to find the configuration to add to your

openlane/user_proj_example/config.json

file.

@Nikola Babić Mitch is right and there's an extra line that should help too. Add both of these lines somewhere to

openlane/user_proj_example/config.json

(and delete existing 'DECAP_CELL' and 'FILL_CELL' lines if you any in that same file):

"DECAP_CELL": "sky130_fd_sc_hd__decap_3",
"FILL_CELL": "sky130_fd_sc_hd__fill_2 sky130_fd_sc_hd__fill_1",

Please let us know your results: i.e. whether you reharden and this then allows you to pass tapeout, or if it does not work.

@Anton Maurovic (efabless support) I'll try it, thanks. I'm currently away but when I get back I will inform you about the results.

@Nikola Babić At the moment we're waiting on an updated PDK with a different fill/decap solution, and for customer tapeouts we've been fixing the over-density decap cells on our side after the tapeout job has been run. I understand your need for the UNIC-CASS program, though. Can you tell me which shuttle you created your project on (CI2409 or CI2411), and what the repository URL is?

@Anton Maurovic (efabless support) Thank you. The shuttle is CI2411, and the repository URL is https://repositories.efabless.com/nikolab2809/UNIC_CASS_FP_Adder

What was the density after the change?

I am not sure. The only thing I noticed is that the number of DRC errors didn't change. I would have to look into it later as I am not at my computer right now.

It would be nice if you could indeed verify that there were no

decap_12

cells used in the layout.

@Nikola Babić I can see in your latest commit on repositories.efabless.com that you added the correct lines to your config.json files, but the updated GDS (i.e. files in

gds/

) is not included in this commit. Our server uses whatever is in

gds/user_project_wrapper.gds

(for your type of standard Caravel project) to run the precheck & tapeout jobs, so in addition to changing config.json, you will also need to re-run the OpenLane flow on your side if you didn't already do so after the change (i.e. reharden by doing both

make user_proj_example

and then

make user_project_wrapper

) and ensure your updated GDS files are committed and pushed to your repo on our server.

@Anton Maurovic (efabless support) @Mitch Bailey It works now. Thanks a lot to both of you!

Great to hear, thanks for the feedback @Nikola Babić 🙂