How do computer engineers design CPU cores and logic?

[SPARC]

I understand the physical process of layering photoresists on silicon substrates, shining a UV light on what you want to get rid of, and then rinse and repeat until you have a CPU die.

 

But what I don’t understand is how engineers come up with the complicated and intricate designs of processor cores, cache, memory channels and PCIe lanes, etc.

 

It feels like since these things are many miles long I feel like it would take a lot of space on a computer and a genius to take a glance at it and say “oh, this does that!”

 

Whenever I look up “CPU die schematic” I only get gross oversimplifications that show blocks and lines, not all the individual logic gates and mosfets. The closest I can get to this is the Intel 4004 die schematic. Not even a Motorola 68000 schematic.

 

 

[AbydosOne]

20 minutes ago, Ionosphere said:

But what I don’t understand is how engineers come up with the complicated and intricate designs of processor cores, cache, memory channels and PCIe lanes, etc.

It's a lot of boot-strapping. First, you design an NAND/NOR gate, then you design a carry-lookahead adder or a memory cell or ... from those, then ... from those, then (after a lot of iterations and optimizations) you have a CPU from those. Every CPU component can be broken down into simpler and simpler parts that can be reused in other parts. It's not far from how complicated code is just a thoughtful combination of many simple bits of code.

 

The top level designer isn't thinking about the specifics of silicon-level transistors, and the electrodynamics engineer doesn't care about the instruction each transistor will be used for as long as his substrate is up to par.

 

20 minutes ago, Ionosphere said:

It feels like since these things are many miles long I feel like it would take a lot of space on a computer and a genius to take a glance at it and say “oh, this does that!”

There are existing patterns in the implementation that allow for better understanding, but I don't think anyone can look at bare silicon and tell you how everything works beyond the simplest divisions (cache, CPU, GPU, etc). Most CPU implementation is from the top down (working with existing designs), not the bottom up (starting from scratch every time).

 

20 minutes ago, Ionosphere said:

Whenever I look up “CPU die schematic” I only get gross oversimplifications that show blocks and lines, not all the individual logic gates and mosfets. The closest I can get to this is the Intel 4004 die schematic. Not even a Motorola 68000 schematic.

Probably because even a single logic gate or memory cell can be a dozen transistors when implemented. Even a few dozen gates would rapidly go beyond simultaneous human perception (that's why logic simulators exist), let alone billions.

[akio123008]

You also have to realise that people use software tools/languages for this stuff as well. 

 

For example in VHDL or something similar, you can type in something like "z = x AND y" and then the software internally designs a component with ports x,y and z which performs an AND operation on x and y with the output being z.

 

[Mister Woof]

I just trust that the magic elves got it right