Friday format: What happens when Moore's law stops acting
Today, Moore’s law refers to almost any exponential or other pattern. What did Gordon Moore actually say? In his article 1965 (paragraph “Costs and Curves”) he talks about doubling the components on an integrated circuit chip every year at the lowest possible profitable price. Later, in the article of 1975, he introduces another time indicator - 2 years.
The figure of 18 months and the very statement about the increase in processor power, along with an increase in their number and clock frequency, in an interview , Moore himself attributed to Dave House, an employee of Intel. However, in some sources, for example, here , this is what Moore's law understands. Even on the Intel website, the law is formulated that way .
/ photo by Steve Jurvetson CC
')
Of course, Moore's law is not a physical law and was merely a stated observation. Therefore, marketers use it as it suits them. But one cannot deny the influence of Moore's law, in any formulation, on the industry. Neil Thompson, an associate professor at the Massachusetts Institute of Technology's international business school, believes that the main advantage of the law was that he coordinated the actions of manufacturers and the market. Everyone knew what to expect and what to prepare for every two years. And without a general plan of action, some companies will have more difficulty.
Difficulties arose before, and there are different versions of the termination of this law. The Economist cites Peter Lee (Peter Lee), vice president of Microsoft Research, who is confident that the number of people who predict the end of Moore's law doubles every two years (see the second paragraph after the chart at the source ). Let's give some examples.
Back in 1996, Michael Malone published the results of a study where, among other things, a survey was conducted on how much Moore's law would still be relevant. And the experts' answers ranged from 3 to 20 years (see paragraph “Is This The Right Question?”). In another opinion, the law a little later ceased to be cost-effective. In 2009, an article about this appeared in the Financial Times, where it was predicted to be permanently terminated in 2014.
Today again they talk about the economic death of the law. This view is supported by the cost of production processors, which all continues to increase. According to Handel Jones, CEO of International Business Strategies (in the same article Economist), today the cost of the factory for the production of the most modern processors is $ 7 billion. The factory for the production of 5-nm processors (approximately 2020) will cost about 16 billion dollars, which will be about a third of Intel's annual income.
Today, the end of the action, or at least the slowdown, is attributed to the fact that Intel has postponed the release date of the new processor with 10-nm transistors in 2017. At the same time, company representatives point out that this is not a one-time action at all. The company will no longer have time to increase the number of transistors on the processor every two years. Intel may soon switch to a new transistor manufacturing technology. Although according to the roadmap (clause 7.5.2), transistors can be reduced (up to a maximum of 5 nm) until 2021 thanks to the maskless lithography technology.
There are many predictions about the development of new technologies and their connection with Moore's law, more precisely, with the version that speaks about processor performance. Experts talk about a number of technologies that can play a role in the development of the processor industry and will reduce the size of transistors without compromising their functionality. We are talking about graphene, or carbon nanotubes.
A team of scientists in the national laboratory of Berkeley managed to create a transistor with a gate size of 1 nm. To do this, scientists used graphene and molybdenum disulfide (MoS2), "weighting" electrons and avoiding leakage. On the other hand, PhD in the field of photonics Arnab Hazari (Arnab Hazari) says that the effective wavelength of photons is 1.3 micrometers, and the electron is 50-1000 times smaller. This may mean that photon processing equipment must be larger than what is used today. But since only a few light sources are needed for a photon processor, the dimensions will be preserved, but it is unlikely to be reduced, the author is sure.
Today, many data processing and storage processes are performed in the cloud . For example, to improve the performance of their servers, IT corporations use AI systems. Google and Microsoft developed processors for artificial intelligence systems and in-depth training, which increased their productivity.
In April of this year, a processor designed to work with artificial intelligence systems was also released in Nvidia, it is also designed for cloud computing. Last month, at the Structure 2016 conference, Urs Hölzle (Urs Holze) said that the new technology will increase productivity by only 30%, but this will be enough, and the sphere of cloud computing is the most favorable for the introduction of innovations.
Another technology with a great future is quantum computing. In November, at the University of Colorado, John Martinis (John Martinis) from Google conducted a seminar entitled “What will happen after Moore’s law: quantum computing”. Microsoft considers this technology promising: at the end of November, Todd Holmdahl, head of the quantum research department, said that the company was ready to move from research to development. Moreover, the goal is not just the creation of a scalable quantum computer based on quantum bits for the sake of increasing productivity, but the creation of an open-source technology that will turn medicine, materials science and create a new “quantum” economy.
Another direction of development of processors is the use of optical computing systems. Photons are much faster than electrons and allow for greater speed and performance. Previously, too high the cost of production was a decisive factor not in favor of such processors. But in some areas of computer technology, such as analog signal processing, ultrafast data processing is required, which only photon processors can provide.
A team from Princeton University in New Jersey managed to create a working optical processor, which also includes neuromorphic technology. At the heart of development is a neural network, each node of which must have the same frequency response as that of a neuron.
Nodes are shaped like tiny round waveguides cut into a silicon substrate in which light can circulate. Light is a signal at the input and adjusts the output power of the laser. Scientists have concluded that such a 49-node photon neural network handles the task three times faster than the central processor.
By the way, with Moore's law, confusion exists not only in the world of electronics. The first "laws" arose at least in the technical field: "Moore's law on software" or "Moore's law on throughput." But then the trend spread far beyond computer science: “Moore’s law in the field of biology”, “Moore’s law on economic well-being”, etc.
For several decades, the number of emerging “laws” named after Gordon Moore has doubled every five years. If this continues in the future, then by 2080, for every English word (and there are from a few hundred thousand to a million), there will be a separate law of Moore.
It is a little about our project ( 1cloud ) on Habré:
We also collect such digests:
The figure of 18 months and the very statement about the increase in processor power, along with an increase in their number and clock frequency, in an interview , Moore himself attributed to Dave House, an employee of Intel. However, in some sources, for example, here , this is what Moore's law understands. Even on the Intel website, the law is formulated that way .
/ photo by Steve Jurvetson CC
')
Beginning of the End
Of course, Moore's law is not a physical law and was merely a stated observation. Therefore, marketers use it as it suits them. But one cannot deny the influence of Moore's law, in any formulation, on the industry. Neil Thompson, an associate professor at the Massachusetts Institute of Technology's international business school, believes that the main advantage of the law was that he coordinated the actions of manufacturers and the market. Everyone knew what to expect and what to prepare for every two years. And without a general plan of action, some companies will have more difficulty.
Difficulties arose before, and there are different versions of the termination of this law. The Economist cites Peter Lee (Peter Lee), vice president of Microsoft Research, who is confident that the number of people who predict the end of Moore's law doubles every two years (see the second paragraph after the chart at the source ). Let's give some examples.
Back in 1996, Michael Malone published the results of a study where, among other things, a survey was conducted on how much Moore's law would still be relevant. And the experts' answers ranged from 3 to 20 years (see paragraph “Is This The Right Question?”). In another opinion, the law a little later ceased to be cost-effective. In 2009, an article about this appeared in the Financial Times, where it was predicted to be permanently terminated in 2014.
Today again they talk about the economic death of the law. This view is supported by the cost of production processors, which all continues to increase. According to Handel Jones, CEO of International Business Strategies (in the same article Economist), today the cost of the factory for the production of the most modern processors is $ 7 billion. The factory for the production of 5-nm processors (approximately 2020) will cost about 16 billion dollars, which will be about a third of Intel's annual income.
Today, the end of the action, or at least the slowdown, is attributed to the fact that Intel has postponed the release date of the new processor with 10-nm transistors in 2017. At the same time, company representatives point out that this is not a one-time action at all. The company will no longer have time to increase the number of transistors on the processor every two years. Intel may soon switch to a new transistor manufacturing technology. Although according to the roadmap (clause 7.5.2), transistors can be reduced (up to a maximum of 5 nm) until 2021 thanks to the maskless lithography technology.
Alternative development options
There are many predictions about the development of new technologies and their connection with Moore's law, more precisely, with the version that speaks about processor performance. Experts talk about a number of technologies that can play a role in the development of the processor industry and will reduce the size of transistors without compromising their functionality. We are talking about graphene, or carbon nanotubes.
A team of scientists in the national laboratory of Berkeley managed to create a transistor with a gate size of 1 nm. To do this, scientists used graphene and molybdenum disulfide (MoS2), "weighting" electrons and avoiding leakage. On the other hand, PhD in the field of photonics Arnab Hazari (Arnab Hazari) says that the effective wavelength of photons is 1.3 micrometers, and the electron is 50-1000 times smaller. This may mean that photon processing equipment must be larger than what is used today. But since only a few light sources are needed for a photon processor, the dimensions will be preserved, but it is unlikely to be reduced, the author is sure.
Today, many data processing and storage processes are performed in the cloud . For example, to improve the performance of their servers, IT corporations use AI systems. Google and Microsoft developed processors for artificial intelligence systems and in-depth training, which increased their productivity.
In April of this year, a processor designed to work with artificial intelligence systems was also released in Nvidia, it is also designed for cloud computing. Last month, at the Structure 2016 conference, Urs Hölzle (Urs Holze) said that the new technology will increase productivity by only 30%, but this will be enough, and the sphere of cloud computing is the most favorable for the introduction of innovations.
Another technology with a great future is quantum computing. In November, at the University of Colorado, John Martinis (John Martinis) from Google conducted a seminar entitled “What will happen after Moore’s law: quantum computing”. Microsoft considers this technology promising: at the end of November, Todd Holmdahl, head of the quantum research department, said that the company was ready to move from research to development. Moreover, the goal is not just the creation of a scalable quantum computer based on quantum bits for the sake of increasing productivity, but the creation of an open-source technology that will turn medicine, materials science and create a new “quantum” economy.
Another direction of development of processors is the use of optical computing systems. Photons are much faster than electrons and allow for greater speed and performance. Previously, too high the cost of production was a decisive factor not in favor of such processors. But in some areas of computer technology, such as analog signal processing, ultrafast data processing is required, which only photon processors can provide.
A team from Princeton University in New Jersey managed to create a working optical processor, which also includes neuromorphic technology. At the heart of development is a neural network, each node of which must have the same frequency response as that of a neuron.
Nodes are shaped like tiny round waveguides cut into a silicon substrate in which light can circulate. Light is a signal at the input and adjusts the output power of the laser. Scientists have concluded that such a 49-node photon neural network handles the task three times faster than the central processor.
By the way, with Moore's law, confusion exists not only in the world of electronics. The first "laws" arose at least in the technical field: "Moore's law on software" or "Moore's law on throughput." But then the trend spread far beyond computer science: “Moore’s law in the field of biology”, “Moore’s law on economic well-being”, etc.
For several decades, the number of emerging “laws” named after Gordon Moore has doubled every five years. If this continues in the future, then by 2080, for every English word (and there are from a few hundred thousand to a million), there will be a separate law of Moore.
It is a little about our project ( 1cloud ) on Habré:
- Customer focus as the basis of business: Experience IaaS-provider "1cloud"
- A bit about security in the "cloud": Experience IaaS-provider "1cloud"
- How to make virtual infrastructure more accessible
We also collect such digests:
Source: https://habr.com/ru/post/317204/
All Articles