Passive circuit models from our extraction, checked by an outside tool

What it shows

A field solver’s answers are only useful to a circuit designer once they are written as a circuit model a simulator will accept. A fitted model can be non-passive, meaning that in simulation it creates energy, and some simulators then fail or give wrong answers.

The record states the result this way:

In plain words: the fitted models are made passive by an enforcement step. The export check plants a non-passive fit, enforces passivity, and checks the written file with scikit-rf, an outside library; with enforcement switched off the file stays non-passive, so the enforcement is what makes it pass.

The second half is a measurement against ourselves. Of the test geometries behind our impedance figure, 78839 in all, 35.5% have an exact twin in the training data. That makes the headline accuracy of our impedance stand-in a measure of interpolation on a fine grid, not of generalisation, and we say so here.

Why it matters

A circuit model that is not passive can break a simulation downstream, far from where the error was made. And an accuracy figure measured on test data that overlaps the training data says less than it seems to. Both are the kind of thing a buyer finds late; this page states them first.

What is ours, and what is not

Fitting circuit models to frequency data and enforcing passivity are standard methods, and data leakage is a known failure (see the prior art below). What is ours is this export path for our extraction, the check of the written files, and the published leakage measurement.

Who should care

• Signal-integrity teams who take extracted models into circuit simulators.
• Reviewers of our accuracy figures. The leakage measurement is the reason to read them as interpolation.

The limits, in the record’s words

In plain words: the export is made passive by enforcement, checked on the tested case, but enforcement does not converge on one larger model type, multi-conductor transmission lines. The leakage figure belongs to the training data, not to any single model, and it is why our impedance stand-in’s headline accuracy is not a claim about new designs.

Open source for this step

Tools and datasets we publish for the package step of building a multi-chip package. They are the checkers around this work, not a copy of the result itself.

• physics-lint: One command that checks a folder of physics models against a fixed set of named physical rules, with findings straight into CI.
• maxwell-lint: Flags a coupling extractor whose answers no passive set of conductors could produce.
• sparam-lint: Is your signal-response model physically possible? Five physical laws checked from the command line.
• interval-core: The interval arithmetic core behind our proofs over whole families of layouts.
• touchstone-tools: Read, write and convert Touchstone files, the standard text files that record how signals pass through a package's connections, and refuse to write one that cannot be read back.
• physics-lint-mcp: The physics checks, callable by an AI agent.
• physics-lint-action: A GitHub Action that fails the build when a model breaks one of a fixed set of named physical rules.
• Signal-response validity corpus: A labelled corpus of physically invalid signal-response networks, and a scorer that grades any checker against it.
• screening-ceiling: The screening-ceiling family as an open dataset.