Explained
How a processor die works.
Inside a processor die
A processor die is a small rectangle of silicon covered in circuitry. From above, a die photo looks like a city map. Large repeating blocks are the CPU cores. Big, even-looking areas are cache memory. Graphics, memory controllers and input/output circuits have their own regions, usually around the edges. This model is a generic layout, not a copy of any real chip.
Cache is used so much because memory is slow compared with a core. Fetching data from the RAM sticks can take hundreds of clock cycles, so chips keep recently used data in fast on-chip SRAM. A typical design has small, very fast caches inside each core and a larger shared cache that all the cores can reach over the interconnect.
How chips are made
It starts with silicon refined to extraordinary purity and grown into a single crystal by the Czochralski process, in which a seed crystal is slowly pulled and rotated out of molten silicon. The crystal is sliced into thin wafers, polished, and taken through hundreds of steps. Photolithography projects the circuit pattern onto light-sensitive resist, and the newest machines use extreme ultraviolet light with a wavelength of 13.5 nanometres.
Transistors are built first, at the surface. Then layer after layer of copper wiring is added on top, separated by insulation and joined by vertical vias. The lowest layers are incredibly fine and connect neighbouring transistors. The top layers are thick and carry power across the whole chip. Finally the wafer is tested and cut into individual dies. Names like "3 nm" describe a manufacturing generation, not the actual size of a transistor.
The transistor
Every block on the die is made of transistors. Most modern chips use FinFETs, where the channel is a thin vertical fin of silicon and the gate wraps around it on three sides. Wrapping the gate around the channel gives it better control, so the transistor switches cleanly and leaks less current when it is off. Newer designs go further and surround the channel on all four sides.
Use Slice to cut through the stack, Explode to lift the wiring layers apart, and Zoom in Play to go from the whole die to one core and then to a single transistor.
Parts glossary
- Silicon substrate
- The die is cut from a wafer of ultra-pure crystalline silicon. Transistors are built into its top few micrometres. Fun fact: Silicon crystals for chips are grown by the Czochralski process, slowly pulling a single crystal from molten silicon.
- CPU cores
- Each core can run its own program. It contains instruction decoders, execution units, registers and private cache.
- Shared cache
- Large blocks of fast SRAM keep recently used data close to the cores, because fetching from main memory is far slower.
- Graphics
- An integrated graphics processor with many small execution units for drawing the screen.
- Memory controller
- This block talks to the RAM chips, sending addresses and moving data in and out.
- I/O
- Input and output circuits drive signals off the chip to PCIe, USB and display connections.
- Interconnect
- A ring or mesh network links the cores, cache and other blocks so they can share data.
- Metal layers
- Over a dozen layers of copper wiring sit above the transistors, finer at the bottom and thicker at the top. Each layer runs mostly in one direction.
- Solder bumps
- Tiny solder bumps on top connect the die to its package when it is flipped face down.
