Reset answers live or die on ordering arguments. Say what releases when, relative to what, before naming any circuit.
Field | Answer |
|---|---|
Interviewer is testing | Scope thinking. |
Bad answer | "Reset is just another input." |
Better answer | "Reset fans out everywhere and releases everything at once: assertion races no clock, deassertion races every clock, and release order between domains is a system property." |
Strong answer (a 4) | Adds that CDC asks per-path while RDC asks per-domain-plus-order, and cites intent (membership, polarity, release rank) as the input CDC never needs. |
Follow-ups | What reset intent lists; one versus two reset domains on one net. |
Common trap | "The CDC flow covers resets". |
Evidence to cite | RDC structural report against reviewed intent. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | Pattern precision. |
Bad answer | "Two flops on the reset." |
Better answer | "Asynchronous assert into a two-flop chain in the destination clock, synchronous release; no logic between flops, chain clocked by the domain it releases." |
Strong answer (a 4) | Adds recovery and removal as the reason (release races the clock), covers the synchroniser's own reset domain, and refuses simulation as proof. |
Follow-ups | Recovery versus removal; what happens with logic between the flops. |
Common trap | Synchronising the assert edge instead of the release. |
Evidence to cite | Structural legality plus recovery/removal timing. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | System argument. |
Bad answer | "Release both together; order does not matter." |
Better answer | "Rank the releases, hold the dependent domain until the first is out, check as an ordering violation." |
Strong answer (a 4) | Names the dependency (clock and reset stable first), the holding mechanism, the simulation of the ordered sequence, and the variant-order trap from the crunch scenario. |
Follow-ups | Nested versus sequenced domains; warm reset. |
Common trap | Assuming one product's order covers all variants. |
Evidence to cite | RDC ordering check plus reset-aware simulation of the sequence. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | The intent reflex. |
Bad answer | "Yes, it boots." |
Better answer | "No: simulation samples some sequences at one corner, and without written intent the checks had nothing to verify against." |
Strong answer (a 4) | Lists the uncovered sequences (mid-operation assertion, partial reset, every variant order) and names the full package (intent, structural re-run, reset-aware tests, assertions). |
Follow-ups | X-propagation in reset simulation; which corner matters. |
Common trap | "Gate-level boot passed". |
Evidence to cite | Intent plus analysis plus sequences. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | Report literacy. |
Bad answer | "Any reset warning is a bug." |
Better answer | "Unsynchronised deassertion, cross-domain reset fanout, ordering violations: each dispositioned as fix, waive with evidence, or accept." |
Strong answer (a 4) | Walks one example end to end (intent row, finding, classification reason, evidence or fix, re-run) and states what a waiver needs. |
Follow-ups | A finding you would refuse to waive; expiry. |
Common trap | Severity labels as verdicts. |
Evidence to cite | Classified report with per-item rationale. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | Definition under pressure. |
Bad answer | "One: it is the same net." |
Better answer | "Two: a reset domain is defined by shared release behaviour, not the net name." |
Strong answer (a 4) | What a 4 sounds like: "Assumption: the two synchroniser branches have independent latency. Mechanism: the blocks observe deassertion on different cycles, so state handed between them crosses an ordering boundary. Trade-off: unifying the branches costs skew control; keeping them costs per-crossing checks. Evidence: intent rows naming both domains with their release relationship, structural analysis of every crossing between them, and a reset-sequence test covering the skew. The net name is block-diagram thinking; the release behaviour is signoff thinking." |
Follow-ups | Shared-tree skew; which register belongs where. |
Common trap | Merging same-named hierarchies in intent. |
Evidence to cite | Intent plus analysis plus sequence test. |
Taught in |
A sentence about release order across the affected domains: which releases first, what holds the second, and what check proves it.