Set micro-network to launch integrated circuit micro-letter public number: ' Every day IC ', major news immediately released, every day IC, every day set micro-net, product micro-Cheng! Replication Laoyaoic Micro Letter public number search adds attention.
1. Intel open 22 and 10nm process to arm architecture foundry business;
In 2017, the ARM Techcon conference, in some areas has formed a mutual relationship between semiconductor manufacturers, Intel and silicon smart financial property companies AMD, both announced that will establish a broad partnership. In such a relationship, one of the ways to work together is based on the ARM core architecture of the action chip, is expected to use Intel's 22 nm FFL process technology, as well as 10 nm HPM/GP process technology for OEM production.
In the past, Intel focused on the x86 core architecture market, with the ARM core architecture focused on the action market, each other is almost not quite the intersection. In the past, though, Intel has tried to enter the smartphone field with a x86 core architecture. And with the ARM core architecture of Qualcomm, also announced in 2017 combined with Microsoft Windows 10 operating procedures, into the past to x86 core architecture as the editor-in-chief of the electricity market. However, so far a failure to quit, the other has not yet launched the finished product. Therefore, in the current Intel on the semiconductor OEM market operating more and more active, with the former rivals, in some areas to shake hands, to expand the market together seems to be a viable thing.
And such a thing, in fact, has been implemented recently. For example, the Cortex-a55 core architecture, published by ARM in 2017, has been made using Intel's 22 nm FFL process foundry, and has experimented with 0 on smartphones. 45V voltage, frequency 2.35GHz performance. In addition, the ARM architecture SoC, which will be produced with the Intel 10 nm HPM/GP process technology, will also be streamed by the end of 2017. As for the updated generation core architecture, it will be expected to achieve 3.5GHz frequency, 0.5V voltage, that is, the single core maximum power consumption is less than 0.9 watts of performance. And this performance will be Gaotong 820 chip single core less than half of the power consumption.
At present, Intel's 14 nm process has been used in the x86 action chip products. However, because it is the core structure of x86, it further restricts the development of the exhibition in the consumer-level market, but also affects Intel's performance in the semiconductor foundry industry. Therefore, in order to further increase revenue, Intel in the arm Techcon conference stressed that the semiconductor foundry part of the arm of the core structure of the product will be released to the foundry.
The same 10 nm process, Intel has a process technology that can place 100 million transistors per square millimeter, TSMC has only 48 million transistors, and Samsung has only 51.6 million transistors, according to data released by Intel recently. So, according to Intel, Intel, the same-node process technology, is the leading competitor for more than 3 years. But who would be interested in Intel's 10 nm process technology to produce ARM chips? So far, the only thing that has been leaking is LG.
Technews
2. Ai wave to attack chip manufacturers to accelerate mergers and acquisitions;
Artificial intelligence (AI) wave to attack, the international big business to accelerate the strategic layout, the major chip industry to take the purchase and join the strategy, to urge the rapid jump in AI warfare. Among them, Intel smashed heavily to strengthen the full ecological chain advantage, in order to contend with Nvidia graphics chip technology strengths, in the face of the strong attack, Nvidia also got up to start to follow up the purchase and trade, the past year, the Silver bullet, in 5 countries invested in more than 10 companies, including the mainland automatic driving new entrepreneur Jing, As the application of technology matures, the AI industry to buy and integrate the tide will be more hot. While the global PC, data center platform city dominates the map, and the new generation of AI is coming, Intel has failed to take the lead, watching the Nvidia GPU parallel computing as the mainstream of technology and market focus. GPU technology is never as Super Micro, NVIDIA, but with a rich silver bullet in the hands of Intel, decided to adjust the strategy, in addition to all its AI business integration into the newly established AIPG (AI products Group) and AI Labs, and by purchasing and Alliance strategy, reinforcement in the unmanned aircraft, automatic driving, Medical and intelligent plant, such as multiple AI application platform combat power. Intel spends no money, which is more concerned about the 2015-year price of $16.7 billion to buy the FPGA manufacturer Altera, FPGA battle Power, 2016 with about 410 million dollars to buy machine learning new creation company Nervana Systems, In 2017, it hit $15.3 billion to buy the Mobileye, and the Intel Capital, the global investment arm, also continued to invest in several new ventures, including Mighty AI, Data Robot and Lumiata, such as the development of AI field companies, the investment amount of more than US $1 billion, recently announced in more than 60 million U.S. dollars to invest in 15 new companies, covering AI, 3D Medical vision, retail robotics, network security and other technical areas, the largest common denominator and ' data ' Closely related, the determination of Intel's transformational data company, including the mainland Beijing Horizon Robotics, mainly provides low-power, high-performance embedded AI solutions. The revenue scale and the silver bullet are not as good as Intel, Qualcomm's Nvidia, and finally got up to expand the Purchase and Alliance plan, recently its GPU Ventures investment division announced participation in the mainland, led by the mainland Qiming venture to automatically drive the new entrepreneur landscape technology investment case, the total investment of 52 million U.S. dollars, Early layout of the mainland will soon be the automatic driving, AI battlefield; The Jing department was founded in early 2017 by a senior executive at the Auto driving vehicle Department of Baidu, June has completed the first automatic driving mode test on public roads, which plans to put 50 self-driving vehicles on the mainland streets before the end of 2017, and to deploy test teams with full level 4 automatic driving capability on the mainland, The goal is to launch a similar bestäuber automatic driving car service in 2018, with its local and technological advantages gaining Nvidia's favor. In addition to Jing, NVIDIA has continued to expand its investment in new ventures, having invested over 10 companies in 5 countries over the past year, including Abeja, a new Tokyo start-up company specialising in the AI retail Analytics system, SoundHound, a new company in Silicon Valley, which focuses on creating a voice-priming AI solution, Zebra Medical, a new Israeli-created company, uses AI to read medical images, as well as new ventures from the mainland, Tucson the future, to develop self-driving trucks. Global AI War continues to heat up, in addition to hard technology, more than who's eyes precision and pockets deep, chip manufacturers buy and tide will continue to burn. Digitimes
3. Intel has a bright eye, but the AI field will meet AMD strong;
Intel revenue of $16.15 billion in the 3rd quarter of 2017, $1.01 per share, net income of $4.52 billion, an annual increase of 34% per share, estimated to be $3.25 in revenue for the whole year, exceeding market expectations. In terms of profitability, Intel's fundamentals have changed significantly, and share prices have been bullish, but the outlook is not a smooth one. According to Wccftech reports, Intel has highlighted opportunities for new markets such as artificial intelligence (AI) and cloud infrastructure, including self-driving cars, to further raise the 4th quarter and Full-year Outlook. Goldman Sachs predicts that the market for advanced driving aids and self-driving vehicles will be around $3 billion from 2015 to $96 billion in 2025 to $290 billion by 2035, one of Intel's most important markets. As a result, Intel's transition from the PC market to a new area looks like a success, but the problem is that Intel buys and Mobileye the Nvidia-led market, although Intel uses a serial method on its self-driving vehicle, and Nvidia uses a versatile graphics processor (GPGPU), If the market thinks that the serial method is not the right approach, Intel will face serious risk. For example, Mobileye plans to launch the EYEQ4 chip in 2018, with 10 of the 14 operational cores being accelerators, 3 different types of accelerators (Programmable macro arrays), VMP (vector microcode processors), and MPC (multi-threaded processing clusters). The actual number of EyeQ4 is 4 cores, 6 VMP, 2 PMA and 2 MPC. But Nvidia's lowest-order version of the processor is 3.2 times times faster than the best Mobileye chips available today. NVIDIA also publishes the $number DL tops AI supercomputer Pegasus, which is a magnitude faster than the EyeQ4 method, and can achieve level 5 levels of unmanned driving. This is why Intel buys and Nervana to launch a highly parallel approach to artificial intelligence, with Nervana neural network processors using parallel computing and very much like the general GPU. Nervana is not likely to lead Nvidia in the field of autonomous driving, but Intel must identify its own arms that enter the artificial intelligence battlefield, as well as the conditions for stepping into the deep neural network training industry. However, it is not yet possible to speculate on the impact of the chip, since Intel does not disclose the full specifications of the Nervana chip and has not published benchmark reviews to compare it with the CUDNN GPGPU network running. AMD, on the other hand, provides a higher core number of steps to Intel's domain at a cheaper price. Shortly after the launch of Intel's Skylake, AMD's Zen series products hit the Skylake processor, AMD's Threadripper processor also beat Intel, causing Intel to lose CPU market share. Although Intel quickly responded with the Core-x series and Coffee Lake. In fact, Intel's Coffee lake is facing a supply problem, while Hedt core-x is too expensive for mainstream consumers. The problem is that Intel has been building very expensive monolithic grains, but AMD uses a multiple-chip module (MCM) approach to build relatively inexpensive CPUs. Because operations are one of Intel's most important businesses, this is a rather serious problem, and in the coming quarters, Intel must be very careful with this particular market, or the city will be handed over to AMD. Digitimes
4. Microsoft developed HoloLens Artificial intelligence chip and will be used in other equipment;
Sina Science and technology news Beijing time November 2 Noon, Microsoft recently revealed that the next generation of HoloLens augmented reality will be equipped with a dedicated artificial intelligence chip. For now, however, Microsoft has a much bigger plan for such a custom chip, which is likely to be used in other devices.
Panos Panay, vice president of Microsoft Equipment business, said in an interview with CNBC that the company is still working to develop HoloLens chipsets, Parnolth Panai. Panai gave a positive answer about whether the chipset would be used for a wider range of Microsoft products. In addition, he said, these artificial intelligence chips may be authorized for partners to use.
"I think one of the most important things we do in surface and chip development is to explore opportunities to ensure that we have the technology inside surface and to provide partners with access to these technologies," he says. '
Microsoft says the ultimate goal of the AI chip in HoloLens 2 is to add specialized computing power to perform complex tasks such as image recognition and speech recognition. This is likely to give HoloLens unique functionality and faster processing speed without having to send data to the cloud Platform for processing.
After HoloLens, we are likely to see these technologies being used in other products, including PCs developed by Microsoft and its partners. Interestingly, in this case HoloLens became a plot. (Victoria Gold)
5. Edge operation will subvert the mobile communication network architecture?
The edge system can be used to reduce the load and transmit data of mobile communication core network. It also provides higher quality for existing services and is expected to create new services.
The American at Labs has been quietly carrying out its definition of the concept of ' marginal computer ' (Edge computers) and is slowly pushing the cloth department; In the long run, this work will have a profound impact on the future design of cloud and mobile systems. At defines edge operations as a cluster of servers and storage systems around its network to provide low latency services; The company foresees a wide range of systems that vary in size and position depending on application and demand.
In a recent speech at the Fog World Congress conference in Silicon Valley, at Labs, senior director of Alicia Abella, said: ' Edge computing is the next step we make the network fully effective, and we are busy building an edge computing architecture; She said: ' We want to operate nodes at the edge of the mobile data center, located in the building, the customer's location and the headquarters office, depending on where the demand is and where our spectrum is. We are developing a way to optimize location. '
The limbic system provides a wide range of uses, with the goal of assisting at to reduce the volume and delivery of data on the core network, as well as providing better quality for existing services, and is expected to spawn new services. One of the notable applications is to perform video analysis for security cameras. This kind of edge system may use the graphics processor (GPU), FPGA or other accelerator program, and the department in the city.
A more challenging application is the task required to support the automatic driving of vehicles, as such applications require substantial investment in the roadside infrastructure and are currently uncertain about their investment returns; Notably, Abella revealed that there are 12 million smart cars on the at network, and that the number is growing at a rate of 1 million per quarter.
At indicates that intelligent vehicles are one of the use cases where edge operations can be supported (source: at)
Seek small and medium GPU accelerator solutions
The different edge systems will be connected through different passes and at core networks, some of which may be in the same way, while others may be larger; And Abella says the hardware challenge the company has faced so far is to look for medium-and small-version GPU accelerators, because such schemes are typically large, high power consumption versions.
At has an early application prototype is to put the camera into the car, Raspberry Pi Development Board to control, the video transmission to the remote monitor; Abella said: ' The size of the edge operation system may also be small, but more realistically, we foresee some of the size as a cloakroom (walk-in closet) ... We are only starting to think about the strategy of this system. '
At does not provide technical details or timelines for its limbic operational system architecture (source: at)
Once such systems are widely used, they may have a significant impact on mobile systems; The mobile device can allocate part of the task to the edge operation system to save the battery power; Abella explains: ' Imagine a new generation of mobile devices, or a AR/VR device that does not need to be tied to a computer, using a moving edge operation ... There is also a way to get signal processing at the edge of death. '
The ultimate goal of at is to define a software platform for edge operations, which provides developers to write applications; But the concept has just come into being, and it may not be true until 2020 years later. Abella expects that, in the far future, the platform will achieve similar seti@home, the "Mass Outsourcing operation" (crowdsourcing of Compute) project that uses home PCs to search for extraterrestrial intelligence, and that participants can get paid for such things as Bitcoin (bitcoin) .
FOG World Congress assembled professionals from various fields, such as edge operations, object computing and cloud computing, where some vendors demonstrated miniaturization, rugged brakes and industrial servers for edge computing applications.
Compiling: Judith Cheng
(Reference text: At previews Edge Compute plans, by Rick Merritt) Eettaiwan