Skip to content

8.3.695

Choose a tag to compare

@github-actions github-actions released this 06 Nov 08:03
· 30 commits to master since this release

commit 8.3.695

"Tagging version 8.3.695"

See documentation at https://github.com/RTimothyEdwards/magic/blob/48d0ea2a/appimage/7/README.md
See documentation at https://github.com/RTimothyEdwards/magic/blob/48d0ea2a/appimage/8/README.md
See documentation at https://github.com/RTimothyEdwards/magic/blob/48d0ea2a/appimage/9/README.md
See documentation at https://github.com/RTimothyEdwards/magic/blob/48d0ea2a/appimage/10/README.md

Change Log (since previous tag):

48d0ea2a Modifications similar to the last commit, which are designed to prevent magic from generating output for capacitance in a .ext file (and subsequently in a .spice file) in order based on memory address, which is effectively random from run to run. Output is now confirmed to be reproducible on every run for basic SPICE netlist extraction (this commit) as well as R-C netlist extraction (previous commit).  The solution was implemented by Claude Opus 5.5.  It was tested for performance with no appreciable degredation in performance due to the sorting routines, which take negligible time compared to the rest of the extraction, and have relatively low memory overhead.
187205b3 A fairly straightforward set of changes to the "extresist" code done by Claude Opus 5.5 which applies sort routines at various points in the output to ensure that the output is consistent from run to run.  Otherwise, the order in which nodes and terminals are output changes from run to run and makes run-to-run comparisons difficult.  This resolves github issue #538.  It does not address a similar problem stemming from capacitors being written to .ext files in unpredictable order.
5bd74f64 Claude Opus 5.5 modified code related to resistor extraction, resolving a few loose ends which I was previously aware of but had never addressed.  Most or all of these changes are also addressing a number of recently posted issues on github between correctly handles resistor length for modeled resistors (e.g., when length is a parameter passed to the model).  Coupling capacitance is now handled sanely for coupling between resistors and nearby objects as well as between two resistors, with appropriate differences for the "pi" and "tee" models.  Extracting resistors no longer produces floating coupling caps to a resistor node that is otherwise undeclared in the netlist.
b224efbf A bit of an overhaul of the timestamp and bounding-box mismatch checks, precipitated by continued difficulties with magic trying to re-write known clean vendor GDS files.  Eventually settled on several modifications of existing methods, effectively treating vendor (PDK) GDS as "trusted".  Fixed a surprising problem with simple "paint" not generating check areas for DRC.  This leaves "gds drccheck off" being the only way to make a cell be marked both clean and checked without having run a check.