Hello, @Tim Edwards. I've tried to make CACE work with my project and it refuses to. It seems like a user problem. It fails to generate the netlist for the simulations. I already have the xschem testbenches working for the typical case, but I can't adapt it to CACE. Maybe it is a file naming problem. Can you look at it? My project is at https://github.com/lhrodovalho/sky130_rodovalho_ip__lpopamp/

I'll do that right away.

@Tim Edwards, assume that you already have not only a reasonably decent implementation of a de Langen-Huijsing mesh Class-AB amplifier, but also quite sophisticated test functions based on the ngspice let statement only (so CACE isn't really needed for anything but avoiding to get eliminated by the chipalooza reviewers) how would CACE fulfill the promise made by Kevin Zheng (who might be a Suit) and Kevin Kundert (who most likely isn't a Suit) to facilitate design re-use, e.g., saving and re-using a DC operating point solution that takes forever to converge, by judicious use of ngspice .NODESET and .IC statements? "There's plenty of room at the bottom." — The man, the myth, the legend.

@Christoph Maier: CACE isn't needed for anything. The point of it is to make it easy to run the same testbench many times across different corners and conditions, and do it in a way that is consistent across lots of different circuits, different processes, and different tools, so that if somebody needs a particular circuit, they can find one on github, immediately know whether it (roughly) does what they need it do, and can clone the thing and run simulations that show that it does, in fact, meet the specification. They also have access to lots of testbench schematics that demonstrate exactly how to measure any property of the circuit that they might want to measure. It's there for re-use, it's there for learning, and it's there to expand the open-source ecosystem.

I can think of two business plans. In one of them, CACE is very nice to have but optional. In the other, CACE is absoultely essential, and its bugginess is not a bug, but a key feature. But maybe I should think about some variations on the subject of composite transistors for a while. If you want to design building blocks to be shared with others, it probably be helps to design them based on concepts that the potential collaborators are very familiar with.