eli5 how transistors can get so small


I was watching the new 4nm transistor size for the new Qualcomm snapdragon 8 gen 2. How does the parts of processors get so small that they’re the same size as around a few dozen atoms?

In: 7

They don’t build transistors in a CPU one at a time.

Instead they make a plate that makes a big shadow of the desired circuit, then use lenses and mirrors to shrink the shadow.

Aim the shadow at a very flat layer of light-sensitive chemical, and the non-shadow parts are weakened. So when you wash it with acid, only the light-exposed parts are washed away. With a plate that makes an extremely intricate shadow, you can make extremely intricate structures all at once.

That’s the basic idea of how millions (or even billions) of transistors, wires and other structures can be made small and cheap enough for a CPU that only costs hundreds of dollars and fits comfortably in your hand.

The components aren’t built individually, instead there’s a series of large scale “passes” of lens-shrunken light / shadow patterns and various chemical treatments. The entire structure is created at once as a single piece, including all the millions / billions of components in specific positions and the connections between them.

1) Transistors don’t have moving parts. They use electrical fields to either induce or prevent electrons from moving through them. This means that they can be scaled down extremely small, and the limits on their size are related much more to capacitance leakage issues.

2) Transistors are manufactured using a process called photo-lithography. An extremely thing layer of silicon (or other material) is laid down then parts of it is etched away by exposure to light to created the desired shape. Then a new layer is applied and the process is repeated until the transistor(s) are complete.

3) Entire chips are manufactured at once rather individual transistors being manufactured and assembled together afterwards. Even then, the chips are tiny when compared to a silicon wafer, so many chips are built at once in parallel.

I usually describe it as putting a stencil on then then blasting it with a fire-hose. They cover the wafer with a material then shine light on it through a design so that only certain areas are hit by the light (photolithography) Then they wash off those areas to make the stencil. Then it gets blasted with an element ion beam to change the properties of the material at the open locations (ion implantation). You can also use diffusion to change the material. Then you wash off the stencil and repeat from the start. This whole process can take months making all the different chips but it is the main cycle. It is very precise with how it is blasted to get the right depth and angle of ions into the material to give it the exact properties it needs. I only worked on ion implators so there may be other methods but fundamentally it’s stencil on, color in, stencil off.

Something else that other people haven’t mentioned yet is that the node name no longer correlates to the actual size of the transistors. So the transistors in a “4 nm” process are not actually 4 nm wide, it is just a marketing term. It used to be that the node name was linked to the gate size, but the actual size of the transistors has stayed constant for a few years now. Performance gains are made with other architecture changes.