A little about the "law of Moore"

According to Moore ’s Law , the number of transistors placed on an integrated circuit chip should double every 24 months, and their cost should remain at the same level. But almost immediately after this hypothesis was put forward, talk of the “death” of this law began — after all, in the real world nothing can grow endlessly (even if growth is exponential).



As Moore himself stated, his “law” assumed not only a reduction in the size of transistors, but also a drop in their value. As the Economist notes , a few years ago there was a tendency that could lead to the "end of Moore's law." When the 28-nm transistors appeared, manufacturers noted the fact that the cost of creating them is increasing. The construction of semiconductor manufacturing plants using new technologies requires more than $ 6 billion. In other words: a further reduction in the size of transistors is possible, but their cost will grow.

')

At the time of the emergence of Moore’s law, microprocessor design did indeed occur in accordance with this law. But with the reduction in the size of transistors, the technology for their manufacture becomes more and more complex and sophisticated. So, in the early 2000s, it also seemed that this progression was coming to an end, but various technical tools being developed at that time supported the “life” of Moore's law. Was presented "stretched" silicon, which provides the production of 90-nm transistors.



For the 45nm transistors, new materials were invented to increase the capacitance (gate) of each transistor. A three-dimensional transistors ( tri-gate transistors ), manufactured using 22 nm technology, only supported Moore's law.



But these technologies can not develop forever. The photolithography used to produce chips works at the limit of its capabilities: light with a wavelength of 193 nm is used to create chips with elements as small as 14 nm. Too long a wavelength of light can be reduced, but this leads to even more complex and costly production. So, hopes were pinned on an “extreme” ultraviolet with a wavelength of just 13.5 nm, but it turned out that it was rather difficult to use in production.



Even with ultraviolet light, it is not known how much transistors can be reduced. After all, at a size of 2 nm, their width will be only 10 atoms, and then they are unlikely to work reliably. In addition, in this case, engineers will face problems of energy consumption and cooling: the tighter the transistors are placed, the more difficult it is to deliver energy to them and take it away from them.

Do not forget about the cost factor - the end of Moore’s law can be put not by physics, but by economics.

Large companies like Intel and AMD will invest billions of dollars as long as they are confident that the profits will cover the initial investment. However, every time transistors become smaller, their production becomes more expensive.



Thus, in the near future there may come a time when the "giants" will not finance the development of smaller transistors, since it simply will not be profitable. And if someone like Intel stops chasing the reduction of transistors, other manufacturers will follow his example, leading to the “death” of Moore’s law.



The economic factor already affects the production of microprocessors. So, Intel planned to switch to 10-nm transistors in 2016 with Cannonlake processors - a smaller version of the now sold 14-nm Skylake. But in July 2015, she changed her plans. A new generation of processors - Kaby Lake - will be released in the third quarter of 2016 using 14-nm technology. Plans for Cannonlake and 10-nm remain, but the release of these processors is expected not earlier than the second half of 2017.



All this is complicated by the fact that new transistors are harder to use. In the 1980s and 1990s, the advantages of additional transistors were obvious - each subsequent generation of processors was faster, and computers of that time received a significant increase in power only through the improvement of processors. But such a simple increase in productivity began to decline in the early 2000s. The frequency of the processors was severely limited due to heating, and the power of one processor core grew relatively slowly. Instead, manufacturers began to make processors with multiple cores: theoretically, this increases the power of the processor.



All these difficulties suggest that the production plans of companies will soon cease to rely on Moore's law. Nature publisher writes that the new ITRS plan will also take a different approach. Now chip makers are targeting the rapidly growing mobile device market. They need chips not only with a logic module (transistors) and cache memory, but also with RAM modules, power management, analog components for GPS, cellular communication, Wi-Fi and so on. All these modules are now being created in different technological processes, and for chip manufacturers, integration of old and development of such composite microprocessor technologies is more relevant than the next doubling of the number of logical transistors.



Still, new technologies can still give Moore's chance. Current microcircuit technology (CMOS), which uses silicon, can be replaced by something else, including Intel 7nm transistors. More promising (compared to silicon) materials: indium antimonide (InSb) and gallium-indium arsenide (InGaAs), as well as (possibly) carbon, both in the form of nanotubes and in the form of graphene.



If you make any predictions, then in the future, Moore's law will not disappear completely. The use of non-standard materials can bring this scale back into operation. A large enough breakthrough may awaken demand for processors, which are simply faster, and not less or less.



But now, when we cannot say for sure about the upcoming breakthrough innovations, the term of Moore's law (in its “last edition”) is coming to an end. And there is nothing catastrophic about this: chip makers are opening up new markets such as mobile devices and data center machines. Therefore, even if Moore's law “dies”, it will not have a strong effect on the life of the man in the street, nor on the life of the producer.



Materials on the topic from our blog on Habré:

• Is Moore's Law Dead: What About the Clouds?
• Is Moore's Law Alive: Intel vs. IBM
• How Moore's Law Affects the Evolution of Storage Technology

PS Additional materials on the development of a virtual infrastructure provider 1cloud :

• How to create a virtual infrastructure provider: Experience 1cloud
• Client-oriented "cloud": Experience 1cloud
• Photo tour of the "cloud" of 1cloud
• A little about data storage and experience 1cloud
• UI Optimization: Experience 1cloud