SE TALK ABOUT years several times, but an article in Nature.com anticipates a few weeks to the conclusions reached in the Semiconductor Industry Association (SIA), that brings together the leading manufacturers of chips the United States (such as Intel, AMD, Micron and IBM), along with similar organizations around the world: Moore’s law is the last stop.
the rule formulated by Gordon Moore, businessman computer Use, has proved its worth by in 1965, with different evolutions. Says “processor performance, and the number of transistors available for it, doubling every 18 months”. In recent years it became increasingly evident that the extreme miniaturization (and the costs associated with it) has led to a slowdown in the exponential growth of the power capacity of the chips and the number of transistors inside. The heat has become a significant problem with increasing density of transistors and within the next ten years will reach fundamental limits. According to Paolo Gargini, head of the ISA and former director of Intel’s technology strategy, “even with super aggressive efforts will reach the limit of 2-3 nanometers, where the components will be the size of 10 atoms.”
At that level the behavior of the electrons will be governed by quantum uncertainty that will make transistors unreliable. And despite massive research efforts, to date there is no clear successor to silicon, the main material of the electronics industry. So the new industrial roadmap – which will have the task of defining the development of the entire industry, including software – will no longer be based on Moore’s Law, but you look at what might be called a strategy “More than Moore”: instead of making chips best and ensure that arise as a result of solutions that exploit them, the paradigm could be reversed by focusing on the end goal (the software), and then work backwards in order to understand the characteristics that must have a chip to meet the demands.
Among these chips there will be new generations of sensors, power management circuits and other silicon devices required by a world where computing is increasingly mobile. This, however, does not mean the end of progress. “Innovation will continue, but will be more nuanced and complicated,” said Daniel Reed, computer and vice president of research at the University of Iowa. The first cracks in the Moore’s Law have appeared in the 2000s, when the production process fell below 90 nanometers, the electrons were moving too quickly into ever smaller silicon circuits, generating a lot of heat.
So the producers opted for two solutions: stop the race to the increase in clock frequencies, limiting the speed of the electrons and their ability to generate heat, and redesign the processors to contain not one but multiple cores – leading to problems in the distribution of loads of work, addressed in the following years and on which he still works. These two solutions have allowed manufacturers to continue to reduce the size of circuits and keep the number of transistors in line with Moore’s Law. But around 2020 we can no longer continue to climb the silicon because of quantum effects. As a workaround? According to An Chen , an electronics engineer who works at GlobalFoundries, there is still a big debate about. The alternatives are there, but many are ripe to be adopted on a large scale.
There is talk of totally change the paradigm and embrace the quantum computing, which to perform operations using typical phenomena of quantum mechanics. Or to adopt the neuromorphic computing, which aims to model the processing elements on the basis of brain neurons. In the case of quantum computing many believe will offer benefits only a niche of applications, while that of neuromorphic computing research is ongoing. Another route is to go to a new material, the spintronic materials cited recently by Intel 2D graphene-like compounds. Unfortunately at the time no material has proven faster than the silicon counterparts, capable of generating less heat and usable without problems in a production chain at a high volume.
It is a third approach, which is an architectural change, keeping the silicon but configuring it in totally different ways. It speaks 3D approach, ie the stack “layers” of logic components on each other, like a skyscraper, but at the moment this solution only good for the memory chips, are less affected because of the heat (the circuits consume only when accessing a cell).
For processors stack multiple layers is complicated, because they become warmer. One solution would be to integrate the processor and memory, since separating them impacting on 50% of the total heat generated by the continuous exchange of data between the two elements.
To do this you have to redesign the structure of chip, which the electrical engineer Subhasish Mitra and his colleagues at Stanford University in California have developed a hybrid architecture that stacks memory unit along with transistors made from carbon nanotubes. According to the researchers this architecture could reduce energy consumption to less than one thousandth of that of a standard chip.
Perhaps the underlying problem in the end of Moore’s Law is the change taking place in the computing industry . It has gone from a world dominated by desktop computers and data centers in a scenario in which the mobile that – it must be said – “the law.”
This has led to a change of priorities: Devices Mobile must have high autonomy (thanks to special chips) and the chips must be able to manage the various sensors and connectivity options. The chips also have to do less and less work because many operations are done on remote servers, in data centers of some great hi-tech giant.
And then, finally, there is the theme already mentioned cost . At every step of production they need new more advanced lithography machines. Set up a new production line requires investment of many billions of dollars and market fragmentation brought by the mobile devices complicates the giants of the return on investment. “My bet is that we will end up the money before the physical”, ruled Daniel Reed .
The new roadmap will give priority to energy efficiency, especially in optical Internet of Things , ensuring for example that the sensors have not need batteries, making use of waste produced by heat and vibration energy. The connectivity will be equally important and so is security.
“In a way,” said Shekhar Borkar , head of advanced research for Intel’s microprocessors, “the Moore’s law will evolve. ” “Seen from the perspective of the consumer, the law says that the value for the user is doubling every two years, and in this context the various innovations in the program should continue to ensure this.”
No comments:
Post a Comment