Analog IO routing in OpenRoad using ODB

[smunaut]

I need to route analog signals. I don't expect the autorouter to do that of course. For that, I use an ODB script using "Non-Default Rules" and then manually using a wireEncoder and such.

This works perfectly fine for internal analog signals between blocks. Convenient and easy to merge with the normal flow.

Problem comes when I need to connect to analog pads. Both the IHP and the GF180 Analog IO pads only have a single signal ( PIN in the LEF ) to represent both the actual bonding pad ( that will show up as a BTerm in the top level chip ) and the internal side of the analog signal.

This means in OpenROAD the net ends up marked as "SPECIAL" and in the "SPECIALNETS" section of the DEF. And it seems creating routing using wireEncoder for such nets doesn't work ... the ODB script works but nothing is written in the DEF output about the created route and then at the DRT steps it just segfaults.

So I'm a bit lost there. How should it work ?
Should OpenROAD accept routing using wireEncoder for special nets ? Should those not be marked special ? ( and how to do that ) ? Should the PDK IO pads have different PIN for the inside and outside part of the analog signals ? ( Even though electrically they are directly connected with nothing in between ).

I also posted in the OpenROAD channel on the fossi matrix chat ( https://matrix.to/#/!xoVuRnQEpAxGcPJbVq:fossi-chat.org/$YtUioev2B3qwsOZu8VCqSwy3gC6cx9pkR8phaRDaKF4?via=fossi-chat.org&via=matrix.org&via=fau.de )

[maliberty]

Wire encoder is for regular nets not special nets. Special nets shapes belong to dbSWire

[smunaut]

Looking at the LEF specs you can also specify routing points, but syntax is different and we'd need something like a dbSpecialWireEncoder which doesn't exist.

I've now taken the approach of calling clearSpecial() on both the net and the ITerm.

That unfortunately doesn't work really either because in DRT I get Net pad_raw[19] already has routes.
coming from io::TopLayerBTermHandler::processBTermsAboveTopLayer(false) call made in readDesign

And I can't find any code path that lets me escape that, it just doesn't seem to be a way to have pre-routed/manually top level IOs that have shapes above the routing layer, it will always hit that.

[maliberty]

LEF has nothing to do with this. There is no meaning to dbSpecialWireEncoder as special nets have geometric nature and not endpoint connectivity (they are essentially a pile of shapes, which is what makes them special). Just write the shapes directly with dbSBox/dbSWire.

drt doesn't expect to see nets pre-routed.

[smunaut]

We pre-route nets all the time in TinyTapeout, that's the only way to make it work.
DRT deal with it just fine for internal nets, it just skips them, but not in this case because top level IOs seem to have special handling which doesn't cope with it.

[maliberty]

@osamahammad21 thoughts?