Stimulus alone proves nothing. A monitor passively rebuilds transactions from DUT pins, a reference model predicts what the DUT should have done, and a scoreboard compares the two streams.
Three rules carry the whole topic. The monitor never drives a signal, the scoreboard compares observed against predicted, and end-of-test checks prove compares happened before any PASS.
Lesson | What it covers |
|---|---|
Input-only sampling, accepted-beat reconstruction, one broadcast copy per consumer. | |
FIFO compare for in-order DUTs, by-id lookup for out-of-order DUTs. | |
Field masks, first-difference dumps, drain windows, watchdogs, leftover checks. | |
Transaction-level predict functions and coverage sampled from observed traffic. |
Nothing checked anything. The PASS is hollow: a broken generator, a miswired monitor, or a disabled driver all produce zero compares. End-of-test checks must fail an empty scoreboard.