Incredibly high memory usage when characterizing entire SCL

[89Mods]

Problem Description
Trying to characterize the entirety of the gf180mcu_as_sc_mcu7t3v3 SCL runs out of memory after consuming slightly over 16GiB at 27% progress. Assuming linear scaling, this means just under 64GiB would be required to characterize the whole library.

To Reproduce
A reproducible can be found here

System Configuration

• OS: Manjaro Linux
• Python Version: Python 3.12.3
• CharLib Version: Installed from cloned repo
• PySpice Version: 1.6
• Spice Simulator: ngspice
• Spice Simulator Version: ngspice-42

Splitting the task up over multiple YAML files and then merging the resulting libs is a possible workaround, but this would not be straight-forward to automate since CharLib does not include a tool to merge lib files.

[infinitymdm]

It actually shouldn't be too hard to merge multiple liberty files with CharLib's current structure. There are a couple of pieces missing:

• A parser to read Liberty files into charlib.liberty.Group objects
• Some improvements to subgroups such that we can ensure all Group objects are uniquely identifiable (group name & id are not enough) during group merge.

I'd personally prefer an approach that keeps the YAML all together, as one of the main goals of CharLib is to keep all configuration information together in one place. It should be possible to defer additional simulation runs until system resources are available.

Finally, as a quick workaround to your specific issue: Have you tried running with fewer threads? If it's an issue of too many simulations running at once, running with e.g. -j4 should help quite a bit. See charlib run --help for details.