Wafer yield
Also known as: die yield, chip yield, manufacturing yield, semiconductor yield, binning
Wafer yield is the percentage of chips on a silicon wafer that come out working after manufacturing.
Every wafer contains defects, so some chips are always lost. While yield sounds like an internal factory metric, it's one of the biggest drivers of chip prices, and it's why huge chips and brand-new manufacturing processes cost so much.
Wafer yield, explained at three levels
At home
Imagine baking a big sheet of cookies where a few random crumbs of burnt sugar ruin any cookie they land on. The more cookies that come out fine, the higher your 'yield.' Chipmakers face the same problem: tiny defects ruin some chips on every wafer.
At work
Yield affects how much chips cost. If only half the chips on a wafer work, each good chip has to cover the cost of the bad ones. Higher yields mean cheaper chips and more supply.
At home
Bigger chips are more likely to contain a defect, so yields drop as chip size grows. That's one reason the largest AI chips are so expensive. It's also why chipmakers sell slightly flawed chips as lower-end models, with some parts disabled, a practice called binning.
At work
Yield is typically expressed as the share of good dies per wafer, often modelled from defect density (defects per square centimetre) and die area. New process nodes start with higher defect densities, so early production is more expensive. Product managers often plan launches around expected yield curves.
At home
A simple model says yield falls exponentially with die area times defect density. Doubling a chip's area can cut yield far more than in half if defects are common. In my opinion, that maths is why chiplets took off: splitting a big design into smaller pieces keeps more silicon usable.
At work
Yield includes several layers: line yield (wafers surviving the process), die yield (working dies per wafer), and packaging or test yield. Foundries treat yield data as highly confidential, and contract pricing can depend on yield commitments. Design for manufacturability, redundancy (such as spare memory cells), and binning strategies all improve effective yield.
Real-world examples
At home
Aisha noticed two versions of the same processor, one with fewer active cores at a lower price. She learned they came from the same production line, and the cheaper version used chips with a defect that disabled a few cores.
At work
Jerome is a product manager at a chip startup. His first design was large and expensive to make because yields were low on the new process. For the next version, he split the design into two smaller chiplets, which improved yield and cut cost per working unit.
Common confusions
- A chip factory either works or it doesn't. — Every wafer has some defects. The question is how many chips survive, and that percentage keeps improving as a process matures.
- Flawed chips are thrown away. — Many are sold as lower-end products with defective parts disabled, a practice called binning.
- Bigger chips are just proportionally more expensive. — Yield falls faster than size grows, so very large chips can cost much more than their area alone suggests.
For analysts
When you see a chip price, think yield. I'd factor in that first-generation products on a new process often cost more and ship in smaller quantities until yields mature, which affects timing for hardware purchases.
History & origin
Yield has mattered since the first integrated circuits in the 1960s, when early production often had very low yields. Engineers developed statistical models, such as Poisson and Murphy yield models, to predict how defect density and chip size affect output. The industry moved from 200-millimetre to 300-millimetre wafers around the early 2000s, more than doubling the area per wafer. Clean rooms, automation, and defect inspection improved yields steadily, but each new process node starts with lower yields that must be 'ramped' over months. Reports on yields at new nodes, such as early 3-nanometre processes, became closely watched industry news in the 2020s. In my view, yield is the hidden variable behind almost every chip business decision, from pricing to product design.