Set micro-grid launch micro-channel IC WeChat public number: 'Daily IC', real-time release of major news, every day IC, every day set of micro-network, micro-into a!
1. Into Intel China Research Institute, has developed human-like neuromorphic LOIHI chip;
(Set micro / text) Some people say that 'technology giant is to promote the advance of human civilization pilot', this sentence is not an exaggeration. The world's top technology companies, its technology evolution and research direction, can affect the global industry Trends, while most of the most cutting-edge technologies are born in these companies' research institutes, so it is interesting to see what their institutes are doing.
Recently, Intel opened up the 'Intel China Research Institute' to the media and showed the media the latest research results of Intel China Research Institute in a series of forward-looking technologies such as artificial intelligence, robotics, 5G, virtual reality, etc. Intel China Research Institute Dean Song Jiqiang also led the team collective appearance, shared the Intel China Research Institute's overall orientation, research direction and innovation mechanism.
Intel China Research Institute: The most important overseas research institute, won a number of awards
From the positioning point of view, Intel will 'Intel Institute' as the company's 'technology innovation engine.' Currently, Intel Research Institute has four major functional laboratories in the United States, with about 800 researchers, whose research mainly for Intel The frontier areas of concern for data strategy are divided into four major functional institutes: the Institute of Circuit and Architecture, the Institute of Systems and Software, the Institute of Wireless Communications, and the Institute of Security and Privacy.
There are mainly overseas research institutes in Europe and China Research Institute, which is one of three "regional research institutes" established by the Company overseas and is also an important research institution outside the United States. Our independent research agenda, which we can produce on our own, was established in 1998 to develop cutting-edge technologies in an intelligent, all-encompassing environment to promote the intelligent industrial upgrading, technological innovation and data economy in China and around the world.
It is reported that the entire Chinese Academy of Sciences has about 60 formal researchers, of which doctoral and post-doctoral staff accounted for 70% .At the same time, the Institute receives from key universities at home and abroad as a graduate student internship in scientific research projects, and set up by the national postdoctoral After the establishment of the CMC approved the establishment of post-doctoral research station, perennial and major domestic colleges and universities jointly train post-doctoral.
In addition, Intel Research China has academic partners around the world with global capabilities, including with Tsinghua University, Massachusetts Institute of Technology, Carnegie Mellon University, etc. Only in 2016 and 2017 two years , Intel China Research Institute published a total of 27 papers, with more than 100 patents, and won several major domestic and international awards.
Currently Intel president of China Research Institute for the Song Jiqiang, he joined the Intel China Research Institute in 2008, but also to create a core member of the Intel Edison product prototype.
At this event, Song Qijiang said Intel Institute sees more about what kind of big technological change will happen in the next three to five years. On the one hand, how to help the existing product line through the prospective research of technology Enhance its competitiveness and provide technical impetus for new products to come out.On the other hand, we try to see what kind of disruptive technology may be possible in the future.
In addition, Song Qijiang also told the media, Intel Research Institute is taking an industrial research institute model, the so-called Industrial Research Institute, it is not to send papers as the main indicator, not with everyone in the academic competition, but whether the groundbreaking technology Truly applied to the product to go, to change some of the current state of the technology products, science and technology really promote to the actual use of our power.
Intel China Research Institute of the four areas of black technology
From the field of research, the current Intel China Research Institute consists of six teams, namely the robot system research laboratory, robot interaction research laboratory, cognitive computing laboratory, intelligent storage laboratory, communications infrastructure laboratory and new technology center.
On this opening day, Intel China Research Institute focused on introducing the four technological developments under study to the media, including the robot HERO platform, 5G + VR, visual understanding of face and emotion recognition, and neuromorphic chips. Look:
Robot HERO platform:
Intel Research Institute said that the robot and artificial intelligence, the combination of the point between them is called intelligent robot.As the most central feature of intelligent robot, it must have the ability to learn independently.An autonomous system is a research direction related to artificial intelligence.
Based on this goal, Intel Institute introduced the HERO intelligent robot open platform (Heterogeneous Extensible Robot Open Platform) .It is Intel China Research Institute for intelligent robots (including service robots, medical robots, autonomous vehicles, etc.) to create a Low power, high performance, small size heterogeneous system platform solutions.
In this scenario, the CPU acts as a control center and works with FPGAs and other specialized accelerator chips such as Movidius' VPU to provide efficient performance.The complete HERO hardware system features an Intel Core ™ series of CPUs powered by the Intel Arria 10 GX Series 1150 FPGA as a heterogeneous accelerator, capable of processing large amounts of data in real time and running a variety of intelligent algorithms.
In the field, Intel also provided an intelligent robot equipped with HERO platform for demonstration, and in the field test, the robot possesses the ability to move through the fusion of laser positioning and visual positioning, and can also include the perception of human faces, objects and behaviors ability.
At this open day event, Intel China Research Institute announced a new partnership program around the HERO platform around which Intel China Research is committed to working with the industry's platform, application and ecosystem partners through R & D Cooperation, academic innovation and industry demonstration, together to create a complete, efficient and open platform, continue to expand the application scenarios, accelerate the arrival of technology and products.
5G + VR:
Intel has been emphasizing end-to-end capabilities, and communication is the bridge to connect.China Intel China Research is also studying 5G and its related applications.Researchers said that our research focuses on the communication and computing integration, the typical application is to create high-quality VR experience.
Commenting on the current VR experience is not good, the researcher said that at present see mobile phones, TVs and other screens, the perspective of 10-30 degrees, while the VR eye can be extended to 100-150 degrees, an increase of several times, in addition to vertical Therefore, the same resolution of the picture or video on the phone, the TV is very clear, but the VR is very vague, you want to achieve VR display HD requirements, you need to increase the resolution of 25- 40 times, so you need to occupy a large amount of traffic.While the combination of 5G and VR, through the communication and computing integration can create an efficient communication and computing system that can solve some of the key issues in the mobile VR side.
At the scene, Intel demonstrated the ultra-high-resolution panoramic video capture system, which is mainly used for wireless VR devices, users can watch panoramic video immersion, or participate in VR game interaction.It uses the ISP bus synchronization camera shutter, through distributed The architecture processes sampled images and splices panoramic video and interpolates stereoscopic information using the optical flow method, which simultaneously uses Intel Core GPU high-performance compressed images to render host output to remote terminals over wireless and uses the OFFLOAD technology to reduce display delays. With these technologies, television viewers can experience the spectator's experience in front of the Olympic Games even at home, or choose different viewing angles within the venue.
Visual understanding of face and emotion recognition
Researchers from cognitive laboratories have introduced some research techniques in face analysis and sentiment analysis in recent years.
It is reported that, in terms of face analysis and emotion recognition, Intel China Research Institute has long been the most advanced algorithms, and won the EmotiW game (emotion recognition competition in the natural environment), the speed of the algorithm is second 200 times more than its predecessor, can be used in driverless scenarios. Based on the Intel Real-Time Portrait Style Migration System, Intel Research China real-time migrated multiple art portraits into static target images and dynamic target video. Technology can be widely used in AR / VR, live online, video, multimedia and social networking.
In addition, Intel China Research Institute has developed Intel 3D face facial expression capture technology based on machine learning and depth learning algorithms.Relevant researchers from Intel hundreds of thousands of face image data, training multiple face recognition models for video images Intelligent analysis and 3D modeling, not only can detect and identify faces, but also accurate reconstruction of 3D faces, and real-time tracking changes in facial expressions, pre-designed effects of the material attached to the 3D face and superimposed on the video, In order to achieve realistic cool facial effects.
Nerve Mimicry Chip:
As the finale, Dean Song for everyone to share the most advanced neuromorphic calculation, the English name is LOIHI.
In exploring new computing models, Intel is divided into four phases: traditional, deep learning, neuromorphic computing, quantum computing.
The traditional model is our programming, our mind will first build a model of the problem.Use this model to program and thus solve the problem.Delicated learning is the most mainstream practice.Using a large amount of data to learn, to train out the model .Nine mimetic computing Not the same, it's a human-like way, and quantum computing entangles a large number of answers in a superposition.
At present, Intel has made achievements in neuromorphology calculation .Song Dean said that the latest release of Intel neuromorphology computing chip named LOIHI .First, not von Neumann structure - the calculation of computing, storage back to storage.Neonomorphic computing chip LOIHI is calculated and stored together, will form a lot of distributed units, but also asynchronous computing.
President Song said the chip directly benefits from Moore's Law. The number of transistors on a neuromorphic chip, along with its capacity and power consumption, will continue to increase its density and decrease over the next two to five years Its power consumption directly leads to increased brain capacity of this chip.
From the picture point of view, LOIHI chip has a lot of grids, there are many small blocks in the grid, Intel called a core.On LOIHI each single chip contains 128 cores, each core can achieve a lot of neurons, Each neuron can be interconnected with other neurons .Song Dean explained that we know there are 80 billion neurons in the human brain, each neuron can be connected with tens of thousands of neurons.We are actually simulating This model, to learn some of the rules into each neuron inside.
In addition, many of the neuron parameters on the LOIHI chip can be updated and adjusted in real time, and can be modified and self-learned while working, and all previous chips do not have this capability, The tools are trained on a good model and then placed in a chip, which solidified will not change.
Finally, President Song revealed that at the end of this year, the LOIHI chip will be taken back to China from the US laboratories for testing and open to some cooperating universities and agencies to make corresponding experiments.
For the future, Dean Song believes that neuronomorphometry, quantum computing are not mutually exclusive, no matter whether they are traditional or deep learning. No one has said that the quantum computing is successful after the previous one did not say that the neuromorphic calculation is successful Do not go deep into the future, they complement each other, and everyone has their own suitable job.
2. Intel: Edge computing overlay artificial intelligence to accelerate IoT vertical industry change;
Judging from my judgment on the overall technology trend, there are two important points in the development of the Internet of Things: one is edge computing and the other is artificial intelligence. We think that the synergy between these two aspects Will promote the rapid development of the entire Internet of Things technology in the future, the ultimate realization of the three stages of the development of the Internet of Things, from interconnection to intelligence, from intelligence to autonomy. 'This is November 30 this year, at the second edge computing industry summit, Intel China Region Internet of Things CTO Zhang Yu told reporters the point of view.
Also at this time last year, six industry-leading companies or organizations, including Huawei, Intel and ARM, set up the "Edge Computing Alliance" in Beijing, and according to official reports, after a year of development, it now has 154 Home member units, including leaders from smart manufacturing, smart city, energy, power and ICT industries, as well as research institutes and universities in related fields.
As the name suggests, "edge computing" emphasizes the computing power of local and terminal side.In fact, earlier, the industry believes that the future of technology will be carried out in the cloud, the terminal side to take more to collect and transmit data.However, the last two years Once again, the terminal side still needs to be computationally strong and efficient enough to be truly intelligent.
It is also the point that Huawei, Intel, ARM and other industry leaders initiated the establishment of 'Edge Computing Alliance', which respectively share industry application domain, data domain, network link, sensor, chip and industrial manufacturing according to division of labor. Intel, as a core member, is one of the only end-to-end total solution providers in the coalition to play a pivotal role.
At this summit, Intel demonstrated the major achievements of the past year, of which one of the most interesting products is Intel's joint lab of the United States Institute of Automation show edge computing test bed - intelligent robot. It is reported that this robot is intended To verify the effectiveness of depth-based machine vision solutions in real systems, the system realizes 3D video capture by Intel® RealSense ™ technology and detects and identifies objects based on the Intel Xeon Platform. The results will control the robot arm to complete the object grasping Take the operation.
As can be seen in this scenario, Intel has now applied artificial intelligence technology to the field of edge computing, while edge computing + artificial intelligence is also a hot topic discussed at this conference.
In the keynote speech of the summit, Dr. Zhang Yu, CTO of Intel's China Internet of Things Division, specialized in intelligentizing the edge of the network, said that realizing intelligence at the edge of the network is one of the key links in controlling data floods, With the development of artificial intelligence in full swing, massive data needs to be analyzed and extracted quickly and effectively, which greatly enhances the demand for edge computing.
Speaking of the company's strengths, Zhang said Intel is providing an end-to-end, industry-leading, full-stack artificial intelligence solution that includes: Xeon processors, Xeon Phi processors, Intel® Nucleav ™ nerves Network processors and leading and complete hardware platforms such as FPGAs, networking and storage technologies, as well as a variety of software tools and libraries to optimize the open source framework It is worth mentioning how to balance power consumption and computing power for edge computing With one of Movidius's leading single-watt computing capabilities, Intel is providing the industry with a low-power, high-performance edge computing solution.
Compared to the PC market, the unified program, the shape of the Internet of things strange, but also involves all walks of life. Segment and vertical is the problem-solving strategy.In Intel's view, the edge computing and artificial intelligence technology should be based on different industries Features to optimize.
At present, the Intel Internet of Things Division focuses on four major areas of security and surveillance, retail, transportation and industry in China, and Intel also offers different solutions depending on the characteristics of different industries. For example, some manufacturers such as Hikvision and Dahua in the fields of security and surveillance The introduction of Intel's Movidius products, you can do in the front-end real-time face detection, flow detection.
Zhang Yu said that artificial intelligence is a more common technology, but the real vertical industry, each vertical industry has its own specific needs, because the use of different scenarios, the object needs to detect the characteristics of different, so here will There are a few specialized algorithms that are specifically optimized for user scenarios, and Intel, in addition to providing a general purpose chip, is working with partners in a number of vertical industries to develop new technologies based on specific vertical industry needs.
From the marginal computing industry summit point of view, Intel proposed edge computing superimposed artificial intelligence has become a consensus of the industry, through a combination of appearance technology, will be for a variety of sub-vertical industries with truly differentiated value of the product, This will also accelerate the arrival of the Internet of Things smart era.
3. 富 比士: Intel Review Own EyeQ5 better than NVIDIA Xavier fear of unfair;
Intel (Intel) previously acquired Mobileye high price, showing a proactive access to the global self-driving chip market determination, Intel CEO Brian Krzanich recently also said the company developed by Mobileye EyeQ5 self-driving chip performance efficiency, higher than the NVIDIA product line 2.4 Times, which seems to show that Intel has superior control over NVIDIA self-driving chip technology, but from the Intel EyeQ5 product planning blueprint is expected to begin production in 2020, but NVIDIA Xavier self-driving chip 2018 is expected to put into production point of view, display The difference between the two production processes is about two years, so Intel's EyeQ5 and NVIDIA Xavier may be considered as a comparison with the generation may not be fair, maybe two years NVIDIA will launch a new generation of more efficient and efficient self-driving chip also Perhaps, according to Forbes, the EyeQ5 is Intel's first chipset for Level 4 and Level 5 and is expected to start production in 2020, which means it will not be possible to get started in the car until 2022. The EyeQ5 Mobileye has been introduced before the acquisition of Intel has its original design specifications, after the acquisition of Intel Fine-tune the EyeQ5, EyeQ5 Can double to 24 trillion operations per second (24 TOPS), but the power consumption is also doubled to 10 watts, power consumption is 2.5 watts compared to the EyeQ3 and the EyeQ4 is expected to begin in 2018 by more than 3 watts of power consumption Quadruple and more than three times, Intel and the goal in the technology-leading 7-nanometer manufacturing EyeQ5. Intel EyeQ5 performance data is based on product simulation derived from the original EyeQ5 Mobileye performance doubled, which means that Intel The EyeQ5 needs to be doubled to achieve the Level 5 autopilot technology previously predicted and two EyeQ5 chips may be used in the final SoC, although Intel chief executive said EyeQ5 is more efficient than its competitors The above simulation data shows EyeQ5 performance should not be bad, but EyeQ5 is, after all, the chip is not planned to start production in 2020, there are more than 3 years ago, coupled with the comparison of Intel's NVIDIA Xavier chip has now been sent to the manufacturing (tape out) , Will be put into production in 2018, which is equivalent to EyeQ5 is scheduled to put into production a full two years earlier, meaning that NVIDIA Xavier is equivalent to the expected production schedule in 2020, the previous generation of self-driving chips In this case, comparing NVIDIA Xavier with 16-nanometer process to EyeQ5, which is expected to adopt the advanced 7-nanometer process, is problematic. It can be said that Intel is too early to conclude that any type of energy efficiency of EyeQ5 and NVIDIA self-driving chips In addition, since both NVIDIA Xavier and Intel EyeQ5 will become part of a larger self-driving platform design, it may not make sense to compare who's better on a chip, as Xavier is a converged SoC capable of In combination with other Xavier SoCs and Graphics Independent Graphics Units (GPUs), the Drive PX Pegasus, featuring two NVIDIA Drive PX platforms and two Xavier SoCs and two Volta GPUs, is also available as an example. In addition, NVIDIA is also the world leader in global artificial intelligence AI) Solution Leading vendors, particularly NVIDIA's new Volta architecture and integrated tensor core, also play an important role in the Xavier SoC, making Xavier the highest-performance solution for automotive applications announced to date, with the Volta architecture, Not to mention NVIDIA architecture to support all training architecture.On the other hand, it is still not clear that using Intel EyeQ5 chip platform, will use multi-chip Design configuration, equipped with other accelerators at the same time, or which AI architecture will be supported, etc. Not only NVIDIA, Intel EyeQ5 in the current global market for self-driving chips are still facing many competitors and competitors competing products into this market, Including NXP's BlueBox platform from Qualcomm, the Open Autonomous Vehicle Platform from Renesas, the DRS360 platform from Siemens and Mentor, Samsung Electronics and Harman Of the self-driving platform, as well as the mainland Baidu's Apollo platform etc. It is noteworthy that the new self-driving chips and platforms to be able to be formally shipped to the automotive manufacturing side, it takes at least two years, This assumptions assume Level 4, Level 5 NVIDIA Xavier chip is indeed put into production in 2018, means that the advent of the first wave of the advent of 2020 may become the first cars equipped with Xavier chips. DIGITIMES
4.Qorvo show the industry's first 5G RF front-end module to accelerate the pace of industrial 5G transition;
Qorvo, the leading provider of RF solutions for defense, microgrid messaging, mobile applications, infrastructure and aerospace, defense applications, today introduced the QM19000, the industry's first Sub-6 GHz 5G RF front-end module, and the highly integrated, high-performance QM19000 enables high Linearity, ultra-low latency and high throughput to meet or exceed the development needs of future 5G applications Qorvo presented its prototype 5G terminal based on this module recently at the 2017 China Mobile Global Collaboration in Guangzhou.
5G era will be the era of all things is the era of the wisdom of the network is to connect the future era.From the 5G standardization process point of view, 3GPP first 15G Rel.15 has been officially frozen in December this year, which is NSA ( Non-independent networking) has frozen the core standards.This also means that the first phase of the first version of 5G NR agreement has been completed, compared to the original plan six months in advance, the industry manufacturers in the 5G layout again speed.According to the latest 3GPP plan, 5G SA standard will be frozen in June 2018 and the full version of Rel.16 will be completed by the end of 2019. According to the 5G network deployment, 5G will not start commercial scale until 2019-2020 Deployment.According to the traditional industry conduction process, the terminal generally lag behind the network about 12-18 months.With the scale of the deployment of 5G network, the user's 5G terminal will have a better experience.
Qorvo, as a 3GPP representative, helped shape the 5G standard and worked closely with the world's leading manufacturers of wireless infrastructure, network operators, chipset vendors and smartphone manufacturers to lay the groundwork for a 5G roadmap Qorvo 5G band coverage The 600-MHz Band 71 band, which will be supported by the United States in the near future, will be the only vendor that can support all 5G-band bands up to 39GHz at 6GHz and below, following the expansion of the infrastructure portfolio.
The Qorvo QM19000's exceptional performance includes support for 3.3-4.2 GHz wideband operation, 24.5 dBm linear power output, 9-dB peak-to-peak CP-OFDM modulated signals, carrier aggregation with QPSK, 16QAM, 64QAM modulation, ; Supports 2dB maximum power back-off at 64 QAM single-carrier 100 MHz operation and 3dB maximum power back-off at 64QAM 4 carrier aggregation at 400 MHz; Supports 2x2 and 4x4 multi-stream transmission. Even at 3.5mm x 6mm The small size of the package, QM19000 integrated high-performance broadband filter can also be a good solution to coexistence interference problems.
At the 20th GTI Industry Summit in London in November and the 2017 China Mobile Global Partner Conference in Guangzhou, Qorvo presented its QG19000 based 5G terminal prototype demo.
5.iPhone PM IC cut the weight of Dialog late seal orders 2019 fear of moving the wind
Apple (Apple) has decided to design its own power management IC strategy, supplier Dell (Dialog) shares fell sharply, and the company has come forward to confirm the news.And the Taiwan Semiconductor related companies revealed that in 2018, such as silicon And other orders to undertake Dialog PM IC used by Apple unchanged, but the supply system changes in 2019, Apple's own design of the PM IC to TSMC after the vote, the IC packaging and testing may turn to ASE, but by then, the integration of silicon Hopefully, the overall semiconductor supply chain in Taiwan is still positive for the help.Familiar semiconductor back-end IC packaging and testing industry, said that as the iPhone PM IC rely on Dialog overweight, Apple has long wanted to reduce the single supply However, in the medium and long term, the IC packaging and testing industry believes that if Apple designs the PM IC itself and casts the chip at TSMC, Sealing and sealing test is still a chance to find Taiwan and other industries on the ASE, the sun and moon in the future with silicon products into a new holding company, a large single package test is still in the hands of Taiwan plant.Familiar with IC packaging and testing industry Revealed that in 2018 most of Apple's production plan at the end, and therefore Apple's own power management IC is estimated to 2019 new models will be imported, while the initial Apple will not completely give up the Dialog chip, but will surely reduce the proportion , I am afraid that the transfer of the back-end IC packaging and testing may be difficult to avoid.Familiar semiconductor IC packaging and testing industry pointed out that this year as a large number of communications chips ASE, silicon products, two closure industry, due to the peak season smart phone market deferred, In the warm-up phase, the operation was less prominent. As for the iPhone X shipments deferred, the operating trend for the quarterly rise, and silicon customers across the Taiwan Strait IC design industry's main customers for the Android camp, facing the mainland smart phones Satisfied with the IC packaging and testing industry, silicon products are still optimistic about the continuous fermentation of 5G, AI, IoT and other fields in 2018, while the mainstream communication chips can keep the basic disk. The monthly consolidated revenue is about NT $ 71.37 billion, down 3.68% MoM and 1.84% MoM. Looking ahead, the market estimated that the performance of silicon products in the fourth quarter may only be flat compared with the third quarter. In November, the combined revenue of about 77.477 billion yuan, down 1.5% .Familiar with IC packaging and testing industry, said the amount of small, high-profit niche products, including NVIDIA, AMD, high-end graphics processor (GPU) chip Flip chip, 2.5D advanced packaging orders continued steady, the other silicon products in 2018 also continued to receive Taiwan-based IC design smart audio chip, game controller control chip package test orders.Industry-related industries are optimistic about the mainland, South Korea fully seized the generation of 5G communications, base stations and other infrastructure construction business opportunities sprang up first, is conducive to Taiwan's IC packaging and testing industry orders. In addition, Huawei and its IC design HiSilicon mobile intellectual AI AI chip acceleration, packaging large single Due to the massive Mate 10 market optimism Huawei is still able to hold the market basic disk, the performance contribution to the peak in the third quarter of this year, have the opportunity to gradually warming. Looking into the first quarter of 2018, IC packaging and testing industry That the silicon products will be a detailed response to the reduction of the traditional off-season work days and other effects, due to the overall smart phone market is more mature, silicon products for next year, covering the bulk of the communications industry related market conditions, for the time being cautious. For specific customer and order situation, is not for public comment. DIGITIMES