What changed
v0.2.0 adds a deterministic held-out 4?4 behavioral characterization workflow with explicit device, stimulus, calibration, and operation seeds; paired pre/post-calibration observations; per-density and overall aggregates; and a sealed machine-verifiable receipt.
Evidence
- Maturity: Behavioral-model released
- 4 modeled device seeds
- 384 paired matrix-vector evaluations / 1,536 output values
- Pre-calibration RMSE:
0.0176779832054 - Post-calibration RMSE:
0.00792159411774 - Post/pre RMSE ratio:
0.44810508222 - Saturated outputs:
0 - Characterization receipt SHA-256:
892e6e72ede2f615fc9dd34242335d426b6a170c6462ae00d9cffb4f0e58604e - Independently approved exact Git tree:
473aa908c1df1972568b65dcd4331206d3ce4241 - CI passed on Python 3.11 and 3.12 plus portable C and Icarus RTL checks.
The values above describe only the declared deterministic synthetic experiment. They are not population estimates.
Reproduce
python -m analog_matmul_lab characterize-4x4 --output characterization.json
python -m analog_matmul_lab verify-receipt characterization.jsonSee docs/CHARACTERIZATION.md and examples/characterization_4x4_receipt.json for the full protocol and retained observations.
Claim boundary
This release does not establish SPICE, transistor, layout, silicon, ENOB, measured energy, yield, training-silicon, or commercial-readiness evidence. On-chip analog weight update remains out of scope.
Permanent archive
- Software Heritage request:
2410672 - Snapshot:
swh:1:snp:1f6120c70ed00cc9491cbd76ff367821244885ff - Archived
main:45120539bcafaefa564caaae00f31a6f0f366bb6 - Archived annotated tag:
refs/tags/v0.2.0