The chip-level RDC review is where all the evidence meets judgement. It is a meeting with fixed inputs, fixed questions, and recorded outputs, not a status update. Its verdict per reset domain is one of three: closed, open with named actions, or waived with recorded evidence.
Input | Without it the review cannot |
|---|---|
Frozen reset intent with revision | Judge any finding, because every verdict hangs off the intent. |
Structural reports on the final RTL | Confirm zero open defects rather than trusting a summary count. |
Finding dispositions and waiver records | Check that each finding has exactly one owner and one fate. |
Dynamic-check results and coverage holes | Confirm behaviour, not just structure, and see what never ran. |
Change delta since the last review | Spot RTL, intent, or IP changes that silently void earlier verdicts. |
For each reset domain: is the intent complete and reviewed, are all findings dispositioned, do the dynamic checks cover the required sequences, are waivers approved and unexpired, and has everything changed since the last clean run been re-checked? For the chip: do inter-block assumptions hold at every connection, is third-party IP contracted on every checklist row, and do warm and power-transition sequences have their own evidence? Any no becomes a named action with an owner and a date.
Per domain the verdict with its evidence pointers, the action list with owners and dates, newly approved or renewed waivers with expiry, and the delta statement: which changes since the last review were re-checked and how. The minutes are signoff evidence themselves, so they name revisions and runs, never just conclusions.
Open with a named action: re-run the dynamic checks on the final RTL. Structure and behaviour must both be proven on the same revision.
Because the minutes are signoff evidence. A conclusion without its evidence pointer cannot be audited later, for example to check whether a late change voided it.