Why it matters

An undersized FIFO is a functional bug with a long fuse: it passes every short test and overflows on the first long burst in the system. Depth is therefore a signoff calculation, written in the design document with its inputs named, not a power-of-two guess. The three inputs are the burst length, the rate mismatch during the burst, and the synchroniser latency hiding recent reads and writes.

The sizing argument

During a burst the writer outruns the reader and occupancy grows by the rate difference accumulated over the burst. On top of that, each side works from a stale view of the other pointer, so a few extra entries cover the in-flight synchroniser delay. Almost-flag margins for pipeline reaction time add further entries where the producer cannot stop instantly.

Symbolic worked example (illustrative)

All numbers below are illustrative: a burst of B = 120 words, write rate twice the read rate, and a synchroniser and flag latency worth about 2 entries each way.

illustrative depth sizing, recomputed two ways (B = 120 words, fw:fr = 2:1)
method 1, rate difference:
  accumulated = B x (1 - fr / fw) = 120 x (1 - 1/2) = 60 words
  synchroniser margin = 4 entries (about 2 entries per direction for the synchroniser and flag latency)
  total = 60 + 4 = 64 -> next power of two: 64
method 2, time based (same inputs, counting from the burst duration):
  burst lasts 120 write cycles = 60 read cycles; reads remove 60 words
  net stored = 120 - 60 = 60 words; plus 4 margin = 64 -> depth 64
both countings agree: 64 entries

What the checks look for

Reviewers ask for the sizing calculation and its three inputs, then check the testbench covers the worst case: a full-length burst at the worst rate ratio with the consumer stalled at the start. A FIFO that is only ever tested half-full is not signed off.

Common mistakes

Related topics

• Full and empty

• Latency and throughput