Moore’s Law is one of the fundamental rules of thumb in computing; it states that every 18 months the performance of a CPU roughly doubles. Amazingly, that rule of thumb has held true for over 40 years. Every five or so years folks will tell you how Moore’s Law is dying or dead.
While technologies play a limiting role, new ideas and innovations continue to drive Moore’s law forward now and into the future. Just a decade or so ago, Moore’s law was reaching its limit. As more transitors could be packed onto a single die, the CPU was able to spend more chip space on larger caches, better branch predictors, more parallelized ALUs, instruction level parallelism, et cetera. Chip producers increased clock rates as high as 4 GHz. Then, the single core design simply became unsustainable.
You see, more transistors being used for all those things means more power was required. Not only that: higher clock rates drastically increased power usage – cubically to be specific. Increasing the clock rate means all of the synchronous components use linearly more power. However, without also increasing the voltage, the internal signals couldn’t travel quickly enough to prevent glitches (the speed of the current in a wire is proportional to its voltage). Since voltage is quadrically proportional to power, increasing clock rate thus cubically increases power usage. Obviously nobody wants to spend a fortune powering their PC, but besides that, more power usage requires better cooling to prevent the chip from burning up. So, increasing clock rate simply became infeasible…
…but Moore’s law did not die. Instead of increasing the clock rate, chip manufacturers began to realize that they could simply increase the number of physical processors on a die. Dual and quad-core processors began to hit the market. Performance continued to double ever 18 months. The performance was a bit more difficult for programmers to take advantage of, but it was there.
Fast forward a few more years. The distance between transitors has shrunk to a tiny 10nm. The size of transistors is rapidly approaching the atomic limit on their size – the point at which they will literally only be large enough to allow a single electron to pass through. In perhaps another 10 or 20 years we will reach that limit. Chip manufacturers are already preparing for that moment. They plan to boost future performance by moving away from the idea of a general purpose processor and more towards a specialized purpose processor.
Using GPUs for heavily parallelized programming absolutely exploded. Desktop CPUs began putting special signal processing, encryption, and more special purpose blocks of hardware on the chip. The rapidly approaching next step is for the chip makers to add FPGAs to the chip.
The means of achieving performance changes, but thanks to continued technological innovations, Moore’s law will continue into the foreseeable future.