Processors and marketing in computer technology
Moore's Law Problem. Microprocessor era sunset when there are no alternatives.
Electronics Magazine in 1965 published an article now widely known about the integration of components of integrated systems, authored by Gordon Moore, who later founded Intel Corporation. At one time this was a sensation, especially in those days when the computer seemed unbelievable even to educated people. Moore conducted an analysis of the development of computer technology over the past 5 years and made a forecast for the next 10. Gordon Moore has literally set the pace of technology development, and processor developers have followed it for four decades. This law is not a scientific, physical or any law as such, it is only the observation of one person over the development of one company, which suddenly began to work. It is worth understanding why?
Following this law, it is very easy to predict the development of it technologies and the companies that these technologies provide. It became clear that this is exactly what is needed for companies: the consumer expects a productivity increase of twice every 2 years, and the manufacturer does not need to talk about it, the consumer expects this anyway. You can also say that this law regulates the pace of development. You can make a breakthrough, very modern technology, but after a long time it will be impossible to release something more technological. Generational update is very important for the manufacturer, the advantages of the new should be obvious, the price is usually lower, and productivity is higher. Except for fairly cheap solutions. It is important to sell a large number of not too modern solutions, and only in this way it will bring income, updating them is almost pointless. This was obvious even in those days, and intel adopted this law, began to follow it, and there the whole industry pulled itself up. It is worth saying that such laws are not too good. Their problem is the very small flexibility of the idea of increasing power while reducing size. In the pursuit of power, we are constantly confronted with a wall of limitations of problems that we need to solve here and now, or the company will not be able to release a new generation of processors. Some restrictions were very powerful, and it is a great merit of those people who made the architectures of these processors that they managed to circumvent these restrictions. This drove the whole industry, untimely presentation of technologies in their not very high-quality form could pose problems in the future. At other times, technology allowed for a real breakthrough,
There are several basic limitations on the development path of processors:
-technological problems of printing transistors of very small sizes.
Diffusion against ion implantation.
Two methods of printing semiconductors that are fundamentally different from each other, on which depends how small a technical process we can print. If you do not go into complicated terminology, then diffusion is described as follows: during thermal diffusion, the maximum concentration of an impurity is always on the surface and monotonically decreases with depth i.e. using some factors (temperature, concentration), we can control the diffusion rate, but the process inevitably spreads in all directions (spreads) increasing the maximum possible process technology. But it was replaced by a more modern doping method: during ion implantation, dopant atoms ionize in a strong electric field and irradiate the surface of the plate with an oxide mask prepared in advance, which allows us to control the process in all directions.
The wavelength of light propagation obviously depends on its frequency. A wave with a frequency of 4000 MHz has a length of 7.5 cm. Considering the fact that the wave in the processor does not propagate linearly, the maximum size of the silicon crystal of the processor is even smaller, exactly the one that we used to see, and you can’t make it bigger.
With increasing frequency, the number of transistors and some other factors, heating increases in direct proportion. It should always be within reasonable limits; no one will want to buy a processor that will have to be cooled using liquid nitrogen. Consumption decreases with a decrease in the manufacturing process, because the light travels a shorter distance and leads to less heat, and the processor heats up unevenly, but to efficiently cool it, you need to distribute the temperature.
The progressive increase in consumption from the processor frequency in this case imposes a very strict framework for the maximum processor frequency, once upon a time it was not obvious, everyone imagined the processor of the future with a frequency ten times higher than what we see now. Relatively speaking, increasing the processor frequency by 2 times gives us 2 times the second power consumption and heat dissipation. Nevertheless, even without increasing the clock frequency at the pace that was in the last century, Moore's law in a sense worked, and the processors really became more powerful, in other ways.
<
The minimum dimension of the semiconductor gate of the transistor is limited not only by technologies that achieve the smallest possible element size, but also by the size of the silicon crystal lattice. Its dimension is 0.5 nanometers, but definitely the element cannot be the same size as the crystal lattice and close to it. We can conclude that the minimum possible shutter size can be about 1 nanometer, but this makes no sense, since shutters so close together cannot work together with others, which will cause malfunctions. It is difficult to predict which technologies will be used in the future, but, say, a shutter size of 4 -5 nanometers can be considered limiting, but only for silicon.
Moore's Law and the modern processor market.
The processor market has long been in stagnation. Not only is the development of new technologies slowing down, these technologies also do not bring an obvious increase in productivity. Nevertheless, we do not notice any problems, the new processors are still more powerful and better than the old ones. When core performance reaches its peak, multicore and multithreading come to the rescue. The strategy of popularizing multicore systems is a rather promising development strategy. It is adhered to by AMD. The promotion of multicore is necessary because far from all programs and applications support a large number of cores. Some even need one high-performance stream, which will be weaker in performance on a multi-core processor than on a processor with fewer than 8 threads (with a similar frequency and process technology). But you need to change that. A multi-core processor is much more multi-tasking when the program uses quite a few processor resources, but at the same time takes up a thread. A multi-core processor will obviously do better and be less overloaded. We can only wait.
Electronics Magazine in 1965 published an article now widely known about the integration of components of integrated systems, authored by Gordon Moore, who later founded Intel Corporation. At one time this was a sensation, especially in those days when the computer seemed unbelievable even to educated people. Moore conducted an analysis of the development of computer technology over the past 5 years and made a forecast for the next 10. Gordon Moore has literally set the pace of technology development, and processor developers have followed it for four decades. This law is not a scientific, physical or any law as such, it is only the observation of one person over the development of one company, which suddenly began to work. It is worth understanding why?
Following this law, it is very easy to predict the development of it technologies and the companies that these technologies provide. It became clear that this is exactly what is needed for companies: the consumer expects a productivity increase of twice every 2 years, and the manufacturer does not need to talk about it, the consumer expects this anyway. You can also say that this law regulates the pace of development. You can make a breakthrough, very modern technology, but after a long time it will be impossible to release something more technological. Generational update is very important for the manufacturer, the advantages of the new should be obvious, the price is usually lower, and productivity is higher. Except for fairly cheap solutions. It is important to sell a large number of not too modern solutions, and only in this way it will bring income, updating them is almost pointless. This was obvious even in those days, and intel adopted this law, began to follow it, and there the whole industry pulled itself up. It is worth saying that such laws are not too good. Their problem is the very small flexibility of the idea of increasing power while reducing size. In the pursuit of power, we are constantly confronted with a wall of limitations of problems that we need to solve here and now, or the company will not be able to release a new generation of processors. Some restrictions were very powerful, and it is a great merit of those people who made the architectures of these processors that they managed to circumvent these restrictions. This drove the whole industry, untimely presentation of technologies in their not very high-quality form could pose problems in the future. At other times, technology allowed for a real breakthrough,
There are several basic limitations on the development path of processors:
-technological problems of printing transistors of very small sizes.
Diffusion against ion implantation.
Two methods of printing semiconductors that are fundamentally different from each other, on which depends how small a technical process we can print. If you do not go into complicated terminology, then diffusion is described as follows: during thermal diffusion, the maximum concentration of an impurity is always on the surface and monotonically decreases with depth i.e. using some factors (temperature, concentration), we can control the diffusion rate, but the process inevitably spreads in all directions (spreads) increasing the maximum possible process technology. But it was replaced by a more modern doping method: during ion implantation, dopant atoms ionize in a strong electric field and irradiate the surface of the plate with an oxide mask prepared in advance, which allows us to control the process in all directions.
- crystal dimension constraints
The wavelength of light propagation obviously depends on its frequency. A wave with a frequency of 4000 MHz has a length of 7.5 cm. Considering the fact that the wave in the processor does not propagate linearly, the maximum size of the silicon crystal of the processor is even smaller, exactly the one that we used to see, and you can’t make it bigger.
- the problem of high processor consumption
With increasing frequency, the number of transistors and some other factors, heating increases in direct proportion. It should always be within reasonable limits; no one will want to buy a processor that will have to be cooled using liquid nitrogen. Consumption decreases with a decrease in the manufacturing process, because the light travels a shorter distance and leads to less heat, and the processor heats up unevenly, but to efficiently cool it, you need to distribute the temperature.
The progressive increase in consumption from the processor frequency in this case imposes a very strict framework for the maximum processor frequency, once upon a time it was not obvious, everyone imagined the processor of the future with a frequency ten times higher than what we see now. Relatively speaking, increasing the processor frequency by 2 times gives us 2 times the second power consumption and heat dissipation. Nevertheless, even without increasing the clock frequency at the pace that was in the last century, Moore's law in a sense worked, and the processors really became more powerful, in other ways.
- size limitations of the silicon lattice
<
The minimum dimension of the semiconductor gate of the transistor is limited not only by technologies that achieve the smallest possible element size, but also by the size of the silicon crystal lattice. Its dimension is 0.5 nanometers, but definitely the element cannot be the same size as the crystal lattice and close to it. We can conclude that the minimum possible shutter size can be about 1 nanometer, but this makes no sense, since shutters so close together cannot work together with others, which will cause malfunctions. It is difficult to predict which technologies will be used in the future, but, say, a shutter size of 4 -5 nanometers can be considered limiting, but only for silicon.
Moore's Law and the modern processor market.
The processor market has long been in stagnation. Not only is the development of new technologies slowing down, these technologies also do not bring an obvious increase in productivity. Nevertheless, we do not notice any problems, the new processors are still more powerful and better than the old ones. When core performance reaches its peak, multicore and multithreading come to the rescue. The strategy of popularizing multicore systems is a rather promising development strategy. It is adhered to by AMD. The promotion of multicore is necessary because far from all programs and applications support a large number of cores. Some even need one high-performance stream, which will be weaker in performance on a multi-core processor than on a processor with fewer than 8 threads (with a similar frequency and process technology). But you need to change that. A multi-core processor is much more multi-tasking when the program uses quite a few processor resources, but at the same time takes up a thread. A multi-core processor will obviously do better and be less overloaded. We can only wait.