EMZETT.Die (Semiconductor)
In short: A single piece of silicon cut out of a wafer with a complete integrated circuit — the actual “chip” before it’s put into a package.
In more detail: After separation, a die is mounted on a carrier, connected with fine wires or solder balls, and embedded in a package — only this finished package is the CPU or memory chip you see on a motherboard.
In Depth
From wafer to die
During manufacturing, a wafer is exposed and patterned with a grid of many identical copies of the same circuit — each individual one of these copies is a die. Only after all manufacturing steps are complete (which can include hundreds of individual, precise process steps, from exposure through etching to doping) is the wafer sawn along fine dividing lines into its individual dies. Not every die on a wafer is defect-free afterwards — manufacturing defects usually occur randomly distributed, which is why a certain percentage of a wafer’s dies is sorted out as “scrap” (the so-called yield rate, i.e. the proportion of usable dies).
Partial deactivation instead of scrapping
With some chip manufacturers, slightly defective dies aren’t completely discarded, but the defective areas are deactivated via software or laser cutting, and the chip is sold as a cheaper model with fewer cores or a smaller cache — one reason some “smaller” CPU models technically use the same dies as their more expensive siblings, just with parts of them deactivated. This practice, called “binning”, maximises the economic use of every manufactured wafer, instead of simply throwing away partially defective dies.
Die size and manufacturing cost
The size of a die (die size, measured in mm²) directly affects manufacturing cost: a smaller die fits more often onto a wafer of a fixed size and statistically has less area that can be hit by a randomly distributed defect — which increases the yield rate and lowers the cost per functioning die. That’s why chip manufacturers often try to split complex chips like CPUs with many cores into several smaller “chiplets” instead of a single, huge die, which are then connected afterwards in a shared package.
From die to finished chip
After separation, a die is mounted on a carrier (substrate), electrically connected with fine gold wires or tiny solder balls, and embedded in a protective package — only this finished package is the CPU or memory chip you later see on a motherboard and insert into a socket.
Chiplet design as a modern trend
Instead of manufacturing a complex chip as a single, huge die, modern manufacturers (especially AMD with its CPUs) increasingly rely on “chiplet” architectures: several smaller, specialised dies (e.g. separate compute-core chiplets and a central I/O die for memory and PCIe connectivity) are only joined together in the package, instead of being manufactured as a single monolithic die. This considerably increases the yield rate (smaller dies statistically have fewer defects), and also allows different chiplet types to be manufactured in different, respectively optimal manufacturing processes and combined flexibly — an I/O die, for example, doesn’t need as modern, expensive manufacturing technology as a compute-core chiplet.
Types of manufacturing defects
The defects that can render a die unusable come from various sources: tiny dust particles during exposure, minimal irregularities in the silicon crystal of the wafer itself, or deviations in one of the hundreds of precise manufacturing steps. The smaller the feature size of a modern manufacturing process (today in the range of a few nanometres), the more sensitively manufacturing reacts to even the smallest disturbances — one reason why the most modern chip factories are among the most technologically demanding and expensive industrial facilities in the world.
See also: Wafer, Silicon chip, CPU