Episode News (Text / Xiaobei) Texas Instruments (TI) said on Tuesday that Brian Crutcher resigned as president, CEO and TI board member, the board has appointed Rich Templeton as the new CEO, and Rich Templeton will continue to serve as TI director Long. TI emphasizes that Rich Templeton is not a temporary CEO, will serve for a long time, and the board will not find alternatives.
Brian Crutcher's resignation was a violation of the company's code of conduct. TI said Crutcher's violations are related to its own behavior, which is inconsistent with TI's ethics and core values, but not with TI's strategy, operations and financial reporting.
Mark Blinn, Chairman of TI's Board of Directors, said that TI's core values and codes of conduct have been the foundation of our operations and behavior for decades, and we have a zero-tolerance attitude toward violations of the Code. Over the past 14 years, Rich has conducted TI on TI. A successful leader. TI is confident in Rich's values and ability to continue to lead the company's advancement.'
Rich said that I am proud of TI and are keen to make it stronger.
In January of this year, TI announced a coaching change. Rich Templeton, a former CEO for 14 years, will step down. Brian Crutcher, who has served for 22 years, will replace Templeton as CEO, and Crutcher will take office on June 1.
In a short time, Rich Templeton returned to the old seat.
Rich Templeton is a 'veteran' of TI. In 1990, he joined TI after graduating with a degree in electrical engineering from Union College in New York. From June 1996 to April 2004, he served as executive vice president and semiconductor business unit. President; served as TI COO from April 2000 to April 2004; served as president and CEO in May 2004; served as chairman of the board in 2008; he has served on the company's board of directors since July 2003.
According to industry insiders, Rich Templeton understands the semiconductor strategy because of his many positions. Under his leadership, TI has demonstrated stronger technology and product positioning in its core business. (Proofreading/Spring/Summer)
2. Another 500 million US dollars! Why is this unmanned car company so fierce?
The micro-network news (text / Emil) The unmanned vehicle field seems to be no shortage of money, this is not, the San Francisco-based unmanned car company Zoox has a total of 500 million US dollars, the lead is the co-founder of the office collaboration company Atlassian Michael Cannon-Brookes, and China VC Chunhua Capital.
Valuation of $3.2 billion
At this point, Zoox's accumulated financing amount has reached 800 million US dollars, with a valuation of 3.2 billion US dollars. Looking at the world, it is also considered the first array of players in autonomous driving startups.
In this new financing announcement, Zoox also disclosed the latest developments of some companies: Zoox is a 500-person company, and plans to launch commercial unmanned vehicles in 2020; Zoox does not sell unmanned vehicles as a business model. It will deploy its own unmanned taxi team directly; the operation and deployment will first be fixed in the scene, and then gradually promoted; the current Zoox has been able to drive from Stanford University to downtown San Francisco, about 65 kilometers, with suburbs, high speed, Multiple scenes such as urban areas.
However, Zoox founder Tim Kentley-Klay also said: The new round of financing will also be opened, will continue until the end of next year. But did not disclose how much money to integrate in a new round, and where to use.
Sword slanting Zoox
Although Zoox is not familiar with the face, when Apple's stock price hit a new high, 17 Apple unmanned vehicle engineers resigned and joined an autonomous driving startup, Zoox. Zoox was founded in 2013, its founder It is the Australian designer Tim Kentley-Klay and Jesse Levinson.
Zoox is a bit of a 'sword to the slant', the idea is to create an unmanned taxi ecosystem, that is, what Google+Lyft wants to do, and Uber and Didi are all focused on the direction.
In terms of specific technical implementations, Zoox has to take a road that relies on Lidar to achieve a completely unmanned L4 level and redesign the car itself. From the previously exposed car body picture, the Zoox unmanned car was designed as 4 Seat, no steering wheel, throttle and brakes, it doesn't seem to be the tail of the car – there is no worry about turning around.
The most prominent sensors are the Velody laser radar mounted on the front, rear, left and right corners, and a GPS antenna on the roof. There are also a number of radars placed at the wheel position, as well as sonar sensors.
For this 'hobby gambling', the industry seems to have reason to expect.
Domestic VC favors
It is worth mentioning that Zoox has long been known in the VC circle in China.
In addition to the early investor Defengjie Fund, this time led by Chunhua Capital, Zoox also received the blessing of Tencent, IDG and other Chinese capital.
It is said that after Tencent tried to take a group of Silicon Valley unmanned car companies, it finally compared with Zoox and Drive.ai, but the specific investment details have not been disclosed.
IDG Capital participated in Zoox's Series A financing in 2016. IDG Capital's global presence in the new travel sector began in 2016. In the high-level autopilot field, IDG Capital first invested in the world's leading Level 4 fully automated All-Star team : Zoox and Pony.ai, now the two companies are the most profitable and highly valued unmanned car unicorns in the US and Chinese startups. (Proofreading / Xiaobei)
3. Nokia pushes its own 10nm ReefShark chipset, 5G deployment cost reduction
In recent years, Nokia has actively developed various 5G related application businesses and actively participated in the development of standards to assist in the development of 5G connection technology in smart ports, car networking, smart factories and other application fields. To this end, Nokia has introduced its own chips. Group, to achieve miniaturization of base stations and reduce the cost of 5G construction.
Wang Jianya, president of Nokia Greater China, said that 5G's investment in construction will not be higher than 4G, depending on whether the construction method is based on SA or NSA architecture; however, it is conceivable that in the future, for global telecom operators In other words, the proportion of revenue from voice will be less and less, and the proportion of data revenue will increase. The future is absolutely based on data services, including pure traffic services and vertical application services.
With research and innovation from Nokia Bell Labs, Nokia can develop a complete solution from core chips, network architecture to management services. In addition, Nokia is actively collaborating with experts in various industries around the world to create a comprehensive IoT ecosystem. Circle (Nokia WING).
Among them, Nokia's ReefShark chipset developed by 10nm process technology can reduce the number of massive MIMO antennas by half, which not only increases the selectivity of deployment, but also reduces the power consumption of the baseband unit by 64%. 5G massive MIMO antenna is equivalent Huge, after the introduction of the ReefShark chipset, it can successfully reduce the base station miniaturization and effectively reduce the 5G construction cost of telecom operators. The beamforming technology with machine learning function allows the network to 'track' mobile devices and expand base station coverage. In addition, it can provide the required large-scale capacity for any location. In the future, Nokia's 5G related products will use their own development chips to truly grasp the end-to-end development.
In order to open up more innovative application cases, Nokia is also actively involved in the development of 5G standards to assist in the implementation of 5G smart ports, V2X car networking, intelligent industrial automation and other applications. Wang Jianya further explained that the emergence of various innovative services, not only for telecommunications operations The challenge of business is more competition across industries. New Electronics
4. Secret voice interaction: Personalized AI to create 'digital you'
From more beautiful photos, higher security to a more natural user interface (UI), artificial intelligence (AI) is enriching our lives in a seamless way. Energy efficiency and personalization are the Qualcomm AI Research Core research areas, because they are indispensable for the next generation of ubiquitous smart user experience. Terminal-side AI is supporting the development of virtual assistants, and personalization is becoming the main between today's virtual assistants and future true personal assistants. difference.
AI drives voice interaction revolution
Voice is a transformative interaction that we have been looking for. It provides natural interaction between the user and the terminal without any hand manipulation. Advanced voice interaction has always-on, conversational, personalized and private features that can be supported. We conduct efficient and natural conversations. From smartphones, smart speakers to extended reality (XR) devices and cars, voice interaction is growing and popular in many product categories.
To achieve an end-to-end voice interaction experience, which requires the support of multiple components (see Figure 1). The flow of voice interaction is as follows: First, the microphone collects the voice signal, then the voice preprocessing, voice activation, speech recognition and natural language. Understand, in the end, the terminal responds to the user through voice synthesis. It can be seen that end-to-end voice interaction is a multi-step complex process.
Why is voice interaction so popular today? An important factor is that machine learning has ignited a revolution in voice interaction. The accuracy of machine speech recognition is approaching 95%, reaching levels comparable to human accuracy. When machine accuracy cannot At the level of human beings, the entire voice interaction process can't achieve dialogue and natural and efficient communication, so the user experience will be significantly reduced. Machine learning technology can be applied to almost all voice interaction components, thus improving all aspects and overall Interactive experience.
Voice interaction on the terminal side is crucial
Due to computational, memory and power limitations, the main voice interaction components (such as automatic speech recognition and natural language processing) have traditionally been supported by cloud AI. However, terminal-side processing has many unique advantages, including faster response times. Higher reliability and better privacy protection. Especially in terms of privacy, today's consumers are eager to keep their data in their hands and have the right to choose whether to send data to the cloud for processing or storage. So, more Processing on the terminal and supplementing the cloud function is one of the important trends in the industry. The main challenge is to run and implement complex voice interaction functions under the power consumption and cooling limits of the mobile terminal.
Qualcomm is actively addressing this challenge and applying low-power terminal-side AI to voice-interactive components. Here are a few examples:
o Voice noise reduction processes noisy speech into clear speech and output, which is especially important in noisy environments. Qualcomm AI Research is developing a deep noise reduction learning model that has significant performance improvements over traditional methods and has Higher efficiency can be run under such very limited processing power, power consumption and heat dissipation conditions as hearing aids.
o Automatic speech recognition transcribes speech into text. Tests show that deep learning technology can achieve top-level speech recognition accuracy on mobile terminals. Qualcomm AI Research has completed a deep learning acoustic model training and adapts it to each user's Accent and environment, thus increasing the degree of personalization.
Today, voice interaction on the terminal side has become a reality. Qualcomm AI Research has been able to implement overall end-to-end voice interaction processing on the terminal side, and demonstrated the end-to-end voice interaction experience in smart homes at the 2018 CES conference. After adapting to accents and the environment, this demonstration has been able to achieve over 95% intent to understand accuracy for a specific field of instruction set.
Personalized achievement of real virtual assistant
The real virtual assistant has a 'digital me' in the terminal. Therefore, it needs to analyze personal information and continuously learn user preferences, thus becoming a digital assistant who truly understands the user. Situational intelligence is a prerequisite for personalization. To achieve scenario-based intelligence, it is necessary to combine data from multiple terminal-side sensors (such as microphones, cameras, and gyroscopes) and other terminal-side and terminal-out data (see Figure 3).
One of the great benefits of enhanced personalization is that it can drive personal assistants to further develop and be responsive and proactive. Real personal assistants not only provide personalized answers when users ask questions, but also make decisions based on contextual analysis and Silent conversation.
To further enhance the personalization of virtual assistants, Qualcomm AI Research is working on the 'AI agent' on the terminal side, which is expected to continuously learn personal information from all sensor data, ultimately enabling intuitive action. Key features of AI agent include contextual Convergence and learning, which will ultimately support personalized response and enrich our lives. Preliminary research results have shown a far-reaching perspective on the terminal side. Personalization will not only enhance the user experience of virtual personal assistants, but will also enrich all aspects of our lives. This technology will transform many industries and continue to promote industry innovation.
* The content of this article is from the signature blog of Dr. Hou Jilei, senior director of Qualcomm and head of AI research project.
5.QLC 64-layer 3D NAND Flash is coming, SSD controller is ready
Larger capacity and higher performance have always been the main direction of the development of solid-state drive (SSD) products. To meet this goal, most NAND Flash factories plan to introduce 64-layer 3D NAND Flash with QLC architecture in the second half of 2018. It is expected to be applied to 96-layer products in 2019. SSD controller manufacturers are also following this trend, launching a new generation of products that support QLC.
In order to achieve higher storage density, the number of stacked layers of NAND Flash is increasing, and more information can be stored in a single cell. Currently, NAND Flash chips have entered the 64-layer TLC era, looking forward to 2019, Samsung (Samsung), Toshiba and other companies will further introduce 96-layer QLC particles. In order to meet the specifications of the next-generation NAND Flash that will be mass-produced, Taiwan's two major SSD controllers, Qunlian and Huirong, have prepared corresponding solutions.
Two major Taiwanese SSD operators show QLC firepower
Group Shaolian Electronics spokesperson Yu Shaoting said that 3D NAND Flash technology continues to advance. Currently 64-layer TLC is a fairly stable product. In order to further increase storage density, NAND Flash suppliers are working hard to develop into 96-layer QLC, when a single particle The storage capacity will be up to 1TB. In fact, there are already 64-layer 3D NAND Flashs using QLC architecture, but the technical verification is very strong. It is expected that 96-layer 3D NAND Flash will not be adopted until 2019. QLC.
Zheng Yuanshun, project manager of Huirong Technology Product Planning Department, pointed out that Micron/Intel is currently the fastest growing QLC technology. There is already a small amount of 64-layer QLC NAND Flash applied to the SSD used by the server, but it is fair. Its main feature is cost performance, as far as performance can only be said to be flat. But because QLC is an inevitable development trend in the industry, next Toshiba, Samsung will also launch 64-layer NAND Flash particles based on QLC, therefore, as SSD controller Suppliers, still have to be prepared for product layout.
In response to this technology development trend, the group officially released its first control chip supporting 3D QLC during Computex. Since the QLC architecture can achieve higher storage density, the reliability and read/write speed of data storage will be Affected, the controller is equipped with the fourth-generation SmartECC technology developed by the group. Under the PCIe Gen3×4 bandwidth, the sequential read/write performance can reach 3,200MB/s, and the IOPS is 600K.
The group further explains that when the data is written into the NAND Flash, the controller supporting the SmartECC will also generate a set of correction codes, which are stored together with the data. If the data is read back from the NAND, the control chip will pass through. The correction code corrects the data. If the error cannot be successfully corrected by the ECC correction code, the data will enter the SmartECC remediation process, and the data will be corrected by a specially designed algorithm to improve data reliability.
Not to let Qunlian specialize in the past, Hui Rong also released its next-generation PCIe SSD controller during Computex, supporting the latest 3D TLC and QLC NAND. In order to improve the reliability of data storage, its controller is unique with Huirong. Firmware technology, including end-to-end data path protection, SRAM ECC, the latest fifth-generation NANDXtend ECC technology combined with LDPC and RAID. As for read/write performance, the maximum sequential read speed of its PCIe Gen3×4 controller is up to 3,500MB/s, sequential write speed of 3,000GB / s, random read and write performance of 420K IOPS.
However, some people in the industry believe that because QLC is more reliable than TLC, even if 96-layer QLC particles enter mass production in 2019, it may take a while to apply to solid-state drives (SSD). Using 96-layer QLC The first batch of end-use products for pellets should not be solid-state drives, but applications such as USB flash drives that require less reliability.
SSD performance breaks the table
In addition to the storage capacity expected to continue to grow due to the introduction of QLC particles, the performance of SSD is also evolving. As the e-sports boom continues to burn, the performance of high-end consumer SSDs is no longer locked in the data center, the server Such as professional use of SSD, and this makes its heat has become a serious challenge that the industry must face.
The group said that the PCIe Gen3×4 SSD currently used in the data center or server, in order to ensure its stable operation, and will not slow down due to overheating, usually customers will install the cooling module on the SSD, some more sophisticated customers. In the product integration design, we will try to make the fan airflow in the server take care of the area where the SSD is located. The performance is not inferior to the high-end eSports SSD of the enterprise SSD, and it will inevitably encounter heat dissipation problems. In the SSD reference design developed for the e-sports market, heat sinks are introduced to enhance the heat dissipation capability of the SSD.
Zheng Yuanshun pointed out that the heat dissipation of high-performance SSDs is indeed a matter that must be carefully considered. In order to pursue the miniaturization of the dimensions of the modules, the industry adopts a series of M.2 specifications as the standard for SSD modules, but this specification has a large The problem is that the heat dissipation is not good. This problem is not highlighted when the speed of the particles and the interface is not fast, but with the increasing popularity of PCIe Gen3×4, this specification is not conducive to heat dissipation. Increasingly emerging.
In view of this, Huirong specifically listed low power consumption as a key design goal when developing controller chips. This is not only to save power, but also to reduce heat. Compared with the previous generation controller, the new generation controller The temperature at full speed is reduced from the original maximum of 78.6 degrees Celsius to 55.2 degrees, which helps to reduce the problem of overheating and slowdown operation of SSD products.
However, with the PCIe Gen4 standard, the interface used by future SSDs will be upgraded from PCIe Gen3 to Gen4. It is only a matter of time. Zheng Yuanshun expects that in the PCIe Gen4 era, the heatsink should become the standard equipment of SSD, because Gen4 is too fast. The heat problem will definitely be more serious than now, and the heat sink will become standard equipment. New Electronics
6. Does the autopilot test need to do this?
According to the NI survey, 72% of test engineers hope to integrate into the design of test instruments. Among them, 25% need to provide programming APIs. Customizable hardware and support services, so they can develop their own test systems to meet their needs; 47% need test instrument manufacturers tailor-made test solutions. For 5G, autopilot, Internet of Things (IoT) and other emerging Applications not only drive innovation needs, but also bring increasingly complex functions, test time and cost compression, customization and other challenges to the test industry. Just as Luke Schreier, vice president of sales for NI automated test products, points out that the industry needs to be systematized in response to test challenges. , integrated, open test solution response, and PC-based measurement and automation platform PXI is the ideal choice for its standardization and innovative balance and huge ecosystem.
PXI is mainly composed of PXI chassis, PXI controller, PXI module and software. It is suitable for different stages of test and measurement applications by providing modular and other I/O with dedicated synchronization and main software functions.
There are more than 2,000 PXI modules on the market for emerging applications and challenges, and PXI is constantly 'revitalizing'. Luke Schreier said that NI develops new modules based on different needs and integrates to meet the needs of industry applications. The software is continuously upgraded and optimized based on application development environment such as LabVIEW, and special test management software such as TestStand. Through modular rapid iteration and continuous optimization of software, customers can customize and reconfigure the system to realize platform generalization and on-demand customization. Chemical.
Modularity can be described as a major advantage of PXI. Jia Qingchao, automotive market manager of NI Asia Pacific, believes that hardware is easy to implement by modularization because NI does not need to be swapped, but can be tested according to customers. Need to upgrade or replace boards according to changes in test items. NI is also constantly improving higher performance modules, such as integrating the latest multi-core processors for faster and more efficient parallel testing, through FPGA, data conversion And so on, constantly improve the signal processing algorithm capabilities and measurement range and performance.
The modular flexibility is also reflected in the plug-and-play of the chip. NI RF technology market engineer Tu Fangze said that in the 5G and automotive industry tests, NI found that the data requirements are very high, the amount of data is very large. NI through the change board The chip in the process, which enhances the performance required by the customer, can solve this problem, including the front-end data acquisition, etc., which can use the original architecture without requiring an overall change, thus greatly reducing the customer's testing cost.
At the recent PXI Technology and Application Forum, NI shared the latest research and application examples of PXI, including large aircraft C919 testing, automotive entertainment information system testing, semiconductor testing, etc. In particular, China's semiconductor industry is welcoming a new one. Round development opportunities, the whole industry is moving towards high performance, small size, low cost, which is a serious test for practitioners. In the forum, Tu Fangze shared how NI can reduce from the laboratory through the PXI platform. Mass production test costs, practice of improving test efficiency.
Software change
The French novelist Jean-Baptiste Karr once said, 'The more things change, the less the change.' NI has practiced this principle. In addition to hardware iterations, NI spares no effort in the field of test software development. Continuous investment, the emergence of test platforms and tools such as Systemlink, InstrumentStudio 2.0, and the results in the Chinese market. The latest automotive-oriented FlexLogger data logging software can capture, display and record data without support, support Intuitive workflow and integrated data management capabilities. 'Because some customers may be unfamiliar with programming, they need a software that can quickly perform data acquisition and related processing without programming, and display it in real time for relevant verification. For example In the car lift test, in the brake test, some customers used FlexLogger to make rapid extensions in the laboratory. 'Tu Fangze said.
InstrumentStudio is another representative product for industrial applications. It enables simple and intuitive error correction during test execution in a real-time, interactive multi-instrument environment. In addition, each workflow can work with Three-party software interoperability to reuse program code and IP as much as possible. Tu Fangze said that InstrumentStudio can display a variety of instruments on a single software interface, allowing all instruments to be operated. At the same time, it can export LabVIEW programs. For component auto-verification and aerospace, automotive and semiconductor production testing is a good configuration.
The software of the PXI platform has become 'everything'. Jia Qingchao said that NI software can be divided into three categories, the first is platform software, such as LabVIEW; the second is for industry-oriented applications, such as for the automotive industry. FlexLogge, and InstrumentStudio for specific industries and applications. The third category is enterprise-level software for device and data management, such as Systemlink.
The top priority of the above changes is that NI continues to integrate software flexibility with modular iterations, allowing the test's 'channels and techniques' to be connected to the low cost required by customers, speeding up the market, and high reliability test appeals. .
Car test battle
In the course of NI, the car is undoubtedly the main force of 'acceleration'. NI recently announced the company's auto market sales growth target for the first time: that is, between 2018 and 2020, NI will achieve 10%-12% increase in the automotive test market. .
For the ever-increasing autonomous driving, Jia Qingchao compares the car sensor and various radars to the eyes and ears, and compares the V2X to the mouth. It is believed that only these different technologies form a seamless integration, and the automatic driving era will come.
Whether represented by Tesla, Google, Baidu, etc.