How do they design/create chip with billions of small transistor?

Question

When we talk about a chip, we often think about it having a really complicated circuits, like the Apple M2 where it has 20 billions of transistors and etc. Back in the days, chip weren’t that complicated so I could digest that they could design it. But nowadays since they’re more complicated, do they really design it when they’re very small like nanometer level?

Ps: I’m not a computer engineering student but feel free to use some technical terms if needed since I take related course to this.

In: 13

Answer 1

To start off with they only need to design one part of the circuit and then copy past that all over the place. You design a one bit ALU and then copy past it to get ten 64 bit ALU for a processor core.

But already in the 80s they came up with gate arrays. You might recognize the name as part of FPGA which was one path the technology went in. A gate array is a matrix of AND and OR gates that given any input can produce any output that have been programmed into it. This did replace read only memory for some tasks but was fast and cheap enough to replace a lot of logic circuits with this as well. A gate array is designed by a computer using a simple algorithm but you can run optimization passes on it to remove transistors and pathways that are not needed. So instead of designing a one bit ALU you just put all the possible inputs and the corresponding output into the gate array algorithm and it will spit out the design of an ALU for you.

This is when things start to get fun, because you can easily take machine code like assembly language and make each operation a gate array which allows you to write code and render it in transistors. Again you have optimizes on multiple levels that helps produce efficient hardware using fairly simple algorithms. But why stop with machine code, you can build higher order languages that compiles down to this code. And this is what people do today. They are not actually placing down each transistor but rather are coding in object oriented languages and just have compilers generate the silicon.