Part 5 answers: How do we know the output is right?
Stimulus without checking is a demonstration. Checking decides, for every output, whether it is right: locally, with assertions at the place a rule applies, or end to end, with a scoreboard comparing outputs against a reference model.
This part covers self-checking testbenches, where to check, assertions, scoreboards, how precise a reference model must be, checking reordered and dropped outputs, end-of-test checks, checking performance, and the quiet failure every DV engineer must guard against: a checker that checks nothing.
Choose between local assertions and end-to-end scoreboards for a given rule.
Build checks that catch lost, duplicated and reordered outputs.
Prove that a checker actually works.
Answer each in your head before opening it. If two or more feel shaky, revisit Part 4: Stimulus first.
If a corner has probability p per trial, hitting it once takes about 1/p trials on average; for small p that is far beyond any regression.
The legal environment of the design: what the real neighbours may send.
Something real and rare is wrong, in the design or the testbench; the seed reproduces it, and the failure rate says roughly how rare the triggering condition is.
# | Lesson | Level | Status |
|---|---|---|---|
5.1 | F · Foundation | Ready | |
5.2 | W · Working | Ready | |
5.3 | W · Working | Ready | |
5.4 | W · Working | Ready | |
5.5 | S · Senior | Ready | |
5.6 | S · Senior | Ready | |
5.7 | W · Working | Ready | |
5.8 | S · Senior | Ready | |
5.9 | S · Senior | Ready | |
5.10 | S · Senior | Ready | |
— | all levels | Ready |
Every lesson in this part is ready. Work through them in order: each one sets up the problem the next one solves.
This course teaches the decisions behind verification. These lessons teach the tools and syntax that carry them out, so this part links to them instead of repeating them.