: Why are computer CPUs the size they are? Wouldn’t making them bigger give way to more processing power without needing better technology?

Question

Edit : My first post to blow up. Crazy.

In: Technology

Answer 1

Check out this video by the great Grace Hopper: https://www.youtube.com/watch?v=9eyFDBPk4Yw

The speed of light sounds really really fast, until you’re dealing with the order of nanoseconds and microseconds. I’m sure everyone is familiar with GHz (gigahertz) and such being a very rough measure of processor speed. Consider a simple 3 GHz processor. This means that for every nanosecond, this processor flips a switch (called a “clock) exactly 3 times. The clock is used for synchronizing the entire processor and is almost a hard requirement (some clockless/asynchronous designs exist, but they are very complex) in modern processor design. Consider If you build a large processor where the middle of the processor is the clock, memory is on one side, and the circuitry for adding two numbers is on the opposite side as memory. The memory and addition circuit are each 10cm away from the clock. With a design like this, if you had some addition operation that you wanted to then put into memory, and the processor frequency is 3GHz, then as soon as the addition circuit received the clock signal, the clock signal would change at the source. To put this into distance:

* clock to addition: 10cm/1 clock cycle
* addition operation: typically 1 clock cycle
* addition to memory: 20cm/2 clock cycles

Basically bigger distance means more propagation delay which means much more complexity with keeping the clock rate of a processor high while not running into problems with components of the processor falling out of sync with the clock, or not being able to “stabilize” on a result within the designated clock cycle. Smaller processor size means that you can push the “wiring” of a processor faster, even if the actual components (ie, transistors) are still effectively fixed speed. For one more example, consider some circuit that adds numbers. It is 10cm in area. This means that when you feed data into one side, assuming perfect component behavior etc, that the result comes out the other side in at least 0.33 nanoseconds or more typically much more time. Now, you invent something and shrink the design of this component to fit into 1cm. Now the result comes back after just 0.033 nanoseconds. The reduction in propagation delay almost always gives a huge advantage, even with a reduction in the quality of smaller transistors that might cause more “stability delay”