Chiplet Economics Console

The single most important fact in chip cost is that yield falls exponentially with die area. Double a die's size and you don't just double its cost — you push it down a yield curve where killer defects claim an ever-larger share, so cost per working chip can rise far faster than area. For the largest dies near the reticle limit, monolithic yields can be brutal — the problem chiplets were invented to solve.

The idea is disarmingly simple: instead of one enormous die, build several small ones and connect them in the package. Because each small chiplet yields dramatically better, and known-good-die testing lets you assemble only working ones, the combined product yields far higher than a monolithic die of the same total area ever could. AMD turned this into a generational advantage with Zen.

But chiplets aren't free. Connecting dies requires advanced packaging — an interposer, EMIB bridge or organic substrate — each chiplet needs die-to-die PHYs that consume area and power, and every die must be tested before assembly. Chiplets win only when the yield savings exceed all that overhead, which is exactly the crossover this console computes: for a large die on a process with meaningful defect density it wins decisively; for a small, well-yielding die it just adds cost.

Beyond the pure yield play, chiplets unlock the right node for each block. Model the cost crossover here, estimate the packaging term in the CoWoS Cost console, and ground the yield assumptions with the Yield console.