Deep DRC failure: the dbu case

[Scafir]

Hi Matthias, hope the FSiC preparations are going well,

While cleaning up a few things, I think I have stumbled in a deep DRC bug.

Here is the MWE example:

deep_issue.zip

Running the reproducer with klayout -b -r reproduce_enclose.drc gives:

The only difference between the two structures is that on the left, the enclosing polygon lives in a different cell (TOP) than the internal polygon (BADCELL). On the right, both polygon live on the same level.

This issue does not appear in flat mode, leading me to believe that there is a bug in the deep DRC somewhere.

It is worth noting that this issue happens only when the dbu is 0.005. This leads me to believe that somewhere, deep DRC assuming the dbu to be 0.001 rather than using the one from the design. You can see this because in the reproducer, the drc issue is raised when the distance is smaller than 50um instead of 10um.

[klayoutmatthias]

Hi Clyde,

thanks for that report. I will take a look. There is no specific assumption that the DBU is 1nm, but I will check.

Best regards and see you :)

Matthias

[klayoutmatthias]

Hi Clyde,

I think the problem is because of the scaled instance. The BADCELL is placed with a magnification of 0.2 and apparently that triggers the bug.

If I scale the content of that cell instead of placing the cell in a scaled fashion, it works nicely. So it's not the DBU, it's the scaling of the instance.

I will debug the issue, but basically scaled instances are deprecated, because scaling cells imposes manifold problems like grid compliance or simply because you can't just scale a device without changing the physical properties fundamentally.

Best regards,

Matthias

[Scafir]

Hi,

That would make a lot of sense. I forgot that I used this feature as I encountered the bug on an old design.
I think printing a warning (stdout and/or GUI window) would be the best fix, as I did not know this feature was deprecated.

Thanks for looking into this!

[klayoutmatthias]

Very good. I still consider that a bug of KLayout and I will fix it.

By "deprecated" I don't mean forbidden in general (for some applications like MEMS it may make sense). But it does not make sense in CMOS for example. If you use that feature in CMOS you should receive a lot of DRC errors (provided the feature works :( ).

Matthias