LVS

Layout vs Schematic (or LVS) is an important verification step.

After the design has been finished and we have the GDS2 files that we can send to the foundry, we want to check that the design is the same as the input that was described by the HDL or schematic.

In the OpenLane tool, at the end we have the LVS step. The netlist is extracted using Magic A tool called Netgen can compare this extracted netlist with the one we get after the synthesis step.

You will get some reports in the run directory. In the inverter example I have this file: results/lvs/inverter.lvs.log

The end of the file shows the input and output are equivalent.

Subcircuit pins:
Circuit 1: inverter                        |Circuit 2: inverter                        
-------------------------------------------|-------------------------------------------
in                                         |in                                         
out                                        |out                                        
VGND                                       |VGND                                       
VPWR                                       |VPWR                                       
---------------------------------------------------------------------------------------
Cell pin lists are equivalent.
Device classes inverter and inverter are equivalent.
Circuits match uniquely.

The most common types of LVS errors I’ve seen with OpenLane are:

  • when power supplies aren’t connected for some reason,
  • there is a short circuit during routing.

Verification essentially tries to show that the final design meets the designer’s intention. This takes various stages:

  1. simulate the initial design (HDL or schematic) to show it behaves in the intended way
  2. synthesise this down to a netlist,
  3. simulate this again to show it still works - or use other tools such as formal proof to show it is equivalent to the original design
  4. place and route it to get a final design
  5. use LVS to show that this design is still the same as the netlist from synthesis
  6. use other steps (simulation, or preferably STA) to show the final design performs well enough

Course feedback

If you've got any interest in how the sausage is made you should get on the course and you should dig in and find out more. I mean this was the work of secret witches and wizards in mysterious cloaks casting strange incantations over a cauldron! This was all secret stuff and I love that this project is trying to do to silicon design what the open source community has been trying to do with software for the last 30 years. This feels like the next logical step and I think we'll look back and say well of course you can make your own chips that's just a thing and it will just be obvious and commonplace and I look forward to that.

Jonathan Pallant