Why it matters

Toggling is not the only way to make a pulse wide enough to sample. Holding the pulse high for a fixed number of source cycles (stretching) is simpler to build and simpler to review, and for slow event sources it is completely adequate. A signoff reviewer needs to know both schemes and, more importantly, when stretching stops being acceptable.

The pattern

A stretch counter in the source domain holds the event high for S source cycles. The destination synchronises the resulting wide level through a two-flop synchroniser and edge-detects it, exactly as in the toggle scheme, producing one destination pulse per stretched event.

1
2
3
4
5
6
7

// snippet: stretch an event to S source cycles, then synchronise as a level.
// Source domain:
//   always @(posedge clk_s) begin
//     if (pulse_s) stretch <= S; else if (stretch != 0) stretch <= stretch - 1;
//     level_s <= (pulse_s || stretch != 0);
//   end
// Crossing: 2FF synchroniser of level_s into clk_d, then XOR edge detect.

When stretching is acceptable

Stretching is acceptable when the stretched width S is chosen against the destination period with margin (at least two destination periods, preferably more) and the destination clock rate is bounded: if the destination can be slowed by a power mode, a fixed S can become too short. Toggling has no such dependence, because the level holds until the next event however slow the destination is.

Destination edge detection

Both schemes share the destination side: synchronise the level, delay it one cycle, XOR. The output pulse is one destination cycle wide. A reviewer checks that the detector watches the synchronised level and that nothing downstream expects the pulse to arrive in any fixed number of source cycles: the latency is two to three destination cycles plus synchroniser settling, never a source-cycle count.

Common mistakes

Related topics

• Toggle synchronizer

• Spacing and rate limits