Is anybody aware with Magic VLSI doing some strange extraction for LVS for bulks of same hierarchy level cells? I am trying to extract a cell calles armleo_gpio_esd_wrapped which contains 1 transistor + 1 cell, but the transistor is placed in cell's deepnwell region, however, when doing and LVS extraction the transistor's Substrate does not match expectation: behaving the same way, as the transistor is placed in flattened design. I was expecting it to have same connection as taps near it (vssio), but instead it uses a net that is in completely irrelevant deep nwell (vss). If I just flatten the cell and then connect the transistor, it behaves correctly and substrate pin (bulk) is parsed correctly

Can you please post the layout?

One with issue: https://github.com/armleo/armleo_sky130_ip/blob/main/gds/user_analog_project_wrapper.bulk_issue.gds Correctly behaving one: https://github.com/armleo/armleo_sky130_ip/blob/main/gds/user_analog_project_wrapper.gds

You can flatten

armleo_gpio_ggnmos

virtually before extraction by using the

gds flatglob armleo_gpio_ggnmos

command.

That's actually a good idea, but probably I should flatten armleo_gpio, because armleo_gpio_ggnmos even flattened still causes same issue

@User: In the version with the commit ID that you gave above, I see cells

user_analog_project_wrapper

, which is empty,

armleo_gpio_esd_wrapped

, which has a flattened version of the cell in question, and

armleo_gpio_ggnmos$1

, which is the non-flattened cell, but doesn't appear in anything else. I'm not sure what I'm supposed to be looking at or how to reproduce the problem.

bulk_issue.gds: armleo_gpio_esd_wrapped when extracted armleo_gpio_ggnmos's bulk is connected to vss instead of vssio. There is no way that it can be connected to vss, I checked everything the other gds: armleo_gpio_esd_wrapped's subcell armleo_gpio has been flattened, as a result the DNWELL is no longer in hierarchy. It's the only change, but in that case the ggnmos's bulk is correctly connected to vssio as expected

I haven't changed the basic substrate extraction code since version 8.3.227, but there were a number of corrections to what ext2spice was producing that might be the problem here; that work was done around 8.3.235, so you might find that the behavior improves if you update magic.

Update: this solved the issue