Open d-m-bailey opened 2 years ago
I encountered a similar problem in a DAC design that I did that I surrounded with deep nwell. My hypothesis is that if a deep nwell is large enough, the cookie-cutter extraction method in magic isn't dealing with internal areas of the deep nwell, where it does not see the connecting nwell. But I haven't looked into it yet.
@d-m-bailey : I expected to find some commonality in the instances that are not connected, but they seem to be selected at random. The VNB ports of some cells just don't appear anywhere in the .ext file, where every one of them should be in a "merge" line.
Once again, sorry for the late reply. I've been rather busy for the last two weeks. This one came in while I was on travel, so it got put on the back burner. I'm looking into it carefully now.
There is a point of commonality. If the "cookie cutter" square being extracted doesn't contain one of the taps, then the VNB port of any subcells falling within the same square will be lost. That should be enough information to go on to fix this error.
One hack solution to this (while I'm busy trying to actually fix it properly) is to change the extraction step size in the magic tech file; the existing step size is 7um (step 7
in section extract
in sky130A.tech
). If you increase this to, say, 15, then there should always be a tap in the extraction square, and the issue should disappear.
Magic 8.3 revision 308 Tech file version 1.0.291-20-g05af1d0
MPW-5 slot-009 armleo gpio mpw5 test chip
There are standard cells placed in dnwell region. The pwell connections (VNB) for some standard cells are not being connected for some standard cells. Notice the
<instance>/VNB
nets.To duplicate, extract the attached tarball, and
armleo_gpio.tar.gz