A structural report lists every reset-domain crossing it can see, including crossings the designer intended. Classification is the judgement that turns that list into work: each finding becomes a defect to fix or an acceptable crossing backed by evidence. An unclassified finding is an open finding.
Finding | Usual disposition | Evidence needed |
|---|---|---|
Asynchronous deassertion with no synchroniser | Defect. | None: add a reset synchroniser or re-parent the reset, then re-analyse. |
Reset crossing with a synchronised, ordered release | Acceptable. | The synchroniser structure plus passing reset-aware tests of the release sequence. |
Quasi-static signal sampled while reset is settled | Waiver candidate. | Proof the signal cannot change near release, plus review approval. |
Reset tied to a constant in this configuration | Waiver candidate. | The configuration record showing the tie-off is permanent, plus review approval. |
Undocumented reset or unmapped register | Defect in the intent. | Completed intent and a re-run; never waived. |
Shared synchroniser across unrelated domains | Defect: split the synchronisers. | A waiver with skew analysis exists only for synchronous, related domains; across asynchronous domains the release cannot be aligned to both clocks. |
Ask one question: does the receiving logic see a clean, ordered release under every allowed reset sequence? If the answer is yes and there is proof, usually the synchroniser plus a passing reset sequence, the finding is acceptable. If the answer is yes but only by argument, it needs a waiver with that argument written down and approved. Anything else is a defect.
The review reads the classification, not just the count. It checks that every finding has exactly one disposition, that no defect is closed by argument alone, and that each waiver names its evidence and expiry. A report where the defect count fell because findings were reclassified without new evidence goes back for rework.
Defect. Rarity is not synchronisation: the first assertion can still release into metastability. Add a reset synchroniser or re-parent the reset.
An acceptable finding meets the signoff rule directly, with proof such as a synchroniser plus passing tests. A waiver candidate is safe only by extra argument, such as a settled quasi-static input, and needs that argument written, approved, and expired.