A modern processor has 100+ billion transistors. They all do one thing: they're voltage-controlled switches. The entire stack from those switches to 'your laptop runs Python' is built layer by layer.
A modern processor has 100+ billion transistors. They all do one thing: they're voltage-controlled switches. The entire stack from those switches up to "your laptop runs Python" is built layer by layer. Here's the whole climb in five steps.
A MOSFET has three terminals: source, drain, and gate. Apply enough voltage to the gate, and current flows from source to drain. Don't, and it doesn't. That's it. A switch you can flip with a voltage instead of a finger.
Modern transistors switch in roughly 10 picoseconds and the gate is a few
nanometers wide. You can fit billions on a die the size of a fingernail.
Wire two transistors in series and you get an AND gate: current only flows if both gates are high. Wire them in parallel and you get an OR gate. Add an inverter and you get NAND or NOR.
CMOS — the technology in every chip you've ever touched — pairs an n-type and a p-type transistor for every gate. This is why CMOS draws almost zero static power: in any steady state, one transistor is always off, blocking current from supply to ground.
Combine gates into a full adder: 1-bit + 1-bit + carry-in = 1-bit sum + carry-out. Five gates. Chain 64 full adders together (with the carry rippling through) and you have a 64-bit adder.
Add a multiplexer to choose between add/subtract/AND/OR/shift, and you have an arithmetic logic unit (ALU). The ALU is the part of the CPU that actually computes things. Everything else is just feeding it data.
A CPU is an ALU plus three things:
Modern CPUs add pipelining (overlap fetching and execution), out-of-order execution (reorder instructions for speed), branch prediction (guess which way an "if" will go), caches (avoid going to slow main memory), and SIMD vector units (do 4–16 operations at once). But the core idea — fetch instruction, decode, execute, repeat — is unchanged from the 1940s.
An Apple M-series or Snapdragon chip isn't just a CPU. It's a SoC — system on chip — combining CPU cores, GPU cores, a neural engine, image signal processors, video encoders, memory controllers, USB/PCIe controllers, and a security enclave, all on one die. The interconnect between them (a network-on-chip) is itself a non-trivial design problem.
This is why "a chip" in 2026 means roughly: a few square centimeters of silicon, ~50–200 billion transistors, multiple specialized accelerators, and tens of GB of stacked memory next to it. The transistor count went up. The basic switch hasn't changed.