@Tim Edwards @Mitch Bailey @jeffdi any thoughts on this: https://github.com/efabless/mpw_precheck/issues/141

I'm not an expert on this, but I believe what the is saying is the clear density has to be at least 40%. Which means the pattern density has to be less than 60%. The error indicates a high density of m4. Do you see that in the layout? Can you reduce the m4 usage? I don't think fill effects this because (I assume) fill is only added at low density areas (although this may be a windowing problem).

I am leaning towards the windowing issue. We don't have a particular large m4 usage

Only FOM (diffusion density) is windowed. I'm inclined to think that this might be a klayout error. Try running magic's density check. In theory, magic's and klayout's checks should give the same result. It's a bit hard to debug density checks because you get a result that flags the entire chip area. But you can do some sanity checks like dropping a large metal 4 square on top of the chip and making sure that klayout reports that the density problem got worse proportionally to the amount of metal 4 you added.

We tried the latter and klayout reports it gotworse actually

That is correct. The more metal you add, the smaller the percentage of clear pattern.

Magic reports that metal4 density (metal density, not clear area) within the user project itself is 6.9%.

% cif cover MET4
Cell Area = 10329984000000 CIF units^2
Layer Bounding Area = 9888485739600 CIF units^2
Layer Total Area = 713447009075 CIF units^2
Coverage in cell = 6.9%

This might require getting @jeffdi to manually override the precheck, while an issue is raised to root out and fix the problem, which (given the existing evidence) I conclude is either an error in klayout or in the density check definition.

@Tim Edwards thanks. Can you show me and @Jianwei Jia how to run this locally

Because of an obscure bug, I expect. To run what I did above, I just read the GDS of the user project into magic, selected the whole top level cell, and typed the command "cif cover MET4". There is a script installed with open_pdks under `sky130B/libs.tech/magic/check_density.py`; you can just run it giving the GDS file as argument. Results will be different than the full Caravel chip. Also you will get lots of metal-under-density errors because the script is intended to be run on a layout after fill generation. When I run it, I get all metal layers under density, which is what I would expect from visual inspection of the layout. So, still all evidence points to this as being a klayout or klayout density rule description error.