These separate candidates who say "Gray code" from candidates who know where it stops working. Every strong answer below contains a limit.
Field | Answer |
|---|---|
Interviewer is testing | The torn-bus reflex. |
Bad answer | "Nothing, each bit is synchronised." |
Better answer | "Bits settle at different times, so the destination can mix old and new bits: a torn bus with values the source never sent." |
Strong answer (a 4) | Adds the skew mechanism (capture-window plus routing skew across bits), rules out more stages as a fix, and offers Gray, handshake, or FIFO by rate. |
Follow-ups | An impossible-value symptom; why stages do not help. |
Common trap | "Add a third flop". |
Evidence to cite | Structural multi-bit-without-protocol flag. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | Limits of a favourite answer. |
Bad answer | "Gray code solves multi-bit CDC." |
Better answer | "Consecutive values differ in one bit, so samples resolve old-or-new; it fails when the source moves several steps per sample or the range wraps non-cleanly." |
Strong answer (a 4) | Adds reconvergence with other logic as the third failure and states the enforcement (transition assertion, range discipline). |
Follow-ups | Non-power-of-two handling; Gray plus compare in one gate. |
Common trap | Treating Gray as a synchroniser rather than an encoding. |
Evidence to cite | Transition proof plus structural check. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | Definition with a sketch. |
Bad answer | "Signals meeting again." |
Better answer | "Two separately synchronised paths rejoin in one function, so skew between them can glitch the output even though each path is synchronised." |
Strong answer (a 4) | Draws it (two crossings into one gate), names the classic site (flag comparisons), and states the removal (compare in one domain against one synchronised copy). |
Follow-ups | Why tools flag it; a legal versus illegal convergence. |
Common trap | "Synchronised inputs cannot glitch". |
Evidence to cite | The structural reconvergence report. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | Report triage. |
Bad answer | "Waive it, the pointers are Gray." |
Better answer | "Find which two crossings meet at the flag, check whether the comparison mixes fresh copies from both directions." |
Strong answer (a 4) | Walks the worked shape (local pointer versus synchronised copy is legal; two synchronised copies in one gate is not), names the fix direction, and demands re-analysis after the fix. |
Follow-ups | What proof closes it; when convergence is benign. |
Common trap | Waiving by pointer encoding instead of by comparison structure. |
Evidence to cite | Re-run reconvergence report. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | Proportionality: cheap problem, cheap fix. |
Bad answer | "Async FIFO everywhere." |
Better answer | "Mux recirculation, handshake for buses, or qualified quasi-static capture depending on change rate and latency needs." |
Strong answer (a 4) | Prices all three (logic, latency, contract strength) and picks qualified capture for truly static buses with a stated stability contract. |
Follow-ups | Who enforces the stability contract; what breaks it. |
Common trap | FIFO-by-default for a 4-bit mode register. |
Evidence to cite | Stability assertion or constraint plus review. |
Taught in |
Field | Answer |
|---|---|
Interviewer is testing | Judgement, not just rules. |
Bad answer | "Never." |
Better answer | "When the converging paths are qualified so skew cannot propagate: a stable select with synchronised data, for example." |
Strong answer (a 4) | What a 4 sounds like: "Assumption: state which signals can move together and which are held. Mechanism: accept only convergence where the output is insensitive to the skew window, such as data captured under a qualified enable that is itself stable. Trade-off: accepting it saves restructuring at the cost of a proof obligation on the qualification. Evidence: a formal or assertion proof of the insensitivity plus the waiver recording exactly which convergence, with expiry on RTL change. Blanket acceptance is never the answer." |
Follow-ups | What the waiver records; which change reopens it. |
Common trap | "The tool has a waiver feature, so waive". |
Evidence to cite | Insensitivity proof plus scoped waiver. |
Taught in |
Escalating scope: the torn bus (what breaks), reconvergence (where else the same skew idea bites), and acceptable convergence (when the rule bends). An interviewer walking this arc is testing whether your rule has reasons.