What is the validity of Moore’s Law? and how long does the Moore’s Law period take?
When we talk about processors (and GPUs), the well-known Moore’s Law is a recurring term when talking, especially about lithography. But do you know what it consists of and why it is so important when we talk about hardware? We explain it below.
What is Moore’s Law?
Moore’s Law was written in 1965 by engineer Gordon Moore when he was director of Fairchild Semiconductor Labs (although, as you may know, he later co-founded Intel). He was the first to observe a trend in the early days of microelectronics that would determine the strategy for all manufacturers in the industry in terms of the integration rate of transistors in integrated circuits.
Originally it said that “the number of transistors per unit area in integrated circuits will double every year”, but a few years later, in 1975, it changed its own law and increased this rate to every two years. Many people mistakenly believe that this period is 18 months because of a claim made by Intel’s CEO, David House.
Let’s put this statement in perspective. Let’s assume that the area of the integrated circuit is one square centimeter (this is an example). If 1000 transistors fit into that area today, according to Moore’s Law, in two years, we should be able to fit 2000 transistors into the same area, and in four years, we should be able to fit 4000 transistors, and so on, doubling every two years.
To put this in perspective, this wonderful video from DataGrapha shows how Moore’s Law has evolved from its inception to the present day and how processor launches have evolved accordingly.
At many moments, this law has not been met, and at other moments, technological development has even allowed it to be exceeded.
Moore’s Law is about density, not power
We must remember that Moore’s law doesn’t say anything about the performance of processors, which depends on their architecture and, therefore, on their internal organization and the way transistors are arranged to form the various elements. So Moore’s law only tells us how many transistors we can fit in a given area.
We say this because power growth slows down from one generation of CPUs to the next, repeating the mantra that Moore’s law has come to an end or will come to an end. What happens? Well, the public confuses Dennard’s scaling, which is about increasing performance, with Moore’s Law. The latter is still relevant, while the former has been completely obsolete since the introduction of 65nm. Since then processors have been scaled in different ways and according to different rules.
The bottleneck is that it’s getting more and more expensive to design new chips, and the number of manufacturers that can make chips with the most advanced nodes is getting smaller and smaller like it’s a scary game of armchair. Because of this, processors are being split into several units, some of which can be made on less advanced nodes. These are much cheaper, and certain parts of a CPU don’t need such an advanced node, especially the near-external I/O interfaces.
Why is it so important for hardware?
Logically, for this to be possible, the size of the transistors must be reduced. Otherwise, it would not be possible to fit twice as many transistors where there is no room for them since the law states that the area is always the same (although it does not have to be, it is a ratio of the number of transistors per unit area), more precisely, the unit of measurement is MTr/mm2 or millions of transistors per square millimeter.
This unit is currently used, and probably will be in the future, since we are still decades away from reaching the atom, especially since the more transistors that can be accommodated in 1 mm2, the more complicated it becomes as we get closer to it.
And this is where lithography comes in, which is also talked about a lot in the context of processors. Smaller lithography means that the transistors are smaller, so we can fit more of them into the same space, or in other words, the same number of transistors would take up less space.
This means that the 7nm process can theoretically accommodate twice as many transistors in the same space as the 14nm process. The ability to fit more transistors in the same space affects performance and energy efficiency, so a smaller lithography process means the processor is more powerful and efficient.
The reality is that nomenclature in nanometers hasn’t reflected reality for years and is more of a marketing ploy to tell us that one node is better than another. And there is no clear consensus among the various chip makers on what their nanometers mean. So it all boils down to vague numbers, and although there have been several attempts to sort things out. Right now, everyone is going their own way with their own X-nanometer nomenclature.