In short
The mask set is the smallest part. A re-spin costs a fab cycle, three to six months of the whole team, the customer's launch and often the market window. Most re-spins trace back to a decision made months before tape-out that nobody could see the cost of at the time.
The costs that are easy to count
At advanced nodes a full mask set is the figure everyone quotes, and it is large. But it is a one-time purchase, and it is the one cost that finance can see on an invoice. The rest of the bill is bigger and spread across budgets that never get added up.
A re-spin means a second fab cycle, which at an advanced node is three to four months before the new silicon is back. For that whole period the design team is on the fix, the verification team is on the regression, bring-up is stalled, and firmware is working around a bug instead of shipping features. Three to six months of the whole program is a reasonable estimate; the team's loaded cost for that period is usually several times the mask set.
The costs that decide the program
The customer who was waiting for samples now has a date that moved by a quarter. In some markets that is survivable. In automotive, in a phone launch, in a data-centre refresh, a quarter is the difference between being designed in and being designed out, and the socket does not come back for a generation.
There is a quieter cost too. The next program starts late because this one's people are still on the re-spin. Slips compound across programs, which is why a company that re-spins once tends to re-spin again.
Where re-spins actually come from
Post-mortems usually blame the last visible symptom: verification was late, a corner was skipped, a waiver was wrong. Trace the decision trail back and the root cause is almost always earlier and smaller. A request for two more verification engineers deferred to a quarterly review. A coverage curve that flattened and was called on track. An IP drop that moved and was absorbed by cutting the integration regression.
Each of those was a cheap decision at the time, made by someone who could not see what it would cost at tape-out. That is the real cost driver: not the bug, but the months in which the bug was still cheap to prevent and nobody could see it coming.
The cheapest re-spin is the one that is never needed
Avoiding a re-spin is not about more reviews at the end. It is about seeing the small decisions early, with their downstream cost attached, and making them while they are still cheap. That means reading the coverage curves and the timing reports every day, connecting a supplier's email to the blocks it affects, and putting the trade-off in front of the person who owns the date.
Tymeline exists to make that the normal way a chip program runs: trouble is seen while it is small, the fix is worked out and approved before the date moves, and the program's record of what nearly went wrong is kept for the next one.