What Makes Up a CPU?

Ryan James Spencer

I consider Ben Eater's series on building an 8-bit CPU from scratch to be a remarkable learning journey that anyone keen enough to better understand the hardware they build software for should watch. I liked Ben's course so much that I thought I'd write a little mini summary of it. This isn't trying to be satisfactory replacement to Ben's video course and I strongly urge you to go watch the whole thing from start to finish. For example, I won't cover the details of circuit designs such as s-r latches, d latches, d flip-flops, and j-k flip-flops the way Ben does, but I will try to cover an overall view of the CPU as best as I remember and will also mention some meta things I liked that Ben did regarding the design and assembly of the CPU as a whole.

The Core of a CPU

Modern digital electronics are like irrigation networks for electricity. CPUs are no different. Data is stored either in memory that requires power (volatile, as it will disappear when the power is lost) or not (non-volatile, such as flash memory). Electricity flows to different parts of the system which I'll call units via the bus, which simply holds some fixed size of data at a time. We might generally call this a word and in this case a word is eight bits, as this is an eight-bit CPU. Units have to be gated to talk or read from the bus to avoid cross-talk. Data can be temporarily stored in registers, of which Ben's CPU has six, not including the register he uses to display numbers from binary, which he calls "out", and the memory register on the 64 bytes of Random Access Memory, or RAM, he builds up to store data and code for programs.

All the registers on the CPU are eight-bits wide. One of these registers is the instruction register where instructions from RAM will be fetched and stored so they can be decoded into the actual bits that will control the various units by the control logic unit. These bits enable or disable certain units, as well as configure them to do different things. The act of translating the instructions into these individual bits is called decoding. Decoding instructions is merely the act of supplying the instruction as a key to a lookup table, which we will talk about more later, to get the resulting set of bits. This CPU is based on microcode which means that the instruction plus a counter built into the control logic unit make up the key that is used for translation. Not all CPUs need microcode. For example, a CPU might translate directly to one set of bits rather than several broken up into steps; this is what RISC, or Reduce Instruction Set Computer, architectures are supposed to be like. The reality is that most modern RISC processors most likely include a bit of microcode.

The other two registers are for the Arithmetic Logic Unit or ALU, which only supports addition and subtraction in Ben's case. The registers are simply named "A" and "B". Values stored in "A" and "B" are added or subtracted together and stored back into the "A" register. Ben later augments the ALU with flags to detect overflow and if an operation resulted in a zero so he can install conditional instructions, which is stored in a "flags" register. As part of the change, he also includes the flags as part of the key in the instruction decoding phase.

From a pedantic point of view, the RAM and the "out" register plus its associated display aren't actually part of the core of a working CPU, but they help making writing programs easier by giving us a place to put data and code as well as visualizing the results of our calculations. The output display is actually rigged up using four seven-segment displays. With the instruction decoding phase and the display, Ben uses Electrically Erasable Programmable Read-only Memory, or EEPROMs, as the look up tables we mentioned before; on one end he can feed some bit pattern and out the other end receive a result. A good mental model is that with an EEPROM we 'select' some value in the memory given some provided key, even though this typically called an "address". For the "out" display he can provide the binary value and receive a bit pattern that is just right for showing on one of the four seven segment displays. The display actually display one-by-one but display so rapidly ("refresh") that the shift isn't visible to the eyes.

Behind all of this is a pulse that goes "high", i.e., it emits a five volt signal, at regular intervals, called the clock. Clocks drive a CPU by breaking up actions into discrete steps. When the clock pulse is high, we might call that tik and when the clock pulse is low we might call that tok. Going tik, then tok, is called a cycle. Imagine some crank that shows single images for a movie on a screen for every cycle. If you move the crank fast enough and the images look to be close enough together in time, you wind up with what looks like a fluid image, but you can also single-step the crank or move it more slowly and notice all individual images making up the movie.

A program counter stores the next instruction that will execute on the CPU.You can either increment the counter by one or change it to some absolute value, which is how jump instructions, and therefore conditionals, work. All instructions have two initial steps which involve pulling a value from the program counter into the memory register on the RAM and then pulling the addressed value into the instruction register. This way the program can startup on any instruction but wind up back to where the program counter is pointing. The memory register is needed to ensure that whatever address is chosen to output or read into the RAM won't change simply because the rest of the system has moved on, e.g., the bus value has changed.

Of notable mention is the reset switch he builds into the system, as well as an ability to stop the clock to emulate a HALT instruction. These make starting from the beginning of a program and stopping the program at a particular point easier than simply having the program "spin" at the end.

The Design and Assembly

There's a lot of meta things Ben does I think are applicable to a wide range of projects.

If it's not obvious, I like that Ben breaks the CPU into separate parts or units. This way Ben can focus on one thing at a time, building up earlier tools and units for re-use later, freeing up his ability to think on different problems without having to hold the whole of the CPUs design in his head at once.

He tends to build out units using bare circuits and simple transistors, first, moving onto integrated circuits, or ICs, later when things become tedious. Ben actively takes the time to build out initial circuits to demonstrate some essential electrical patterns. The way Ben leads up to d and j-k flip-flops based on his prior building and use of d and s-r latches informs a lot of the circuitry across the whole of the computer. Knowing how data gets "trapped", forming the basis of memory, and how this plays into how memory is "gated" on and off the bus, can help strip away whatever magic you might have left about what is going on under the hood. He does have some supplemental videos on how semiconductors, transistors, and diodes work that can help fill in fundamentals beyond what the primary playlist covers.

On a number of units Ben goes out of his way to make driving the unit by hand, such as allowing one to single-step the clock or program the RAM with dip-switches. Doing this allows him to rapidly smoke test or make quick tweaks on a given unit.

I like that he builds the EEPROM from scratch because it helps keep the attitude that "there is no magic".

At several points he uses a multimeter and oscilloscope to visualize what is going on with the circuits. This really drives in the metaphor of the electrical current analogy and that the transistors and complicated logic are the gates, feeding data into and out of particular pools.

In the core explanation, I noted that the RAM and LED display weren't crucial parts of a CPU, but it is good that Ben included them because they helped facilitate the process of building up the CPU, testing it, and making it useful beyond simply being a doormat that blinks.

Last, but not least, it's a small thing but I like that Ben actively goes out of his way to build a number of programs for the CPU to run. This verifies the CPU is, indeed, working as intended, and helps highlight a bit of a discussion around "what is Turing completeness?", or, more simply, how can a computer be deemed of enough general-use such that it can be used to compute anything we would generally hope for it to compute?


Ben's courses are great. He covers a number of other topics, such as building up a basic computer using a 6502 chip, talking about networking internals, checking for reliability on a transmitted message, and he also has a pair of videos building up a video card from scratch.

The brilliance of all of this is that there isn't any magic behind modern computers. There is a lot of complexity through many, many layers of abstractions, responsibilities, code, data, hardware, and so on, but the overall view of things need not be that complex. Ben's CPU isn't the fastest or feature-complete, but it does give a mental model for a basis of a CPU.

Modern CPUs are a beast, but you don't need to know everything to feel like you know enough of a subject to be dangerous. Many modern-day, bootcamp-trained software developers have learned specific patterns that might aide them with particular stacks of technology, but I have helped mentor a number of these people who wish they could better understand just what is going on under the hood. Holding the idea that there isn't any magic and you can continually unravel the layers and components to gain a deeper understanding is possible and it is thoroughly rewarding. You don't need to know everything at once, either! Understanding how a CPU works and some fundamentals of electronics both at a rudimentary level will help guide you with other understanding other things, and with time you can refine that understanding.