@User Was there a performance hit between tech files? Using tech file version

1.0.264-0-g27ecf1c

, caravel extraction took about 4 hours. With version

1.0.291-20-g05af1d0

it took over 10 hours.

There is now additional processing of fringe capacitance to account for shielding of the fringe capacitance by neighboring shapes. I would have expected a performance hit but not a 2x performance hit. However, if you are extracting for LVS you can try doing "extract no capacitance" and "extract no coupling" (avoids parasitic capacitance calculations). If that doesn't solve the problem, then it's something else, and could be a change in caravel. There are inadvisable ways to make the extraction run very slowly.

Thanks, I'll look into that.

@User Follow up question: Are there maybe advisable ways (wich might not be obvious to the unsuspecting user) to speed up extraction? 🙂

@User: There are some which are situation-specific. Above I mentioned using "extract no coupling" which works if you are extracting for LVS, but then you have .ext files that are missing parasitic information so you need to remember to re-extract before doing "ext2spice" if you want parasitics to show up in the output netlist. Adding lots and lots of labels, such as by doing "def read ... -annotate" for a large digital block, is a great way to slow magic down to a crawl. Having the wrong extraction step size in the tech file can slow it down, although you can usually assume that the work of figuring out the optimal step size has already been done for you. Shallower hierarchy results in faster extraction. Cells which overlap other cells slows down extraction. The best way to reduce extraction time is to complain to me that it is taking way too long, and send me an example. : )

Excluding the capacitance extraction reduced the run time to under 3 hours.