Wafer Cost Console

Every chip strategy decision — which node, what die size, monolithic or chiplet, which volume to commit — resolves to one number: the cost of one working chip. The arithmetic is simple; the inputs hide most of the industry's hardest trade-offs, and getting them wrong is how products miss margin or never recover their design cost.

The spine is one division: cost per good die = wafer cost ÷ net good dies, where net good dies is gross dies × yield. The subtlety is "good." You pay for the whole wafer regardless of how many work, so a $16,500 wafer holding 760 dies costs far more than $21.71/chip once yield enters — at 80% only 608 are sellable, so each costs $27.14. Forgetting to divide by yield is the most common and expensive mistake in chip costing.

On top sit three terms modern designs can't ignore. Packaging has exploded for AI silicon — CoWoS and HBM can rival the logic die. Test adds real ATE-time dollars. And NRE — one-time design and mask cost — has grown so steep at the leading edge that it reshapes which products can use which nodes. A 3nm mask can exceed $30M, and because NRE amortizes across volume, it's $30/die at a million units and thirty cents at a hundred million.

Use it as the capstone of the cost trio: Die Per Wafer supplies gross dies, Yield the working fraction, and this turns them into cost, margin and break-even — the language executives decide in.