According to reports from Taiwan Media, Formosa Plastics' president Wang Wenyuan took over as chairman of the Taiwan Federation of Industry and Industry Corporation of the United States yesterday. When talking about the current brain drain of the mainland, he mentioned that the situation of semiconductors and DRAM is more obvious. In the past one or two years, there have been reports of high-level semiconductor technicians in Taiwan being attacked by land companies. South Asia Branch also lost 48 high-end technical personnel, but 'had no major influence on the South Asian Branch,' Wang Wenyuan said. The DRAM factory Huayuki is relatively large, running about 400 people.
In regard to the impact of the 31 measures taken by the Taiwan Huitai Group on brain drain, he pointed out that he did not know the specific situation, but if Taiwan lacks opportunities, 'self (talents) will not move to the mainland if they do not have enough opportunities,' or choose to go for high salaries. Land development.
Lu Chaoqun, chairman of Chuangchuang, said that the wages previously paid out in renminbi directly to the Taiwan dollar were now more targeted at the top individuals and directly pay the monthly salary of the US dollar. This powerful siphon force will not be conducive to the development of the semiconductor industry in Taiwan.
Lu Chaoqun called for Taiwan to be more active in formulating relevant policies and to use commercial methods to plan for talent gathering, talent retention and talent cultivation, to make talents willing to stay in Taiwan, and continue to contribute to the development of the semiconductor industry. He cited the past technology industry as a result of stock dividends. Therefore, it is possible for companies to have strong retention tools. Nowadays, the mainland has completely compared this practice and even resorted to high salaries.
Lu Chaoqun said that the flow of talent depends entirely on the strength and policy of economic development. Taiwan’s semiconductor industry is no longer five. Now that it only competed with the United States and Japan 60 years ago, it is now facing global competition, if not talents. Stay in Taiwan, let others have strength over Taiwan.
In all industries, Wang Wenyuan named the semiconductor and DRAM, and thinks that the situation of brain drain is evident in the mainland. He pointed out that although the Formosa Plastics Corporation’s South Asia Branch has cases, he knows there are 48 high-end technical talents, but it has not had much impact. Hua Yanke, the loss of talent from the United States is more obvious. 'He said that although it was expected, but did not expect that people will go so much. It is high-end technical personnel who have gone to the mainland in the past two years. Fortunately, the South Asian Section has not been greatly affected. The South Asia Section has sufficient talents. On the contrary, the Asia Pacific Branch will need to adjust.
For this phenomenon, Taiwan has expectations, but it is not expected to be so obvious. 'I don’t know if I’m going straight to inquiries. Most of them are Inako, and the loss of talent from the United States is obvious. This is expected, but I did not expect it. People will walk so much. 'Wang Wenyuan added.
2. South Asian Branch: DRAM demand is hot until Q3 Q4 to observe changes in capacity;
According to the micro-message message, Taiwan Semiconductor Memory Chip Factory South Asia Branch held a law meeting yesterday (17th) and announced that gross profit margin rushed to 51.8% in the first quarter of this year, which was better than market expectations. Li Peiyi, general manager of South Asia Branch, analyzed the gross profit margin of the previous quarter. High, mainly due to DRAM prices and 20-nanometer products, the proportion of rapid increase, but due to holding more than 40 billion equivalent US dollars in the hands of the quarter, with the NT appreciation of about 2.6% in the previous quarter, resulting in 1.178 billion yuan exchange loss, erosion Earnings per share of more than 0.3 yuan.
Li Peixi said that in the first quarter, the average quoted price of DRAM increased by 6.1%, plus sales volume increased by 8.4% in the quarter, sales ratio of 20nm accounted for more than 60% and R&D expenditure deferred. %, Quarter increase by 2.1 percentage points.
As for the future trend of the DRAM market, Li Peiyi expects that supply will be in short supply until the third quarter. In the fourth quarter, it will be necessary to closely monitor the increase in production capacity of the major manufacturers. At present, the two major manufacturers of Samsung and SK Hynix will regard the market demand. It is expected that this year's DRAM market conditions will be healthy and stable.
At the same time, South Asia Branch announced that it had revised its DRAM sales volume for the full year this year. Its annual growth rate has increased from the original 45% to 48%, and this quarter's sales volume has increased by more than 15% quarter-on-quarter; 20-nanometer production capacity has shifted slightly by 1,000 units. The 30-nanometer product has reduced monthly production capacity from 38,000 to 37,000 units; 30-nanometer monthly production capacity has increased to 31,000 units. At the same time, the self-developed 10nm is planned as scheduled, and it is scheduled for small-scale production in 2020.
In response to the fluctuation of the Taiwan dollar, South Asia Branch adopted a strategy of dynamic balance between US dollar and NT$ cash. He expects this season to be reduced due to exchange rate factors. It is optimistic that the average unit price of DRAM in this quarter is expected to increase by one-digit percentage, plus 20 The contribution of nano-manufactured products continues to increase and the cost decreases. The gross margin has the opportunity to continue to write new highs, and has the opportunity to go all the way up to the third quarter.
Talking about the impact of the Sino-US trade war, Li Peixi believes that this is a means by which both China and the United States want to force each other to the negotiating table. If a consensus is reached, China will pay more attention to intellectual property rights, which will bring positive effects to the DRAM industry. If the results are not as expected, they may impact the global economy, and thus affect the weakening of terminal demand. Not only the DRAM industry, but also all industries.
3. Accelerating Cloud Release A complete set of FPGA-based AI acceleration solutions. The CEO has just said that AI does not play virtual;
Micronet collection April 17 report
This afternoon, Accelerating Cloud held a conference to “accelerate new technologies and drive smart for the future” in Beijing and formally launched its four major innovative products and three major solutions. This series of solutions can meet the ever-expanding demand for data and models, and help in depth. The learning model works efficiently.
'Acceleration Cloud', the full name of Hangzhou Acceleration Cloud Information Technology Co., Ltd., is a leading manufacturer of heterogeneous acceleration and service offload solutions. It is committed to providing integrated solutions to accelerate heterogeneous computing. With their leading technology in FPGA, they are widely recognized in the industry. .
Launched two system hardware acceleration products, two IP libraries, three solutions
Unlike many existing trainings that use GPUs to complete deep learning, Accelerated Cloud's products are FPGA-based deep learning acceleration programs. They believe that the use of GPUs to perform deep learning training has exposed many problems, such as high power consumption and performance. Power consumption ratio is poor, processing delay is large, it is not suitable for large-scale reasoning deployment and some delay-sensitive scenarios.
But the FPGA is different. It has a high performance to power ratio, and based on the gate-level circuit design, FPGA is an ultra-low latency and deterministic delay. The FPGA's programmability and dynamic reconfiguration can adapt to deep learning. Future changes in the algorithm, IO programmability can meet more business needs (network acceleration, edge computing), so FPGA-based deep learning program has become the future direction of technology development.
In this conference, Accelerated Cloud introduced a complete set of FPGA-based AI acceleration solutions, including four major products: two series of hardware acceleration products (SC-OPS, SC-VPX), two IP libraries (FDNN, FBLAS ), Three major solutions (deep learning solutions, high performance computing and digital signal processing solutions, edge computing solutions). Specifically look as follows:
On hardware products, SC-OPS is the world's first Intel Stratix 10 FPGA accelerator card launched by Accelerated Cloud. It uses Intel's latest 14nm process Stratix 10 GX2800 FPGA device, which can be widely used in data centers, cloud computing, machine vision, deep learning, high Performance calculation, simulation, finance and other fields.
In another, SC-VPX is the world's highest computational density VPX blade acceleration platform, using Intel Stratix 10 GX2800 devices, compatible with GX1650, constructing industry-leading, flexible, efficient signal processing and deep learning architecture, and positioning the communication of institutions such as universities , rapid prototyping and algorithm development and application of deep product learning related fields.
In terms of the IP library, the deep learning acceleration library FDNN is the first FPGA acceleration library supporting general convolutional neural networks. Based on RTL-level code, it can provide high performance and flexible configuration features. High-performance computing acceleration library FBLAS is the industry's highest Performance RTL level math acceleration library.
The three solutions include: Deep Learning Acceleration Solution - Accelerating Cloud Introduces a Complete FPGA-based Deep Learning Acceleration Program; Digital Signal Processing Solution - Requirements for Digital Signal Processing Systems such as Communications, Combining Intel's Latest 14nm Process Stratix 10 FPGA Family Accelerated Cloud provides a complete set of hardware and software solutions; Edge Computing Solutions - Accelerating Cloud Intelligent Control Solutions.
Accelerating Cloud founder and CEO Yu Gang said: 'Artificial intelligence has entered our lives, but there are still bottlenecks in the future development, and we need breakthroughs in hardware technology and algorithms. Heterogeneous computing is the future trend of computing architecture, while FPGAs are different implementations. The perfect choice for structured computing. Heterogeneous computing accelerates cloud innovation Accelerated platform solutions, with high performance, high efficiency, low latency and programmability and remote reconfigurability, are ideally suited to the needs of resilient services in the cloud. We hope that through our technology, we can help more companies achieve deep learning and win opportunities in the era of big data.
CEO Yu Gang: AI does not play imaginary, does not intend to pack the IP package
Accelerated cloud is a star enterprise in the field of domestic FPGA. After the company was established in 2015, with deep technical strength, it took the lead in implementing the FPGA-based deep learning algorithm, cloud computing acceleration algorithm, and applying it in the system. High performance-to-power ratio, lower construction and operating costs, becoming a leading provider of next-generation data center acceleration solutions.
Figure: Accelerating Cloud CEO
In fact, accelerating the heterogeneous computing acceleration of the cloud platform is not a simple board provider, but rather provides an acceleration solution from the bottom of the hardware. Accelerating the cloud CEO said that the company’s core technology is to have two IP libraries. : Deep Learning Acceleration Library, High Performance Computing Acceleration Library. Based on deep understanding of FPGAs and technologies, the company is fully supported by customers such as Intel, and can even get Intel's first in-house test chips.
At present, many domestic companies with similar IP technologies will choose to encapsulate IP into AI chips in order to obtain a higher valuation. In this regard, Luan Gang believes that if it is not worth the cost of expensive tapeouts just for valuation. He said that we do not play with AI, and currently accelerating the cloud will not take IP out of the film. In the future, we will continue to provide customers with accelerated solutions in the form of board cards.
It is reported that Accelerating Cloud and Tencent, Jingdong, Midea, Intel, HKUST, Jinshan and other enterprises have in-depth strategic cooperation and provide them with long-term technical service support. In addition, they also cooperated deeply with many domestic universities and established joint experiments. Room, also set up R&D and business center in Hangzhou, Wuxi, Nanjing, Xi'an, Beijing.
In terms of financing, in 2017 the company won a prestigious investment institution, Dachen Venture Capital, Zhenge Fund, and tens of millions of yuan of pre-A financing of Rushan Capital.
4.In the AI field, the global semiconductor output value is expected to grow by another 10% this year;
Gathering micro-messages, According to Taiwan media reports, at the VLSI-TSA and VLSI-DAT seminars held at today's International Large-scale Integrated Circuit Technology, Systems and Applications Symposium (VLSI-TSA), Memory Manufacturers Long Lu Chaoqun said that due to the strong market demand, the outlook for the current semiconductor industry in 2018 is optimistic. It is expected that the global semiconductor output value is expected to grow by another 10% this year (2017), which will be higher than the industry’s previous expectations of 7 %.
For the development of the semiconductor market in the future, Lu Chaoqun pointed out that under the stimulation of the AI field, semiconductor business opportunities will blossom everywhere. Not only does it require advanced processes, but even special process requirements such as 28 nm and 90 nm have the opportunity to grow. Observations currently include silicon wafers, The out-of-stock condition of memory and passive components all showed strong demand in the downstream market. He estimated that this year's semiconductor output growth is expected to be better than expected and will reach a level of 10%, which is also better than the industry's growth in semiconductor output this year. 7% growth forecast.
5. MediaTek and Microsoft join hands to promote IoT innovation and security;
On July 17th, MediaTek announced that it has partnered with Microsoft Corp. to lead the industry's first system-on-a-chip (SoC) MT3620 supporting the Microsoft Azure Sphere IoT operating system, providing built-in security and networking capabilities. MCU-type IoT products that promote IoT innovation and security. Microsoft Azure Sphere is an operating system designed to develop highly secure MCU-related devices that enable companies to deploy a wide range of cloud-based devices. Security mechanism. It is reported that MediaTek has sent samples to the main cooperative customers. The terminal products will be available in the third quarter of 2018. MediaTek MT3620 is a chip jointly developed by MediaTek and Microsoft and is applicable to the Microsoft Azure Sphere Internet of Things operating system. Wi-Fi networking control is built into the processor chip and loaded with Microsoft's latest security protocol. The MediaTek MT3620 chip will be sold with the Microsoft Azure Sphere total solution, allowing many IoT device developers to easily connect The MCU IoT device it built, and ensure the security of the device under the security framework provided by Microsoft. MediaTek Deputy General Manager of Technology and General Manager of Smart Device Business Group Zhan Jie said: 'Lianfa Technology has a long history of cooperating with Microsoft. We are cooperating to develop special-purpose chips to meet various networking needs. MediaTek has leading position in chip design and In close cooperation with Microsoft, and more importantly, we agree with Microsoft's vision and work together to bring security and trust into the overall IoT application from the terminal chip to the cloud. Device manufacturers can use MediaTek chips and Microsoft operating platforms to create a variety of All kinds of networking equipment are sold and consumers can manage these products properly. 'The cost of networking has dropped dramatically, meaning that billions of devices will be able to connect to each other and be controlled by MCUs. From home appliances, Health monitoring equipment to children's toys and industrial equipment are all part of the continuous growth of the Internet of Things. The more connected nodes there are, the more ways that they may be invaded. This means that companies and individuals must trust their equipment, data, and cloud. It is indeed safe and secure to take advantage of these convenient networking devices. As a result, MediaTek and Microsoft have joined hands to promote the market. From stand-alone MCU devices to secure cloud-connected MCU devices and applications. Galen Hunt, general manager of Microsoft Azure Sphere, said: 'Computer computing is becoming more powerful and ubiquitous in the cloud or in the terminal. We will see the next decade One billion devices connected to the network. Azure Sphere brings together Microsoft's expertise in cloud computing, software and silicon, providing unique security solutions from the chip to the cloud. MediaTek and Microsoft work closely together on chip design, and we're happy to be together Announced that the first MT37020 chip with Azure Sphere certification will enter the market. Microsoft and MediaTek will jointly provide a solid foundation for the next generation of smarter, more secure products and solutions for the Internet of Things. 'More than 9 billion built-in systems are built each year. MCU's IoT devices are entering the market, so industry leaders like Microsoft and MediaTek must work together to ensure the highest level of built-in security, regardless of expensive or low-cost networking equipment. Through Azure Sphere , Microsoft is building a new ecosystem consisting of multiple chip suppliers, ODMs and OEMs. Participating partners understand the opportunities and risks that come with the increasing number of IoT products and applications, and thus work together to confirm the establishment and maintenance of safety standards.
6. Qualcomm wants to dig the corner of Wintel? Microsoft: We are pursuing ecological diversity;
According to recent industry sources, Qualcomm will enter the field of PC computing, and Intel’s Wintel alliance with Microsoft will be affected. Apple may no longer like it. However, according to foreign media TechRadar interview with Erin Chapple, general manager of Microsoft Windows. Reports, even if the first Always Connected PC equipped with Qualcomm chips, does not mean that the company will abandon the cooperation with Intel.
In other words, even if Microsoft has made a big resignation on 'ACPC' Windows hardware, the company will also hedge against risk by collaborating with Intel, Qualcomm, AMD, and even some local Chinese manufacturers. Erin Chapple stated: 'We haven’t Always connected to PC's and Qualcomm's equator, our company will choose from the ecosystem and work with partners.'
Although there are many news reports focusing on the recent release of Qualcomm-based ACPC devices, such as HP's Envy x2, Asus's NovaGo, and Lenovo's Miix 630, but don't forget that Microsoft's own Surface Pro LTE also uses Intel's processor.
After all, Microsoft firmly believes that a diversified chip ecosystem will benefit the continuous development of ACPC devices: 'We will not stop at laptops, two in one, or tablets, but we will be dedicated to networking every device in different SoC architectures. '.
Finally, Erin Chapple stated that Microsoft will continue the development of the Windows 10 S operating system, and that the fall of Windows 10 creator updates will bring significant performance improvements to ACPC devices. 'Compiled from: OnMSFT' cnbeta
7. Intel's new manufacturing process is behind. In 2021, it was overtaken by TSMC;
According to the micro-network news, the research institute Linley Group released a report on the chip industry, pointing out that Intel (Intel) is lagging behind in the development of a new process, and its original chip manufacturing technology advantages are almost gone.
Linley Gwennap, chief analyst, wrote that Intel's delay in the 10-nanometer process has caused the sprint of the next-generation process not only to lag behind TSMC, but Samsung and its partner GlobalFoundries are also preparing to overtake.
It is reported that TSMC plans to mass-produce 7nm FinFET chips in June this year. TSMC will then achieve a 100% market share of 7nm chips. Qualcomm, Hass, and Xilinx are major customers of TSMC, and the benefits of the 7nm process will be in The second quarter began to reflect.
Samsung has already completed the development of 7nm process technology and used extreme ultraviolet exposure equipment (EUV) in this process technology. Initially, the company expects that the development of 7nm process technology will be completed in the second half of 2018, but it will catch up in the future. TSMC’s pace has now been completed six months ahead of schedule. It is reported that Qualcomm is preparing to send its new mobile chip sample to Samsung.
The report analyzes the chip circuit density produced by major manufacturers and other process technology indicators. The conclusion is that the level of technology used by other competitors has approached Intel. For example, there is almost no gap between 7-nanometer TSMC and 10-nanometer Intel. The report points out that the three major Competitors are procuring next-generation lithography devices more aggressively than Intel's, and it is not ruled out that the three will surpass Intel in 2021.
Intel Moat was attacked by opponents one by one, and the outlook for future revenues is worrisome. Recently, news broke that Apple will replace Intel processors with some of its Mac computers with its own R&D processor in 2020 at the earliest, threatening Intel's revenue.