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.ASIC AI, the giant can play the game? Not;
Set micro-network message, artificial intelligence chip mainly includes GPU, FPGA, ASIC and class brain chip. In the artificial intelligence age, they each play the superiority, presents the state of blossoming. Now, AI is no longer limited to machine learning, and more new architectures are being developed to run AI systems at a faster rate. NVIDIA, Qualcomm, Intel, IBM, Google, Facebook and other companies are accelerating into the field.
In fact, these devices are not real chips, but rather a system-level package. They typically consist of one or two ASIC based on the latest semiconductor manufacturing processes (16NM and below) with large-scale processing capabilities, and high-capacity ultra-high bandwidth memory (such as the HBM2 stack), all of which are integrated through advanced packaging technology.
AI related ASIC has recently been the market attention, a number of manufacturers such as NVIDIA, Intel, Google and some of the new ventures have been robbed into development, is expected to form a billions of dollar market business opportunities in the future scale. With Nvidia repositioning the GPU into a cloud AI engine role, also determined to drive the ASIC business to follow up, such as Google has now launched the second generation of TPU, Intel after the purchase of Nervana company also made its own Nervana chip, plus a number of former Google TPU staff founder of the new Enterprise GROQ, recently announced in 2018 will launch its own next generation AI chip and so on.
Google TPU, is specifically for its depth learning algorithm tensor flow design, TPU also used in the Alphago system. This year's release of the second generation of cloud TPU theoretical calculation to achieve the 180T Flops, the machine learning model training and operation of a significant acceleration effect, in fact, is an ASIC chip. Choose to do the ASCI customization research and development, on the one hand is the fund is not worried, on the other hand also is the service which the Google needs to provide consideration. Including Google image search, photos, Cloud Vision API, Google translation and other products and services need to use a deep neural network. Google has the need and the ability to develop a dedicated chip and a large scale application (which will cost a lot of research and development).
Nervana was purchased by Intel and plans to launch the first AI dedicated Nervana neural network processor (NNP) by the end of 2017. Nervana NNP is also a ASIC chip, can be extremely high computational efficiency in the depth of learning algorithm training and implementation. Intel abandons the common CPU cache, and changes the chip memory management for specific algorithms by special software, in the hope of elevating the chip's operational density and performance to a new level.
GROQ, a new venture founded by former Google TPU employees, plans to release the first generation of AI chips in 2018. This chip on the standard NVIDIA GPU, is specifically for artificial intelligence to redefine a chip. Official website data show that the chip will be able to operate at 400 trillion times per second, each watt can perform 8 trillion operations. And Google's latest generation of TPU to achieve 180 trillion times per second operation, GROQ chip performance will be Google TPU more than twice times.
MediaTek will also act. MediaTek co-CEO Tsai said that the ASIC layout of MediaTek will play both resources, through the existing team strength, with the customer needs to actively develop the ASIC field. But after all, it is not too much to pick customers at this stage, or to take the overall business growth as a priority.
The first unicorn of the global AI chip, along with the ASIC route.
In fact, due to customization, low power and other benefits, the AI domain ASIC is being increasingly used, driving the ASIC design and rapid market growth. According to a recent survey by Semico Research, the design of the AI voice-activated device ASIC was expected to grow by nearly 20% CAGR by 2021, almost twice times the growth rate of all ASIC design (10.1%) from 2016 to 2021. Last year, the global ASIC shipments grew to 7.7%, and next year, the network of things will be more than 1.8 billion units of the unit shipments. Semico study points out that the main impetus for ASIC growth comes from the growth of industrial and consumer markets, with the growth of many traditional terminal applications slowing down due to market saturation coupled with reduced demand, while applications related to IoT are taking off.
In addition to IoT and artificial intelligence, Semico said, the growth rate of ASIC products related to smart grids, wearable electronics, solid-state hard drives, drones, industrial networking, advanced driving aids (ADAS) and 5G infrastructure is expected to be more rapid in the broader marketplace. Prior to 2021, the basic SOC design project in the Consumer electronics field will grow at 19% CAGR, while the industrial networking ASIC design project will grow by 25%.
ASIC design and development cost is difficult to load, design service bath Fire Rebirth
However, the application of ASIC in AI field still faces many problems.
ASIC is an integrated circuit tailored to the needs of the product and is designed and manufactured by specific user needs and specific electronic systems. In general, ASIC has been specifically enhanced for specific functions, can be designed according to the needs of the complex, can achieve higher processing speed and lower power consumption, relatively, ASIC design, manufacturing costs are very high. It is difficult for general IC companies to take the risk of developing a dedicated processor ASIC chip for deep learning. First of all, the future of the best semiconductor process for performance must be used, and now the latest technology to make the chip one-time cost of millions of dollars, very expensive. Even if you have money, you need to build a team from scratch design, the design cycle is often very long, can say time to market too long, the risk is very large. And ASIC chips will have to be upgraded to keep up with new technology and process levels, and ASIC chip designers have fixed their logic early in the development process, so if new ideas arise in the fast-evolving areas of AI, ASIC chips will not be able to react quickly. If the scale of the application can not be achieved, even if the success of the development of the United States dollar practical value. Therefore, IC companies tend to adopt general-purpose chips such as CPUs, GPU, or semi-custom chip FPGA.
Obviously, with the continuous upgrading of the process, the ASIC's streaming cost has raised the ASIC minimum chip sales, to the end of the world, a few ASIC manufacturers can withstand the huge ASIC cost and failure risk. In this way, ASIC design services return to the industry focus.
For example, the United States amorphous wafer factory FinFET class ASIC design services Enterprise Esilicon recently announced that the success of its own in-depth learning ASIC sent to the manufacturing, Esilicon mentioned this ASIC acquisition of custom IP, Advanced 2.5D packaging technology and one of the industry's large chips, And for the company's first use of the TSMC 2.5D COWOS Packaging technology production chip. Mr. BJ Woo, vice president of TSMC's business development, said that the Cowos packaging technology is designed to meet the needs of this type of chip design depth learning application, this advanced package solution can achieve high-performance and integration requirements to achieve Esilicon design goals. The aforementioned Google TPU, Nervana NNP, GROQ will launch the first generation of AI chips, are sent by the ASIC manufacturing, and to TSMC manufacturing.
At present, the development of AI-like ASIC is still in its early stage. The basic reason is that once the ASIC design and manufacture of the circuit is fixed, only fine-tuning, can not be greatly changed. and hardware research and development design and production costs are very high, if the application of the scene is not a real market is not clear, the enterprise is difficult to rashly try. In addition, the company that can design the artificial intelligence chip necessarily is to have both the artificial intelligence algorithm and excels in the chip research and development company, enters the threshold higher. Therefore, AI algorithm +asic design services + OEM business model development, can let more and more AI ASIC to come out and develop.
On behalf of the factory here, there are many generations of factories around the world, but because the difficulty is too high, can do AI System package manufacturers are not many, TSMC, Samsung and the core are listed in the list. So, which vendors are designing the AI System package? You need to see which vendors are really good at 2.5D integration and have the key IP needed for design (such as HBM2 physical layer interface and high-speed serdes). The HBM2 PHY and the high-speed SerDes module perform mission-critical communication between multiple components within the encapsulated system. These are very demanding challenges in the simulation design, and buying IP from an ASIC vendor can minimize the risk. The three key technical Esilicon mentioned above are all involved. There are not many ASIC companies that specialize in these areas, but these ASIC companies will benefit because of the potentially explosive growth in the artificial intelligence market.
There are also many IC design services companies in Taiwan, driven by the artificial intelligence market, the ASIC business is well again, and the industry expects the boom to last for a long time. TSMC boldly predicted that 2020 High-performance computing (HPC), AI-related chip performance will be as high as 15 billion U.S. dollars, creativity and the world core, intellectual original also optimistic about the demand for ASIC chips from the global AI customers, is expected to continue for a good while from 2017, especially emphasis on advanced process technology, Extremely complex chip design, high efficiency and low power consumption will be the new blue Sea market for IC design service providers.
2. The industry's first support Nb-iot R14 specification chip, MediaTek release dual-mode MT2621;
Set micro-Network news, MediaTek released the industry's first support for Nb-iot (narrowband Networking) R14 specifications of the dual-mode networking chip (SoC) MT2621. The chip supports Nb-iot and GPRS two network models, has excellent low-power and cost advantages, will bring more rich networking applications, such as fitness tracker and other wearable equipment, internet security equipment, smart meters and a variety of industrial applications.
MT2621 is a highly integrated networking platform, in addition to supporting the Nb-iot network, but also compatible with the existing GPRS network, for the IoT equipment to provide excellent network coverage and call quality. MT2621 built-in power and connectivity can support the existing GPRS network and the future of Nb-iot development, will be able to drive things network into the next stage of growth.
In order to help terminal equipment manufacturers to push products into the market faster, MT2621 uses simple and cost-effective single SIM card antenna design. Terminal equipment only a general-purpose integrated circuit card (UICC) and mobile number, you can freely switch between Nb-iot and GPRS dual network, and support single double stay. In addition, MT2621 integrates a broadband front-end module, which supports all frequencies of ultra-low frequency, low-frequency and intermediate frequency in the global range.
MT2621 is powered by a power-saving arm V7 microcontroller (MCU), built-in flash memory and pseudo static RAM (Psram), also supports a variety of audio and video interface, built-in Bluetooth 4.2 communication features can be connected with other nearby devices freely.
3. How is MediaTek in the Nb-lot market ' waiting for the Wind '?
Set micro-mesh/Wen Dunwen label
If you want to find a bright spot at the 2017 China Mobile Partner Conference, then the ' 139 cooperation plan ' is absolutely not negligible. Here the ' 1 ', that is, China Mobile will build a leading technology, covering a wide range of mobile networking, in 346 cities to achieve nb-iot continuous coverage, for the vast number of users with convenient, fast Internet services experience.
MediaTek, the chip maker that pushed domestic nb-iot applications to the ground, is on the move, releasing the industry's first dual-mode networking chip MT2621, which supports Nb-iot R14 specifications, at the China Mobile Partner Conference.
Today in China Mobile Partner conference, heard most is 139 cooperation plan, I think this is the layout of the Nb-lot environment, China is very good opportunity in the Nb-lot and 5G of the era can be ahead, become the world's fastest growing place. Zhenger, general manager of MediaTek's home entertainment products group, said the dual-mode Nb-lot chip MT2621, today's MediaTek, has a low power and cost advantage, and will achieve mass production in the first quarter of 2018.
The first Nb-iot chip MT2621 supporting R14 dual-mode
I believe the industry knows that the motorcycle is MediaTek's 2G chip, through GPRS positioning, take the GSM network, rather than Nb-iot chip.
In fact, shared bicycles should be a typical case of Nb-iot application, but since the Nb-iot definition was in R13 's communications standard, it could only do stationary transmissions, so it's hard to use nb-iot like a shared bike. However, the latest R14 standard support device mobile, MediaTek first cut into the R14 standard, will give Nb-iot application to bring more possibilities.
This time, MediaTek released the industry's first support Nb-iot R14 specifications of the dual-mode networking chip MT2621, support for Nb-iot and GPRS two network models, with low power and cost advantages, can bring rich Internet applications, such as fitness tracker, wearable equipment, internet security equipment, Smart meters and a variety of industrial applications.
As you can see, MT2621 is an integrated platform for networking, which is compatible with the existing GPRS network, in addition to supporting Nb-iot networks. In the low power and universal connection of Nb-iot applications, MT2621 is driven by a power-saving arm V7 microcontroller (MCU), built-in flash memory and pseudo static RAM (Psram), also supports a variety of audio and video interface, built-in Bluetooth 4.2 Communication functions are free to connect with other nearby devices.
In product cost and application maturity, MT2621 uses simple and very cost-effective single SIM card antenna design, terminal equipment only a general-purpose integrated circuit card (UICC) and mobile number, you can switch between Nb-iot and GPRS two networks freely, and support single double stay. In addition, MT2621 integrates a broadband front-end module that supports all the ultra-low-frequency, low-frequency, and intermediate frequency bands around the world to help terminal manufacturers push their products faster into the global marketplace.
According to the visitors Jay, MT2621 will be in the first quarter of 2018 to achieve mass production, the fastest in the second quarter next year to achieve product listing. MT2621 built-in power and connectivity function can support the existing GPRS network and future Nb-iot development, can bring animal networking into the next stage of growth.
While China Mobile launched the ' 139 Cooperation Program ', MediaTek introduced the Nb-iot R14 specification of the dual-mode networking chip MT2621, mainly considering China Mobile to achieve a Nb-iot continuous coverage deployment cycle in 346 cities. Visitors Jay said the launch of the dual-mode MT2621 chip is due to the operator base station layout takes time, the estimated time of this layout may take two time, of course, in addition to China Mobile, MediaTek is also in line with China Telecom/Unicom in the Nb-iot business application needs.
Nb-lot market will grow fast in the second half of 2018
In fact, in June this year, MediaTek launched its first Nb-iot chip MT2625, and joined hands with China Mobile to create a small size (16mm x18mm) of the Nb-iot Universal module, for IoT equipment to provide low-power, low-cost solutions, is gradually realized commercial.
See MediaTek Nb-iot chip to standard support, is ahead of the operator's deployment rhythm. Visitors Jay said that the reason for the introduction of R14 Nb-iot chip, is because of the already formed 2G network applications, will be replaced by Nb-iot. Nb-iot transmission speed will be faster, and its low power characteristics, can make water meters, meters and other equipment television life in years, greatly reducing the cost of battery replacement.
At the same time, in wearable equipment, it needs to do a small chip, but also to save electricity, while Nb-iot can meet this demand, and support the movement of wearable equipment, and Nb-iot R13 standard is not supported.
In addition, due to the latest support R14 standard chip support device movement, for some mobile speed less than 90 km/h devices, can be installed Nb-iot chip tracking. such as pet tracking, valuables tracking, logistics tracking and so on, the scope of application is more extensive.
' Just replace the 2G with Nb-lot, this will be quicker, and we will also choose the target customer to support us in the future. ' Visitors Jay said, formerly MediaTek is to push the handset chip scheme, but the Nb-lot play with the previous business is not the same, because the application surface is too wide, also includes the market and the middle tier, we call the system integrator, this will involve many different special specifications, by the system integrators to complete, MediaTek only provides the core chip.
Overall, the Nb-lot market is still in its infancy, and product applications will be dominated by single mode. However, with the Nb-lot market business process accelerated, Single-mode is the transition to two-mode, dual-mode will be nb-lot future market trends. Based on the advantage accumulation of 2G chip application, MediaTek can quickly realize the transition transfer to Nb-lot chip application.
' The Nb-lot market is expected to see more rapid growth in the second half of 2018. ' Visitors Jay said that this year, China Mobile's access port is about 200 million, estimated 2018 into 320 million, in the increase of 120 million inside, next year may be a double mode will take the majority, the single mode will also develop rapidly. Now Nb-lot business model of the ecological chain is still in the integration, to increase the formation and maturation of the ecological chain will take some time. And MediaTek has been the Bureau, and so on just ' wind '.
4. Bo Tong and Qualcomm have killed the competition of the specter of the strong resistance of regulators
Braodcom's takeover proposal was rejected by Qualcomm (QUALCOMM) at too low a price, and now it is considering raising the prices and even launching a proxy battle to achieve the goal, but even if the Bo pass smoothly, then I am afraid to step on the supervision of landmines. The combination of Bo Tong and Qualcomm will create the third largest semiconductor company next to Intel and Samsung Electronics (Samsung Electronics), a result that is enough to wrinkle the anti-Tottenham enforcers. The Wall Street Journal report notes that the focus of regulators ' concerns on the purchase and plan will be on the dominance of the merged company in key 0 components of smartphones. Qualcomm is the wireless device Wi-Fi and Bluetooth chip largest supplier, according to Gartner Statistics, Qualcomm in these two chip market total control 60% of the Territory, Bo Qualcomm and Qualcomm combined in the GPS chip market will be up to 52% of the city, The two are in a competitive state (which will complement each other in the case of a combination of wireless communications), which is currently dominated by Qualcomm's mobile communications chip market, totaling 59%. As a result, the two merger plans are likely to suffer regulatory resistance in Europe and the mainland, in addition to the US. The US Department of Justice recently dismissed at's $85 billion takeover of Time Warner as a message that even the Warner Republican law enforcers would intervene with the big buy-outs of together. Scholars also believe that the combination of the two unfavorable market competition. Sascha Meinrath, a professor at Penn State University in the United States, says that Pennsylvania and Qualcomm's combination are bad for consumers, hardware makers, software developers and service providers. Such a huge enterprise will have the ability to control the development of wireless hardware. Harry University, a professor at New York University, says U.S. regulators are paying particular attention to companies that package complementary technologies because it could stifle small competitors. Eleanor Fox, an expert on international antitrust Law at New York University, believes that in Europe and the United States, the combination of Bo and Qualcomm may be required to have collateral conditions, such as the sale or sharing of intellectual property rights. For example, European regulators conditionally approved the 130 billion-dollar merger of Dow Chemical (Dow Chemical Co) and DuPont (DuPont) in early 2017 because the latter agreed to sell a research and development centre. On the other hand, the mainland government has often used antitrust rules to prop up local companies, and in 2015 The local regulators argued that Qualcomm had to cut tariffs on the mainland market in response to Qualcomm's licensing fees, which experts said would make a significant profit for its hardware makers. Boton has expressed confidence that once the Qualcomm deal is finalised, regulators will be approved within 12 months, but the situation may be much more complicated. Bo Qualcomm and Qualcomm have encountered a strict audit of the case, of which Qualcomm spent $39 billion to buy the NXP (NXP semiconductors) also card to the European Commission (EC), Bo Tong 5.5 billion U.S. dollars to Brocade communications (Brocade Communications Systems Before the completion of the takeover, it had been shelved for possible concerns about U.S. national security. Digitimes