Most weak CDC answers are unordered facts: synchronisers mentioned, Gray code mentioned, no argument connecting them. The framework fixes that by forcing an order. State your assumptions first so the interviewer can correct them, explain the mechanism so the fix follows, name the trade-off so you sound senior, and cite evidence so the answer ends at signoff, not at opinion.
Step | What you say | One-line example |
|---|---|---|
Assumption | The clocks, resets, rates, and intent you are taking for granted. | "I assume the two clocks are asynchronous and unrelated, and the pulse is one fast cycle wide." |
Mechanism | How the failure actually happens, flop by flop. | "The pulse can fall between two slow edges, so the capture flop never sees it." |
Trade-off | What each fix costs in latency, area, throughput, or complexity. | "A toggle synchroniser catches every pulse but adds two slow cycles of latency and forbids back-to-back pulses." |
Evidence | The check that would close it: simulation, structural check, formal, constraints, or waveform. | "I would cite the structural report showing a legal toggle synchroniser plus a formal proof of no lost or duplicated events." |
Question: "A one-cycle pulse in a 500 MHz domain must reach a 100 MHz domain. How?" A framework answer runs: Assumption, the clocks are asynchronous and the pulse spacing is currently unknown, so I will ask about the minimum spacing. Mechanism, direct capture loses any pulse that falls between two slow edges. Trade-off, a toggle synchroniser guarantees delivery at the cost of latency and a spacing rule; a handshake adds acknowledgement at the cost of throughput. Evidence, structural check for the synchroniser plus an assertion-based proof that every toggle produces exactly one destination strobe. Notice the answer never says "use a FIFO": the mechanism rules out overkill before the interviewer has to.
Whether you argue from mechanism or from vocabulary. Candidates who start with the mechanism can be guided to an answer they have never memorised; candidates who start with a part name stall when the question changes slightly. Saying the assumption out loud also tests whether you check intent before designing, which is exactly what signoff engineers do. For the mechanism itself, the failure modes lesson is the reference.
Because an answer that ends at a fix is an opinion, while an answer that ends at a check is a signoff argument. Evidence is simulation, a structural check, a formal proof, timing constraints, or a waveform, whichever closes that particular crossing.
Mechanism: per-bit synchronisers each settle correctly while the combination tears, so the bus needs Gray code, a handshake, or a FIFO. Naming the torn-bus mechanism first rules out the per-bit answer before it escapes.