In recent decades, chip designers have managed to consolidate more and smaller transistors on a single silicon chip, and the performance of computers continues to increase at an alarming rate: the number of transistors on a single wafer doubles every two years on average. , notebooks and smartphones are constantly being enhanced and their prices are decreasing.
Development meets resistance: Small chips are high in energy consumption However, researchers are now worried that such rapid development will reach the limit. The problem is not whether the designer can install more transistors on the chip. The problem they can certainly do is that all the tubes work with huge amounts of energy, just as a city cannot power its entire streetlight system. At the same time, they will overheat.
The result may be that people who are obsessed with new gadgets have already become accustomed to the retailer's exaggeration of new products, and have to accept the reality: the new generation of electronic products is only slightly improved, rather than soaring speed, falling prices, and more perfect .
Simply put, the next hot product may take longer to appear.
"Indeed, if you just follow the scale of the previous processor and scale it, you will not be able to make any further progress," said William J. Daly, chief scientist of graphics processor maker Nvidia and Stanford professor of computer science. "Think about it now." Making progress requires real innovation."
A paper at the International Computer Architecture Conference in June summarized this issue: Today, high-end microprocessor chips are loaded with too many transistors and it is difficult to supply them with power at the same time. Therefore, some transistors can only be powered down. The terminology is called dark while other tubes are in operation. This phenomenon is called dark silicon.
According to the authors of the paper, as early as next year, 21% of the transistors in these advanced chips will have to be in dark state at any one time. After about three more generations of chips, that is, more than 50 years, this restriction will be even more severe. There will be more transistors on the chip, but up to half of them will have to be in a dark state to avoid overheating.
"I think the chips won't melt, they'll be like liquids flowing on a circuit board. That's an exaggeration," said Doug Berg, a computer scientist at Microsoft Research, who wrote in an e-mail, "But you will The wrong result, some parts of the final circuit will be welded together, so that the chip can not be repaired."
This problem may affect the law that has been established for decades by the computer industry: Moore's Law. This law was proposed by Gordon Moore, one of Intel’s founders. He believes that the number of integrated circuit chips that can accommodate transistors has roughly doubled every two years, which has led to a dramatic increase in the performance of consumer electronics.
If the pace of development slows down, some of the innovations that people take for granted will not appear or they will slow down. There will be no new type of personal computers, smart phones, LCD TVs, MP3 players, or anything that can quickly create billions of industries and thousands of jobs.
In the paper, Dr. Berg and his collaborators simulated the power consumption of more than 150 commonly used microprocessors. It is expected that by 2024 the average speed will increase by only 7.9 times. In contrast, if there are no restrictions on the performance of the tube, the maximum growth rate may reach 47 times.
However, some scientists hold different opinions and believe that as long as new ideas and designs continue to emerge and the computer industry maintains rapid development, this will not happen. For example, Dr. Daley of Nvidia is optimistic about future chip designs.
"The good news is that the original design was really inefficient, leaving plenty of room for improvement," he said. But some experts irrelevant to their research believe that Dr. Berg and other authors from the University of Texas, Washington University and the University of Wisconsin have accurately pointed out a real problem.
Dr. Xega Y. Burka, an Intel lab researcher, believes Berg’s analysis is at the heart of the analysis, but he added: “His conclusions are slightly different from mine. Maybe the future will not be able to reproduce past glories, but it will not be bleak. of."
Dr. Burka cited a number of new design ideas that he believes will help solve the limitations identified in the paper. Intel recently developed a technology that allows different parts of the processor to consume different amounts of energy. The processor also has low-rate, low-power transistors and transistors that can be quickly switched but consume large amounts of energy.
Today's processor chips increasingly use dual-core or even multi-core, ie, two or more central processors, which enables them to run multiple programs at the same time. In the future, Intel computers will use different cores to solve different problems maximally, and only some of them will consume more energy.
Although Intel announced in May that the use of 3-D design can put more tubes on a single chip, this technology does not solve the energy consumption of dark silicon mentioned in the paper. The authors of the paper said that they tried to list some promising innovations. They think the problem is how far the innovators can go on overcoming energy constraints.
"The direction is limited, and how to develop lies in attitude," said Dr. Burg.
For some time, chip designers have been working hard to solve energy limitations. A century ago, it was widely believed that increasing the chip's clock speed (calculation speed) was an easy task. However, at a clock speed of 3 gigahertz, the chip began to melt due to overheating. The development of the industry hit a wall, which triggered a rush of intense design innovations.
Today, some leading designers think there is still much room for innovation. David A. Patterson, a computer scientist at the University of California, Berkeley, said that the dark silicon phenomenon does exist, but he doubts the author's pessimistic conclusion.
"Some papers contain such profound meaning that if we do not innovate, we will die. This is one of them," Dr. Patterson said in an e-mail. "Since we do not want to die, it must mean that we need to innovate. â€
Development meets resistance: Small chips are high in energy consumption However, researchers are now worried that such rapid development will reach the limit. The problem is not whether the designer can install more transistors on the chip. The problem they can certainly do is that all the tubes work with huge amounts of energy, just as a city cannot power its entire streetlight system. At the same time, they will overheat.
The result may be that people who are obsessed with new gadgets have already become accustomed to the retailer's exaggeration of new products, and have to accept the reality: the new generation of electronic products is only slightly improved, rather than soaring speed, falling prices, and more perfect .
Simply put, the next hot product may take longer to appear.
"Indeed, if you just follow the scale of the previous processor and scale it, you will not be able to make any further progress," said William J. Daly, chief scientist of graphics processor maker Nvidia and Stanford professor of computer science. "Think about it now." Making progress requires real innovation."
A paper at the International Computer Architecture Conference in June summarized this issue: Today, high-end microprocessor chips are loaded with too many transistors and it is difficult to supply them with power at the same time. Therefore, some transistors can only be powered down. The terminology is called dark while other tubes are in operation. This phenomenon is called dark silicon.
According to the authors of the paper, as early as next year, 21% of the transistors in these advanced chips will have to be in dark state at any one time. After about three more generations of chips, that is, more than 50 years, this restriction will be even more severe. There will be more transistors on the chip, but up to half of them will have to be in a dark state to avoid overheating.
"I think the chips won't melt, they'll be like liquids flowing on a circuit board. That's an exaggeration," said Doug Berg, a computer scientist at Microsoft Research, who wrote in an e-mail, "But you will The wrong result, some parts of the final circuit will be welded together, so that the chip can not be repaired."
This problem may affect the law that has been established for decades by the computer industry: Moore's Law. This law was proposed by Gordon Moore, one of Intel’s founders. He believes that the number of integrated circuit chips that can accommodate transistors has roughly doubled every two years, which has led to a dramatic increase in the performance of consumer electronics.
If the pace of development slows down, some of the innovations that people take for granted will not appear or they will slow down. There will be no new type of personal computers, smart phones, LCD TVs, MP3 players, or anything that can quickly create billions of industries and thousands of jobs.
In the paper, Dr. Berg and his collaborators simulated the power consumption of more than 150 commonly used microprocessors. It is expected that by 2024 the average speed will increase by only 7.9 times. In contrast, if there are no restrictions on the performance of the tube, the maximum growth rate may reach 47 times.
However, some scientists hold different opinions and believe that as long as new ideas and designs continue to emerge and the computer industry maintains rapid development, this will not happen. For example, Dr. Daley of Nvidia is optimistic about future chip designs.
"The good news is that the original design was really inefficient, leaving plenty of room for improvement," he said. But some experts irrelevant to their research believe that Dr. Berg and other authors from the University of Texas, Washington University and the University of Wisconsin have accurately pointed out a real problem.
Dr. Xega Y. Burka, an Intel lab researcher, believes Berg’s analysis is at the heart of the analysis, but he added: “His conclusions are slightly different from mine. Maybe the future will not be able to reproduce past glories, but it will not be bleak. of."
Dr. Burka cited a number of new design ideas that he believes will help solve the limitations identified in the paper. Intel recently developed a technology that allows different parts of the processor to consume different amounts of energy. The processor also has low-rate, low-power transistors and transistors that can be quickly switched but consume large amounts of energy.
Today's processor chips increasingly use dual-core or even multi-core, ie, two or more central processors, which enables them to run multiple programs at the same time. In the future, Intel computers will use different cores to solve different problems maximally, and only some of them will consume more energy.
Although Intel announced in May that the use of 3-D design can put more tubes on a single chip, this technology does not solve the energy consumption of dark silicon mentioned in the paper. The authors of the paper said that they tried to list some promising innovations. They think the problem is how far the innovators can go on overcoming energy constraints.
"The direction is limited, and how to develop lies in attitude," said Dr. Burg.
For some time, chip designers have been working hard to solve energy limitations. A century ago, it was widely believed that increasing the chip's clock speed (calculation speed) was an easy task. However, at a clock speed of 3 gigahertz, the chip began to melt due to overheating. The development of the industry hit a wall, which triggered a rush of intense design innovations.
Today, some leading designers think there is still much room for innovation. David A. Patterson, a computer scientist at the University of California, Berkeley, said that the dark silicon phenomenon does exist, but he doubts the author's pessimistic conclusion.
"Some papers contain such profound meaning that if we do not innovate, we will die. This is one of them," Dr. Patterson said in an e-mail. "Since we do not want to die, it must mean that we need to innovate. â€
Bimetal Strip,Thermostat Strip,Clad Metal Strips
INT Metal Products Factory , http://www.libometal.com