open-source-silicon.dev

image.png

Hi <@U016EM8L91B>, we used `gds write` command to write a layout with pwell into gds format. Then we used `gds read` to look at the layout and found the pwell disappear. How can we solve this issue? Thanks!

You can't.  Magic does not guarantee that GDS out and GDS in are symmetric.
In this case, "pwell" is not an actual mask layer.  There is no GDS layer for "pwell", so it does not show up when you read back the GDS.

Magic's setup for sky130 is agnostic about the pwell layer.  If you have it or don't have it, all DRC and extraction should work the same (briefly that was not the case, while I was developing a new method for substrate extraction in magic, but that's fixed now).

When we open our lef file, we can't see pwell either. So how can we check whether our pwell is existent or not?

nmos_flat.mag

<@U016EM8L91B> Thanks for explanation. We'd like to confirm this as it is quite weird that even loading from a single cell mag, some part of the pwell couldn't be seen while some part of pwell could been seen after gds/lef write and being gds/lef read again. We are not sure will this lead to any tape out issues or not (such as our VNB is connected to pwell to VGND, while some pwell is missing in the GDS). Is there any good way to fully write the whole gds/lef for pwell? I attached the `nmos_flat.mag` that we are unable to write gds/lef and read back to check the pwell using `gds/lef write` command. Could you please help have a look at it?

<@U030TNQLQG4>:  As I said, "pwell" is _not_ a GDS mask layer.  It is an optional drawn layer in magic, and it represents the p-substrate.  However, absence of pwell (or just "space", in magic's terminology) also represents p-substrate, so it doesn't matter if the drawn pwell layer is there, or not there, or only half there.

It is mostly just left over from the time that the PDK was first created, when I found that magic's substrate extraction method could not deal properly with deep nwell, the way SkyWater had defined it.  Adding the pwell layer was a "quick fix" at the time to get (mostly) correct extraction.  Since then, I have worked on magic's extraction methods, so the pwell layer is no longer required for correct extraction.

<@U016EM8L91B> Just to confirm again before submitting our tapeout, for the nmos, the pwell is appearing in the final user_analog_project_wrapper.gds or not doesn't matter right? In our case, the pwell even doesn't cover the nmos and gnd rail together.

Yes, that's correct.  Pwell is just a substrate marker layer and does not correspond to any mask geometry.