Samsung, the descendent of the AI ​​chip, is coming up with a big knife: According to foreign media, Samsung has nearly completed the development of an AI chip. Its performance is comparable to Apple's A11 and Huawei Unicorn 970. Samsung is very likely to be in 2 During the MWC2018 conference held on the 25th of this month, the Galaxy S9 was announced and its new AI technology capabilities were demonstrated.
According to the website of the “Korean Heraldâ€, Samsung Electronics has basically completed the development of the first NPU and is planning to deploy it to the device later this year. The NPU is widely known as the AI ​​chip, and it is said that Samsung will carry this chip on the upcoming smart phone, which will help its mobile phone catch up with its competitors.
"According to Samsung insiders broke the news, Samsung has basically completed the development of the server AI chip, it is expected that the chip will be sold to the server vendor." An AI expert told the "Korean Herald."
In addition to CPUs, GPUs, etc., NPUs are added to smart phones, and the phones themselves can process, analyze, and store the data generated on smart phones without the help of a cloud server. NPU is widely regarded as the brain of smart phones.
Currently, Samsung lags behind Apple and Huawei as the two biggest competitors in this field, both of which have already launched mobile AI chips. Last year, Apple released the iPhoneX, which uses the NPU to implement familiar features such as facial recognition and animated emoticons. Next, Huawei introduced an NPU that can learn user habits over time and deploy it on the Mate10Pro mobile phone.
Although Samsung is a latecomer to this market, the “AI expert†reported to the Korean pioneer that “Samsung has reached the technological level of Apple and Huawei, and will certainly launch a better chip in the second half of this year.†According to reports, Samsung's chips run faster than Apple's A11 and Huawei's Unicorn 970.
In addition, the same source disclosed that Samsung may demonstrate its new AI technology capabilities while it is releasing the Galaxy S9 at the MWC2018 conference held on February 25.
Samsung is said to have invested heavily in artificial intelligence-related projects and has collaborated with professors and researchers at Korean universities to create chips that are safer and more efficient than their competitors.
The source revealed that Samsung is also developing an enhanced NPU for the GalaxyNote9 smartphone, which is expected to be released in September this year. Like the previous GalaxyNote7 and Note8, Note9 may become a turning point for Samsung.
AI mobile phone entered the Warring States era, chip development Matthew effect fades
As mentioned above, Samsung's AI chip, the main competitors are Apple and Huawei. At present, Apple's iPhone X and Huawei's Mate10 (Pro), V10 and other series of mobile phones have used artificial intelligence chips, becoming the leader of AI mobile phones.
The A11bionicneuralengine chip in the iPhoneX is capable of operating up to 600 billion operations per second, which is equivalent to 0.6TFlops (Cambrian NPU is 1.92TFlops, and can perform 19200 floating-point operations per second). It adopts a six-core design and consists of two high-performance cores and four high-efficiency cores. Compared to the A10, the speed of the two performance cores has increased by 25%, and the speed of the four energy efficiency cores has increased by 70%. On the process side, the A11 uses the TSMC 10nm FinFET process and integrates 4.3 billion transistors.
A11 can help accelerate artificial intelligence tasks, including FaceID, Animoji, and AR applications. A11 also supports CoreML, which is a new type of machine learning framework that Apple introduced at the WWDC Developer Conference. CoreML supports all major neural networks such as DNN, RNN, CNN, etc. Developers can encapsulate training-completed machine learning models into the App.
The Huawei Kirin 970 chip, for the first time, adopts TSMC's 10nm technology, integrates 5.5 billion transistors (3.1 billion for the Snapdragon 835 and 3.3 billion for the Apple A10), and reduces power consumption by 20%. It uses 4 Cortex A73 cores, 4 Cortex A53 cores, GPU is a 12-core Mali-G72MP12, and the processing speed can reach trillions per second. Kirin 970 chip built-in new upgrade self-developed camera dual ISP, support artificial intelligence scene recognition, face tracking, intelligent motion scene detection, while the night shot effect is upgraded again. On the baseband side, the Unicorn 970 adopts a more advanced 4.5G LTE technology and supports the world's highest LTE Cat. 18 specifications, achieving the industry's highest peak download rate of 1.2 Gbps. At the same time built-in TEE and inSE security engine, have higher security.
However, Samsung, Apple, and Huawei's "three-nation killing" will not last long.
As the competition for smartphones enters into a serious homogenization phase, as with full-screen, artificial intelligence chips will be the next “standard†focus for handset manufacturers. In addition to Samsung, Apple, and Huawei's Big Three, there are still a lot of manufacturers rushing into the competition of smart phones and AI chips, such as the recent rumors of the millet will be listed in the second half of 2018 - February 2017, Xiaomi self-developed mobile phone chip 澎湃S1 Since its inception, it has become the second mobile phone manufacturer in China to have its own R&D chip capability.
Artificial intelligence chips have huge imagination for mobile phone performance enhancement, human-computer interaction and other functions. The R&D capability of AI mobile phones, especially high-end flagship AI mobile phones, is an important step that can open up the gap between them and their opponents. The mobile phone market landscape will also follow R&D. The ability gradually differentiates, the stronger is stronger, and the weaker Matthew effect becomes more prominent.
Mobile end AI chip, who can laugh to the end?
Because the GPU has a power bottleneck, and users are increasingly concerned about personal privacy security and timeliness, the development of terminal smart chips is in line with the needs of the times.
The terminal's smart chips need to have both high performance and low power consumption. In addition, the terminal also covers different application scenarios. All of them need to be adjusted and optimized according to specific requirements in the selection of power consumption, delay, data throughput, and accelerator solution.
Xin Zhiyuan has combed the chips that can support AI functions on some mobile terminals:
Google: PixelVisualCore
In October 2017, Google disclosed on its official blog a dedicated image processing coprocessor, PixelVisualCore, used in Pixel2. This is the second chip launched by Google after the server's TPU, this time for the mobile. PixelVisualCore is designed and developed by Google and Intel. It is mainly used for image processing and machine learning. This chip is composed of 8 IPUs (each containing 512 ALUs) and 1 Cortex-A53 core. It can provide 3TFLOPS floating-point computing capability.
Qualcomm: Xiaolong 845 processor
When it comes to cell phone chips, how can we not improve communication? Qualcomm officially announced the Xiaolong 845 mobile platform in early December 2017. Xiaolong 845 processor adopts 10nm LPP process technology, including GPU adopts Adreno630, X20LTE modem, WiFi, Spectra280ISP, and Hexagon685DSP coprocessor, Qualcomm AqsticAudio in terms of sound quality, and four 2.8GHz cores + four CPUs. 1.8GHz small-core + 2MB cache Kryo385 CPU, mobile security chip, another in 845 added a separate memory. In terms of AI, Xiaolong 845 mainly uses the Kryo385 custom architecture, Adreno 630, and Hexagon 685 to asynchronously compute data on the terminal. Compared to the 835, the Snapdragon 845 has three times the computational power of the 835, and is now able to support several mainstream deep learning frameworks such as S845GoogleTensorFlow, FacebookCaffe, and OpenNeuralNetworkExchange.
MovidiusVPUMyriad2
In September 2016, Intel issued a statement to acquire Movidius. Movidius focuses on developing high-performance visual processing chips. The current CEO is the original general manager of the OMAP division of Texas Instruments. Its technical steering committee is also a powerful, veteran of the semiconductor and processor industries. Daniel Doberpühl, founder of PASemi, acquired by Apple. Kimi Takeo, an expert in computer science/computer vision at Ned Mellon University, and David Tupman, a former vice president of engineering and a senior engineer at Apple's iPhone and iPod divisions, took part.
Its latest generation of Myriad2 vision processors consists of a SPARC processor as the main controller, plus a dedicated DSP processor and hardware acceleration circuitry to handle specialized visual and image signals. This is a visual processor based on the DSP architecture. It has a very high energy consumption ratio in visually relevant applications and can spread visual computing to almost all embedded systems. The chip has been widely used in Google 3D project Tango mobile phones, DJI drones, FLIR smart infrared cameras, Haikang series cameras, Hua Rui smart industrial cameras and other products.
Cambrian 1H8 and 1H16
In the Cambrian era, the world’s first commercial deep learning processor IP, the Cambrian 1A processor, was released in 2016. Cambrian 1A broke many records and was widely followed by the industry. It was selected as one of the fifteen “World Internet Leading Scientific and Technological Achievements†by the Third World Internet Conference. In November 2017, at the company's first conference, CEO Chen Tianshi introduced three new intelligent processor IP products, including the Cambrian 1H8 for visual applications in low-power scenarios, with more universality and higher Performance of the Cambrian 1H16.
According to Chen Tianshi, compared with the Cambrian 1A, the new products are optimized in terms of power consumption, energy efficiency ratio, and cost, and the performance/power ratio is once again a leap. The scope of application covers image recognition, security monitoring, smart driving, and no Man-machine, speech recognition, natural language processing and other key applications.
Horizon Sunrise and Journey
On December 20, 2017, Horizon held a press conference in Beijing to launch Journey and Sunrise, two computer vision-oriented processors for driverless and smart cameras. The Rising Sun and the journey are all embedded artificial intelligence vision chips, respectively, for smart driving and smart cameras. Zhou Feng, chief chip architect of the horizon, introduced Xin Zhiyuan. With two chips, the chip performance can reach 1 Tops, and 1080P@30 frames can be processed in real time. Each frame can detect, track, and identify 200 targets at the same time. Typical power consumption is 1.5w.
Shen Jiantao
On October 24, 2017, Shenzhen Shen Technology held a press conference to formally announce the completion of the A+ round financing of approximately US$40 million, led by Ant Financial and Samsung Ventures. CEO Yao Song also announced a series of chip plans. The “Tingtao†and “View of the Sea†chips developed by Shen Jian will be available in the third quarter of 2018.
Among them, "Tingtao" will complete the product loading in the first half of 2018. This series of chips adopts TSMC's 28-nanometer process, and the core uses its own Aristotelian architecture with a peak performance of 1.1 watts and 4.1 TOPS. The Aristotle architecture is designed for convolutional neural networks. At present, convolutional neural networks are generally used to deal with image-related intelligence problems. The flexibility and extensibility of this architecture make it applicable to a variety of different specifications of terminals.
Heterogeneous Intelligence: NovuTensor
At this year's CES, NovuMind (Chinese name "heterogeneous intelligence") demonstrated to the industry for the first time its own research and development of the first high-performance, low-power AI chip NovuTensor, claiming that may be in addition to TPU, the world's most running Fast single chip.
NovuMind said that this is the only chip in the world that can actually run, performance reaches the mainstream GPU/TPU level, and the performance/power ratio is far superior to the mainstream GPU/TPU. In the case of power consumption 12w, NovuTensor Seconds can identify 300 images, and each image can detect up to 8192 targets. Compared to the most advanced desktop server GPU (250W, 666 images per second can be identified), only 1/20 of the power can be used to reach it. One-half of the performance; compared to the most advanced mobile or embedded chips, the performance is more than three times that of the same power. It is understood that this CES shows only the FPGA version, and other ASIC chips that are currently being streamed will be officially shipped, the performance will be increased by 4 times, the power consumption will be reduced by half, the power consumption will not exceed 5 watts, and 15 trillion operations can be performed. Ultra high performance.
GTI light spear processor
Gyrfalcon Technology Inc. (GTI), which was founded in early 2017, has recently attracted industry attention with the Lightspeeur 2801S, a company founded by Chinese chip veterans and headquartered in Silicon Valley, USA. Its chip solution is based on the APiM architecture and has 28,000 parallel neural computing cores. It truly supports on-chip parallel and in-situ calculations, eliminates the need for external memory units, overcomes performance bottlenecks caused by memory bandwidth, and excels in efficiency and energy consumption. 9.3 Tops/Watt achieves high-density computing performance both in training mode and inference mode.
At this year’s CES, GTI introduced the Laceli artificial intelligence computing stick with the built-in Lightspeeur 2801S chip, which can provide performance exceeding 9.3 trillion floating point operations per second at a power of 1 watt, which surpasses that of Intel’s Movidius neural computing stick. The calculation power per watt power range is 0.1 trillion times. Laceli artificial intelligence computing sticks can be applied in a variety of deep learning scenarios, including image and video recognition, understanding and description, natural language understanding, natural language processing, and the like.
Using Solar Panels is a very practical way to produce electricity for many applications. The obvious would have to be off-grid living. Living off-grid means living in a location that is not serviced by the main electric utility grid. Remote homes and cabins benefit nicely from solar power systems. No longer is it necessary to pay huge fees for the installation of electric utility poles and cabling from the nearest main grid access point. A solar electric system is potentially less expensive and can provide power for upwards of three decades if properly maintained.
182Mm Solar Cells,Solar Cells Materials,Solar Cells Voltage,Pv Mono Solar Cell
Jiangxi Huayang New Energy Co.,Ltd , https://www.huayangenergy.com