To gather micro-message information, Japanese microcontroller manufacturer Renesas Electronics plans to assemble all automotive microcontrollers (MCUs) from TSMC, focusing on software and semiconductor R&D, in order to reduce the high cost of chip production equipment.
Renesas today announced the industry’s first integrated Flash MCU using the 28nm process and will begin shipping samples on the same day. To build the next generation of more efficient and reliable environmentally friendly cars and self-driving cars, this revolutionary RH850/ E2x series microcontrollers incorporate up to six 400Mhz processor cores, becoming the industry's first automotive control flash MCU that can achieve 9600MIPS instruction processing capability. This series of MCU also has up to 16MB of built-in flash memory and more Complete security features and functional safety.
Renesas 28-nanometer MCU has 3 times the computing performance than the current 40 nanometers. With the increasing of autopilot technology, it can meet the needs of MCUs with low power consumption and high processing performance. Renesas said that the 28-nanometer MCU is the current The world's most advanced products have been shipped to DENSO and many other well-known auto parts factories in March. The goal is to mass-produce them through TSMC in 2020. Renesas’s own fab will gradually withdraw from manufacturing automotive semiconductors.
Following the announcement of the process development for the 28nm embedded flash memory in February 2015, Renesas Electronics announced in September 2016 that it will co-produce 28nm MCUs with TSMC. Renesas stated in today’s press release that the world’s first product was introduced to the market today. 28nm embedded flash MCU will become another important milestone for Renesas Electronics. Renesas Electronics has verified the application of fin-shaped MONOS flash memory technology in 16/14nm and next-generation MCU products.
The Nikkei pointed out that Renesas can reduce production costs, and TSMC can also use depreciated equipment to produce 28-nanometer MCUs.
According to the financial report released by Renesas, the Renesas Semiconductor business revenue was based on non-GAAP basis in the previous quarter (October to December 2017), a significant increase of 28.0% from the same period last year to 206.5 billion yen. Yuan, in which automotive semiconductor revenue grew by 14.7% to 107.8 billion yen.
For the full year of 2017, Renesas Semiconductor’s revenue grew 23.4% to 765.7 billion yen, including automotive semiconductor revenue growth of 13.8% to 408.1 billion yen.
2. Foxconn's 8.66 billion US dollars to buy peripherals Belkin needs US government approval;
According to foreign media reports, a subsidiary of Foxconn, an Apple supplier, announced that it will acquire Belkin, a well-known peripheral product manufacturer, for approximately US$866 million. Belkin also owns Linksys, Phyn and Wemo. Brand.
The subsidiary is called FIT (Foxconn Interconnect Technology Limited) and a joint statement on Belkin's official website revealed the news. The joint statement stated that the acquisition will be completed in a cash transaction. .
'Hong Teng Precision is pleased to acquire Belkin and its capabilities in high-end consumer products,' said Sidney Lu, CEO of Hong Teng Precision. 'By integrating Belkin's best-in-class capabilities and solutions into Hong Teng Precision We hope to enrich our high-end consumer product portfolio and accelerate our penetration of smart homes.
Belkin was founded approximately 35 years ago in California, USA. Later, during the boom of the personal computer, he was known as a manufacturer of high-quality peripheral equipment. It started with various peripheral devices, such as surge protectors, USB hubs, and of course cables. Line, Belkin later caught up with Apple's popularity in the early 21st century iPod products.
Currently, the company produces a range of products that provide services to manufacturers of computers, smart phones, tablets, smart watches and other electronic products. Belkin is known as WeMo's independent brand and is well-known in the emerging home-connected market.
Recently, Belkin and pipe supplier Uponor established a new company named Phyn in 2016. The first product launched by the new company - the water monitoring device connected to the iPhone, Phyn Plus - was exhibited at CES earlier this year. .
In accordance with the terms of the agreement reached with Foxconn, Belkin will operate for a subsidiary of Hong Teng Precision, and its CEO and co-founder Chet Pipkin will take control of Shuai Yin. As part of the agreement, Chet Pimpkin is expected to participate in Hong Teng Precision Management.
According to a report in the Financial Times, the merger must be approved by the Foreign Investment Committee of the US Treasury Department, which recently expressed its opposition to chip maker Broadcom’s hostile takeover of competitor Qualcomm.
President Donald Trump vetoed Broadcom’s acquisition to endanger national security reasons. In view of this, Foxconn’s acquisition of Belkin’s case may also become unstable. Belkin owns network technology. However, it is important for the United States. Investors Foxconn has announced that it will invest US$10 billion to build LCD factories in Wisconsin before 2020. (Tianmenshan) Netease Technology
3. Toshiba sells chip subsidiaries or misses deadlines will seek more options;
According to Reuters, China’s Ministry of Commerce has not yet considered Bain Capital’s lead in a consortium’s plan to acquire Toshiba’s (6502.T) chip subsidiary for US$18 billion, making the transaction unlikely. Completed before the upcoming deadline. Toshiba seems to be looking for more options for the subsidiary.
China’s Ministry of Commerce briefly told Reuters on Tuesday that it is evaluating the transaction but did not elaborate further.
A person directly informed said that if the transaction is to be completed before the March 31 deadline, it must be approved by China's anti-monopoly authority early this week, as administrative procedures and transfers will still take some time to complete.
If it cannot be completed as scheduled, Toshiba will have the right to abandon the sale of the chip subsidiary without paying a fine. Some investors have urged Toshiba to consider this option. Toshiba’s chip subsidiary is the world’s second-largest NAND chip maker.
A Toshiba spokesperson said that the company has not yet given up its efforts to complete the transaction by the end of the month. Even if it misses the deadline, it will still sell the chip business as soon as possible.
Previously, some analysts predicted that the flash memory business will be split and listed. There are also media reports that the lack of anti-monopoly approval may be a good thing for Toshiba. Since the financial data has improved, the company can re-raise the bid price even more than the current The price was $4 billion higher. Some active shareholders opposed Toshiba’s transaction and believed that the value of the assets was underestimated. They thought that they should negotiate with Bain Capital on the purchase price, or make the flash business split.
It is generally believed that the biggest difficulty in the review of Toshiba deals comes from South Korea's Hynix. The industry is also concerned that if the acquisition is completed, Hynix will gain greater scale and influence in the market and damage the market competition. Historically, Hynix has passed Abnormal means to obtain a case of Toshiba Semiconductor Technology.
4. MLCC continue to increase prices need to be vigilant in price manipulation;
Traditionally, Japanese and Korean companies mainly produce mid- to high-end MLCC products, while mainland China and Taiwanese companies are based on mid- to low-end markets. The industry estimates that the strategic adjustment of Japanese companies will release 20% of the standard MLCC market. .
The thrilling surge of storage prices gradually came to an end. However, the MLCCs (chip multilayer ceramic capacitors) that entered the price increase channel during the same period remained 'increasingly loud'. Recently, market sources said that the world’s third largest MLCC supplier, Taiwan’s giants In April, the price of MLCC was further raised by about 40%-50%.
In less than a year, Guo Da has raised the price of MLCC several times, and the total price increase has been close to 30%. Thanks to the price, the sales volume has risen in synchronicity. Guoju first achieved a monthly income breakthrough of 30 in September 2017. TWD TWD, since then, its monthly revenue has increased by more than 30% year-on-year, and its revenue from December 2017 to February 2018 has increased by more than 50% year-on-year. In 2017, the country’s huge revenue was NTD 32.26 billion. , YoY growth of 16.1%, gross margin increased from 23.6% to 32.5%, an increase of 8.9 percentage points.
The stock price of the company is even more alarming. Before 2017, the stock price of the company has never reached NT$70. In March 2017, it broke through NT$80 for the first time. By March 2018, the stock price of the company has exceeded NT$500. , The year-on-year increase was more than 520%. The market, the market value of the double surplus, making the giant of the country to become the biggest winner of MLCC price increase.
Supply structure adjustment
Similar to the incentive for the increase in memory prices, the MLCC's price rise is also due to the industrial restructuring within the industry.
MLCC is the most common capacitive device in electronic products. It is widely used in mobile phones, automobiles, home appliances and other consumer devices. In a mobile phone, hundreds of MLCCs are used, and the MLCC usage of a car is 5,000. .
However, the MLCC's price is extremely low. According to the Guoju 2016 earnings report, its MLCC output totaled 343.1 billion, and its revenue was approximately NT$9.1 billion, approximately NT$0.026/piece, and approximately RMB0.5 cent/RMB.
'Because it was too cheap, Japanese companies began to abandon low-end and mid-to-high-end products in favor of higher-profit cars, and MLCCs for industrial applications.' An industry source briefed reporters. Mid-2016, Japan TDK announced its withdrawal from the general MLCC market. Japan is the world's leading supplier of MLCCs. Murata and TDK alone have produced 50% of the world's MLCC products.
Traditionally, Japanese and Korean companies mainly produce mid- to high-end MLCC products, while mainland China and Taiwanese companies are based on mid- to low-end markets. The industry estimates that the strategic adjustment of Japanese companies will release 20% of the standard MLCC market. .
This opportunity became the trigger for price increase. Subsequently, Taiwan’s Guoju, Huaxin Branch, Chaozhou Sanhuan in China, and Fenghua Hi-Tech Co., Ltd. successively announced price increases for MLCC products. According to the 2017 financial report, ASIMCO 2017 MLCC products Revenue was approximately US$12.14 million, an increase of approximately 19% year-on-year. In the most recent year, AAC's stock price rose from NT$44 to NT$111, an increase of approximately 152%.
In China, although MLCC-related stocks did not show such strong increase, major brokerage institutions are optimistic about stocks such as Sanhuan Group and Fenghua Hi-Tech. It is worth mentioning that Fenghua Hi-tech Co., Ltd. specifically mentioned in its 2016 financial report. At that time, sales of Fenghua Hi-tech MLCC products increased by 22.1%, and net profit increased by 426.27%.
Although Murata, the world's largest MLCC manufacturer, has not publicly stated that it has raised prices, it has expressed its optimism that MLCC is in short supply and invested 260 billion yen in capacity expansion.
Alert price manipulation
'Although it sounds like a big increase, it has no effect on mobile phones, and the overall value is too low. ' A veteran of the mobile phone industry told reporters that the MLCC’s price hike has not been passed on to consumer products such as mobile phones, automobiles, and home appliances. , Consumers have no intuitive experience.
'Moreover, mobile phone manufacturers of a certain scale can lock prices for such devices for half a year or even one year in advance. 'It is precisely because of this that the person said 'we haven't felt the demand that is so obvious at the time of previous memory prices.'
It should be pointed out that the MLCC's price increase is mainly low-end products. An industry insider said, 'The current MLCC, high-end products do not dare to increase prices, only the low-end is rising. And the price increases, many are channel providers' Shantou. The supplier's original price increase rate will be multiplied by the channel provider here.
Murata Japan, which has never paid a public price increase, has maintained an increase in the revenue of its capacitor products by more than 30% in the past five years. The financial report shows that in 2017, Murata capacitor products revenue was 369.5 billion yen, an increase of 32.6% over the same period of last year. huge.
At the same time, Japan’s TDK, which has announced its abandonment of the general MLCC, recorded revenue of 115.5 billion yen, or about US$102 million, an increase of 13.6% year-on-year in the three quarters of April-December 2017. The growth rate was the same as before. The increase in the company’s revenue was mainly attributed to the growth in sales driven by smart phones, automotive electronics and other markets.
According to customs data, in 2016, China imported MLCC products worth 5.226 billion U.S. dollars. In 2017, China imported MLCC products worth 5.62 billion U.S. dollars, an increase of 7.6%. But in January-February 2018, MLCC imports reached 1 billion U.S. dollars. The U.S. dollar was up 35.5% from the 738 million U.S. dollars for the same period last year.
According to the forecast data of China Industry Information Network before this, the global MLCC market size in 2017 totaled approximately US$10 billion, while China’s imports of MLCC alone reached US$5.6 billion. The above-mentioned mobile phone manufacturers introduced that 'we need to be vigilant against these companies to manipulate the market, after all, China. Is the world's largest consumer of electronic devices. '
In fact, the price monopolistic behavior in the passive component market is not uncommon. On March 21, 2018, the European Commission determined that nine Japanese companies had monopolized the prices of aluminum electrolytic capacitors/tantalum electrolytic capacitors between 1998 and 2012 and aggregated them out. A fine of approximately 254 million euros. In January 2018, the Singapore Competition Commission also imposed a monopoly penalty of about 100 million yuan on the price monopoly of five aluminum electrolytic capacitor companies during 1997-2013.
Previously, the NDRC had repeatedly discussed Samsung on the issue of increasing memory prices. Currently, memory prices have started to decline. Passive components such as MLCC are still rising prices. 21st Century Business Herald
Embedded neural network gives machine vision, hearing and analysis capabilities;
Youval Nachum, Senior Product Manager, Audio and Speech Product Line, CEVA
The potential applications of artificial intelligence (AI) are increasing. Different neural networks (NNs) have been tested, adapted and improved to solve different problems. There have been various ways of using AI to optimize data analysis. Most of today's AI applications, such as Google Translate and Amazon Alexa voice recognition and visual recognition systems are also leveraging the power of the cloud. Depending on the always-on Internet connection, high-bandwidth links and web services, IoT products and smart phone applications can also integrate AI functionality. Most of the attention is focused on visual-based artificial intelligence, partly because it tends to appear in news reports and videos, and partly because it is more similar to human activities.
Voice and visual neural networks (Image courtesy of CEVA)
In image recognition, a 2D image is analyzed (a set of pixels is processed at a time), and a larger feature point is identified by a continuous layer of the neural network. The first detected edge is a part with a high degree of variability. For example, the earliest identified edges are around the features of the eyes, nose, and mouth. As the detection process goes deeper into the neural network, the features of the entire face will be detected.
In the final stage, combined with feature and location information, a specific face with the greatest degree of matching can be identified in the available database.
Neural network feature extraction (Image courtesy of CEVA)
The objects captured or captured by the camera can be found in the database with the highest probability of matching the human face through the neural network. It is particularly good that the objects do not need to be photographed at exactly the same angle or position, or under the same lighting conditions.
The popularity of AI so quickly, in large part because of open software tools (also called frameworks), makes it easy to build and train a neural network to achieve a target application, even when using a variety of different programming languages. Two common common frameworks are TensorFlow and Caffe. For known recognition targets, a neural network can be defined and trained off-line. Once trained, the neural network can be easily deployed to an embedded platform. This is a smart one. Divide, allows to train the neural network with the ability of PC or cloud, and the power sensitive embedded processor only needs to use trained data to identify.
The ability of humanoids to identify people and objects is closely related to popular applications, such as industrial robots and self-driving cars.
However, artificial intelligence has the same interest and ability in the audio field. In the same way as image feature analysis, the audio can be decomposed into feature points for input to a neural network. One way is to use the mel-frequency cepstrum coefficient (MFCC). Decompose audio into useful features. Initially the audio sample is decomposed into short time frames, for example 20 ms. Then the signal is Fourier transformed and the power of the audio frequency spectrum is mapped to a non-linear scale using overlapping triangular windows. .
Sound neural network decomposition diagram (Photo: CEVA)
Through the extracted features, the neural network can be used to determine the similarity of vocabulary or speech in the audio sample and audio sample databases. Like image recognition, neural networks extract possible matches for a particular vocabulary in the database. For those who want to copy Google For people with Amazon's 'OK Google' or 'Alexa' voice triggering (VT) features, KITT.AI provides a solution through Snowboy. Triggering keywords can be uploaded to their platform for analysis, exporting a file, integration In the Snowboy app on the embedded platform, voice trigger (VT) keywords can also be detected offline. Audio recognition is not limited to language recognition. TensorFlow provides an example project on iOS. Can distinguish between male and female voices.
Another application is to detect the animals and other sounds in the cities and neighborhoods where we live. This has been verified by a deep learning bat surveillance system installed at the Queen Elizabeth Olympic Park in the United Kingdom. It provides the integration of visual and auditory recognition neural networks into The possibility of a platform. For example, identifying specific sounds via audio can be used to trigger a security system for recording.
There are many cloud-based AI applications that are unrealistic. On the one hand, there are data privacy issues. On the other hand, services due to poor data connectivity or insufficient bandwidth cannot be sustained. In addition, real-time performance is also a concern. For example, industrial manufacturing. The system requires a transient response to operate the production line in real time, and the delay in connecting to the cloud service is too long.
Therefore, moving the AI function to the terminal device is getting more and more attention. That is, exerting the power of artificial intelligence on the device being used. Many IP providers provide solutions such as CEVA's CEVA-X2 and NeuPro IP cores. And supporting software, it is easy to integrate with the existing neural network framework. It provides the possibility to develop embedded systems with artificial intelligence, while providing the flexibility of low-power processors. A speech recognition system as an example You can use power-optimized artificial intelligence integrated on the chip to identify a set of Voice Triggered (VT) keywords and a minimal set of Voice Commands (VCs). More sophisticated voice commands and features can be used in applications. Wake up from low-power voice triggered state, completed by cloud-based AI.
Finally, a Convolutional Neural Network (CNN) can also be used to improve the quality of text-to-speech (TTS) systems. TTS has historically integrated many high-quality recordings from many small pieces of the same voiceover into a continuous sound. The output is human-readable, but it still feels like a robot's voice because of the strange tone and pitch of the output. If you try to express different emotions, you need a new set of recordings. Google's WaveNet improves the current In the situation, TTS waveforms are generated at a rate of 16,000 samples per second through a convolutional neural network (CNN). Compared to previous sound samples, the output is seamless, with significantly more natural and higher-quality sound.