In short
A phase gate is only useful if it proves something. Most gate reviews prove that a slide deck was made. The gates below are listed with the evidence each one should require from the tools, which is the only way a gate can stop a slip instead of recording one.
The gates, and what each one has to prove
Names differ between companies; the sequence does not. Each gate below is a decision to spend the next, larger block of money, and it should be made on evidence that the previous block produced what it promised.
- Concept gate: the market requirement, the target node and package, a first cost model, and the IP make-or-buy list. Proves the chip is worth starting.
- Architecture freeze: a frozen feature list, block owners, power and area budgets, and dated IP commitments. Proves the chip is defined well enough that changing it later is a decision, not a drift.
- RTL freeze: every block's RTL under change control, coverage plans with targets, and the ECO process in force. Proves that physical design can start without being undone.
- Physical design release: timing closed at agreed corners, power within budget, DFT inserted. Proves sign-off can be reached in the weeks planned for it.
- Tape-out: sign-off evidence per block, the waiver list with owners, and the downstream dates checked. Proves the chip can go to the fab.
- Samples: silicon back, bring-up firmware running, the test program debugged. Proves characterisation can begin.
- Qualification and production release: reliability lots complete, customer criteria met, yield at target. Proves the chip can ship in volume.
Why gates reviewed from slides let slips through
A gate review built from status slides measures the team's confidence, not the chip's state. Confidence is high at every gate; that is what teams are for. The slip is visible in the data for weeks before the slide says anything, in a coverage curve that flattened, a corner that keeps failing, a drop date that quietly moved.
The second failure is that gates are events, and slips happen between them. A gate every quarter means a problem can grow for eleven weeks before anyone with authority sees it. The fix is not more gates; it is reading the gate's evidence continuously, so that the gate itself has no surprises.
Running gates from evidence
Each gate's criteria map to things the tools already know: coverage databases, timing reports, DRC results, Jira, the foundry's and vendors' mail. A gate run from evidence opens those directly, with the owner of each item paged ahead of the review for anything not closed.
That is how Tymeline treats a gate: as a set of commitments held to a date, checked from the tools every day, with trouble worked out and put in front of the gate's owner while there is still time to fix it. The review becomes a decision meeting, and the decision stays with a person.