BeCuriouzLearn something new every day
Why chipmakers stopped building one giant chip
💡 Chiplets beat giant monolithic chips because splitting a processor into smaller dies means a manufacturing defect wastes only a tiny piece instead of the whole chip.
For decades, a processor was a single slab of silicon — a monolithic die. But as chips grew larger, a brutal economic problem appeared: silicon wafers always have random microscopic defects, and a defect anywhere on a big die can ruin the whole thing. Double the die size and you don't just lose more area to each defect — you lose a higher fraction of dies entirely. This is the yield problem, and it's why the biggest chips became wildly expensive to manufacture.
The fix, popularized by AMD's Ryzen and EPYC processors, is the chiplet: instead of one huge die, you build several smaller dies and stitch them together inside one package using a high-speed interconnect. Because each chiplet is small, a defect only ruins that tiny piece — the rest of the wafer is still usable — so overall yield climbs and cost falls.
Chiplets bring another win: mixing and matching. A single package can combine chiplets built on different, cost-optimized manufacturing processes — say, dense compute cores on a cutting-edge node and input/output circuitry on a cheaper, mature node. The tradeoff is complexity: moving data between chiplets is slower and more power-hungry than staying on one die, so the interconnect and advanced packaging become the hard engineering problem.