FeCIM Lattice Tools

The engineering problem

A parameter sweep is only useful if another engineer can identify what ran, inspect the assumptions, and repeat the analysis. FeCIM makes those boundaries explicit for exploratory ferroelectric compute-in-memory studies.

This is our own research software, not evidence of commissioned customer delivery.

Define → run → reuse → inspect

The repository’s documented HZO example expands three parameters into eight runs. Its recorded first execution evaluates all eight. A repeat with a different worker count reuses all eight SHA-256 run identities.

First execution: 8 total · 0 reused · 0 failed
Repeat execution: 8 total · 8 reused · 0 failed
Configured study: 6 feasible points · 1 Pareto point

These are repository-reported results at the linked revision, not a new execution performed by this website. “Feasible” means meeting the example’s configured constraints. The Pareto point is a trade-off result under its objectives—not an experimentally established optimum.

What makes the workflow inspectable

• Explicit inputs: project, design, and sweep files define units, assumptions, constraints, and objectives.
• Stable identities: each run retains its resolved design, manifest, and result under a SHA-256 identity.
• Reusable outputs: JSON, CSV, and Markdown reports expose the study without requiring someone to reconstruct the author’s session.
• Visible failure states: failed or unusable points remain distinguishable from successful and feasible results.

Repeatability is not physics validation

The example demonstrates execution and reuse, not physics validation, measured silicon performance, foundry sign-off, or calibrated energy and area predictions. Simplified models and simulation defaults remain estimates even when their outputs repeat exactly.

The repository’s evidence review explains why a measured-versus-simulated headline figure is not presented. Scientific acceptance requires suitable experimental evidence and domain review beyond this demonstration.

Relevant to your workflow?

The transferable engineering pattern is a bounded input contract, identifiable executions, explicit checks, and a usable handover. It is not a promise that your scientific model is correct or that every integration is already supported.