The hottest Intel meso transistor is expected to a

  • Detail

Intel's meso transistor is expected to achieve a huge leap in the processing power of artificial intelligence

less than a decade ago, neuroscientist Amir khosrowshahi was drilling a hole and inserting a needle into the skull to understand the human brain

he transformed his neuroscience knowledge into the success of entrepreneurs, and co founded nervana, a startup company that helps large companies run neural networks. This technology has now promoted the explosive achievements of artificial intelligence. Intel is the world's largest computer chip manufacturer. Only two years later, in 2016, it purchased nervana for more than US $350million

since then, khosrowshahi trained by Harvard University and the University of California Berkeley has become a key AI thinker of Intel. Artificial intelligence and chips are colliding with each other in new and profound ways

in an interview with VentureBeat, khosrowshahi, the current chief technology officer of AI, said that because there were a group of researchers there, he would stay at Intel. The team is building a new type of integrated circuit (IC), which is full of transistors and can one day operate at a very small amount that can make the upper and lower jaws different from each other - as low as 100 mV. He said that this would be a step in the direction of low voltage for brain communication

the new IC will change the rules of the game. Khosrowshahi said that this will promote the explosive growth of energy greedy artificial intelligence applications needed to solve some of the world's biggest problems, including climate change, waste management and increasingly complex food and transportation system logistics

what might be more important? He asked. This will change the field. You know, for the most part, we're talking about 30x performance

most of the researchers involved in the project are called meso (magnetoelectric rotating orbit) - admitting that it may take a decade to bring it to market to the decisive breakthrough required

a long-term project with short-term impact

khosrowshahi has no dispute that it may take ten years to complete the project, but he is also excited about the derivative results that meso may produce in the next two to five years. These can help push AI forward faster. We will see a series of new AI products, he predicted

khosrowshahi's role in this project deserves clarification. In many ways, he is an outsider. He is a member of the AI product design team, focusing on development and projects to have a direct impact. In addition, he lacks a background in physics and circuits, which is the basis for meso to focus on key research. Although his achievements in nervana have given him credibility in the AI field, he is trying to catch up with others at Intel. He admits that he has surpassed him. Here, he relies on people like Ian young, a senior researcher at Intel, who is a circuit designer and chief researcher

cmos crisis - and revolutionary alternatives

young Australians have been working in what he calls beyond CMOS research for nine years. In the past 35 years, CMOS has been the main method to build transistors, but it has problems in expansion. Although CMOS may allow transistors to use nodes as small as 7 nm or even 5 nm (100000 nm wide for human hair), its power requirements make it expensive and commercially infeasible for some types of computing, such as the AI algorithm that may be needed in the future. At the same time, most other research approaches fail to produce better alternatives, which is why everyone is running around claiming that Moore's law is dead

in December, Young's team (Intel will not disclose the size of the team) cooperated with some physics researchers at the University of California, Berkeley to make breakthroughs in key components such as precision reducers, servo electromechanical and controllers of industrial robots. They published an article on nature and publicly proposed a new meso device for the first time. This is the first time in many years that someone has provided a credible and revolutionary CMOS alternative - a solution that can enable Intel to regain Moore's law curve. Meso provides a commitment to general-purpose processors, said young and khosrowshahi, so it must be done anyway

the 100 mV switch of this device will be one fifth of that of advanced CMOS devices. Its Joule free switching energy will be reduced by about 30 times. In addition, its density is 5 times, non-volatile, and can provide very low standby power

the team provided the research roadmap needed to realize meso: mainly breakthroughs in the field of quantum materials, especially the topological states of related oxides and substances. To achieve this goal, Intel recognizes the need for open source research to work with universities and others. (see our interview with khosrowshahi and young, which details this.)

meso can generate derivative products. For example, asic

young for AI is not afraid of the expected time frame. Why is 10 years a problem? He asked. Young's mission is to ensure that meso devices are widely available - for all computing purposes. However, khosrowshahi said that even before the completion of meso, things will become interesting, because Intel does not need to wait for technology to enter its general-purpose CPU - with well-designed circuits, hairy interconnection requirements, production planning and wide applicability. In addition, chip design research will include breakthroughs in quantum material structures, which can be used for progress in various customization fields. One of the most promising areas is AI, because - a bit coincidentally - the structure of the meso transistor is strikingly similar to the computational elements in the AI neural network architecture

in fact, the integration of meso and AI was in our interview. Khosrowshahi and young were very excited

the reasons are as follows: since the transistors and interconnects of meso move the voltage in a way that can help data move more effectively in AI applications, the team can first try some meso methods in AI. Khosrowshahi is convinced that this will bring significant improvements to the application specific integrated circuit (ASIC) design of AI, and may even be a new product for learning and reasoning in the data center

Khosrowshahi explained that AI is simpler than CPU in key aspects. For example, AI data flow is very easy. Therefore, artificial intelligence is a good place for experiments on the substrate platform. We will replace the existing substrates for matrix multiplication, and possibly Silicon Photonics (we will use light). Or we will use room temperature quantum materials. Or we will use analog circuits. We can try all kinds of things

meso transistor characteristics reveal the commonality with AI

to understand the incredible connection between meso and AI, remember that computers use numeric strings (0 and 1) to store data. The computer's processor contains billions of transistors that are turned off or on during computer activity to reflect 0 and 1

The advantage of the

meso project is that it uses quantum materials to do this switching at the nano scale. Meso uses switches made of ferromagnetic ferroelectric materials. Magnetic switch solves a problem of CMOS, in which the switch uses more power

(things become strange here. For insiders: the switches in the meso system are similar to the neural network architecture in AI, as young explained. Using the meso magnet, multiple inputs can be introduced through most gates. This is similar to how the neural network uses weights to represent the influence of nodes. Similarly, you can see how low-voltage and high-efficiency transistor interconnection in meso can help AI. The challenge in artificial intelligence is great Amount of data movement and reduce the energy consumption involved in moving all data.)

young has the idea of more derivative products, including artificial intelligence algorithms (not yet envisaged), otherwise it will be limited by the available power

the connection between the low-energy demand of the chip and the low-voltage state of the human brain itself is a good match between circuit expert young and neuroscientist khosrowshahi who studies the brain. They agree with this. Young said that this is a giant that they have gathered together. It can be eluted with no-load gradient once (is there really a problem?) Big and interesting coincidence, their cooperation has profoundly affected the direction of young. For khosrowshahi, the human brain is evidence of existence, proving that there are better materials, and we should continue to look for them

(despite the analogy between meso and the human brain, and the operation in millivolts, young said that the human brain is still many orders of magnitude higher than the energy efficiency of meso.)

potential incentives require more investment

young said that a lot of work required to enter the final state of meso requires a lot of investment

he said that the industry needs to come together and cooperate more actively. Intel is a member of Semiconductor Research Corporation (SRC), which is also investing in beyond CMOS. The group includes IBM, micron, Samsung, TSMC and arm. President Trump's recent initiatives to support AI include $2billion for DARPA and funding organizations that go beyond CMOS research. But Yang still has more to do. The social benefits of this research will be enormous, but the project is still relatively unknown - even within Intel

khosrowshahi1 some PEBA variants can also prevent static electricity. It is difficult for him to listen to the opinions of some Intel leaders, because the company has many competing priorities, and it is difficult for meso to explain. When I go to executives to talk about device physics and quantum materials, it is difficult for them to see the correlation. This is a difficult story to convey, khosrowshahi said. Why does it matter? It is the core technology to run everything. If you don't have this, you have nothing

the momentum of artificial intelligence may help.

khosrowshahi said that there has been such a virtuous circle of innovation in the field of artificial intelligence, and things happen faster than people expected. The amount of work required by artificial intelligence provides a huge impetus for innovation, which makes him believe that meso can realize its potential faster than expected. I think we will get there faster than people expect, he said. If people don't pay attention to it, I don't care, because it is a competitive advantage for our company. Either way, it's good

Copyright © 2011 JIN SHI