A tape-out review is scheduled for an hour. The first forty minutes are slides of regression pass rates and coverage charts; the last twenty are spent on a question nobody prepared for: "what about the low-power mode the main customer uses?" The meeting ends without a decision and is rescheduled for next week, a week the schedule did not have. The review was a presentation, not a decision process.
How do you run a tape-out review?
The obvious review is a status presentation: show the good numbers, take questions, ask for approval. Status presentations hide open items in summary charts, leave reviewers to find gaps by luck, and end in deferral when an unprepared question arrives. A review should be structured around the agreed criteria and the open items, with the material sent beforehand.
Area | Question |
|---|---|
Plan | Which features have no evidence, or only one kind? Was the plan reviewed by architects? |
Coverage | Which holes remain, by cause? What was excluded, and who approved it? |
Checks | How do we know the checks detect bugs (planted bugs, mutations)? |
Formal | Which properties, which assumptions, which depth, full or bounded? |
Bugs | What does the discovery curve look like with effort beside it? What is open, triaged how? |
Change | What changed in the last month, and how was each change re-verified? |
Use | Which customer configurations and modes were verified, and which were not? |
Risk | What is the residual risk, and what goes first in silicon validation? |
Outcome | When |
|---|---|
Go | All criteria met or waived with owners; residual risk accepted |
Go with conditions | Small items to close before the final database, or follow-ups in silicon validation, each with an owner and date |
Not yet | A criterion unmet without an acceptable waiver, or a risk nobody will own |
Whatever the outcome, record it: the decision, the criteria status, the waivers accepted, the conditions with owners, and the residual risk. The record is what silicon validation and the next project start from.
Practice | Purpose |
|---|---|
Review packages sent days in advance | Questions prepared, not improvised |
Reviewers independent of the team being reviewed | Fresh eyes on assumptions and gaps |
Checklist-driven agendas from the sign-off criteria | No criterion skipped, no time lost on good news |
Decisions and conditions tracked to closure | Conditions actually closed before the final database |
The opening story (illustrative) is the classic failure: a status presentation with no package, no criteria walk and no open-items list. The same project's second review, run from a package sent two days earlier, reached a decision in forty minutes: go, with two conditions and the low-power mode first in silicon validation.
So reviewers can prepare questions about gaps and assumptions, rather than improvising them, and the meeting can reach a decision.
Each condition has an owner, a date and a check that it was closed before the final database or in silicon validation.
A tape-out review is a decision process, not a status presentation.
Prepare a package, walk the criteria, review open items and waivers, invite hard questions.
End with a recorded decision: go, go with conditions, or not yet.