Home > News content

BAT AI chip layout inventory Who is the "Chinese core" new hope?

via:博客园     time:2018/4/21 21:31:37     readed:962

How popular is the AI ​​chip in the technology field? If the traditional chip giants such as Qualcomm, Intel, and Nvidia have all begun to manufacture AI chips, it is not enough to prove their popularity. Then e-commerce giant Amazon, social giant Facbook, and software giant Microsoft, etc. all began to independently research and develop AI chips. This not only shows that AI chips have Pay more attention and prove its importance.

Compared to the U.S. technology giants, the three domestic technology giants BAT didn't seem to be as fast. In the context of the ZTE event once again triggering the general public's lack of discussion on China's high-end core, some people place their hopes for the development of the Chinese core on BAT, which does not lack money. So, what are the chip layouts that BAT has completed? What is the focus of attention?


Chinese and American technology giants' chip gap

According to Lei Feng network (public number: Lei Feng network) understand that the current strength of artificial intelligence chip manufacturers is the United States.QuantitativelyAccording to incomplete statistics, there are 33 chip manufacturers in the United States and 12 in China. Among the US manufacturers are technology giants such as Google, Intel, and IBM, as well as large companies such as Qualcomm, Invista, AMD, and Xilinx that have absolute advantages in their respective fields, as well as some well-developed medium-sized companies. Active start-ups, mainly in China, mainly small and medium-sized companies.

Chip categoryLook, the U.S. manufacturers are spread over four major schools of artificial intelligence chips CPU, GPU, FPGA, and ASIC. The industrial structure of the IC design link is also very balanced, and monopolized in the GPU and FPGA fields. However, Chinese companies can only do slight things in FPGA compilation, ASIC and brain-like chips.

In addition, the innovation of AI chips involves artificial intelligence algorithms, programming languages, computer architectures, integrated circuit technologies, and semiconductor processes. thereforeTalentIt is also very important that the 2017 data shows that there are approximately 78,700 employees in 1,078 artificial intelligence companies in the United States and 39,200 employees in 592 companies in China, which is only 50% of that in the United States.

It can be seen that many technology companies have tried to enter the AI ​​chip field in the past decade. However, due to high patented technical barriers, small market size, fierce competition, and long investment cycles, many technology companies have failed to close.

However, the AI ​​chip, as the core of the artificial intelligence industry, is also the highest link of technical requirements and added value. The industry value and strategic position are much larger than the application layer. This is also an important reason why technology giants such as social networking, software, e-commerce and other countries pay attention to the layout of AI chips.

BAT rushes to layout AI chips, Ali Alto wins

Since the US technology giants have already begun to lay out AI chips, and artificial intelligence chips from companies such as Google and Invista have already achieved good results, how have domestic technology giants progressed?

Ali is eager to invest in AI chip companies, self-developed Ali-NPU fresh exposure


Ali has been accumulating AI. In addition to the numerous spectators who recently attracted RMB1.5 billion from AI Unicorn to invest in Tang Technology, Ali has invested a lot of AI at home and abroad. Vertical fields involve blockchains and edges. Computing, new storage, quantum computing, etc. Among them, in Ali's "investment history", Ali and Ali showed enthusiasm for AI chip companies.

Zhongtian Microsystem

In 2015, Alibaba cooperated with Hangzhou Zhongtian Microsystems Co., Ltd. (Zhongtian Microsystems) for in-depth cooperation and developed a cloud chip architecture for various segments of the Internet of Things. Zhongtian Microsystems uses 32-bit high-performance, low-power embedded CPUs and chip architectures as its core business. The CPU business lines for various embedded application scenarios are developed and 7 embedded CPUs are used to cover high, medium and low embedded applications, including Internet of Things, digital audio and video, information security, networking and communications, industrial control, and automotive electronics. In January 2016, Ali became its largest shareholder. In June 2017, Ali injected 500 million yuan into Zhongtian Microsystems and formally entered into the field of chip infrastructure design. In April 2018, Alibaba acquired a wholly-owned subsidiary of Zhongtian Microsystems. .

Barefoot Networks

In November 2016, Ali and Tencent together led the programmable chip company Barefoot Networks. Barefoot Networks has developed the world's first programmable chip capable of processing network packets at a rate of 6.5 megabits per second. It is reported that the successful joining of Alibaba and Tencent has pushed Barefoot’s Series C financing to a peak of US$80 million.


In August 2017, Ali invested in Cambricon Technologies Corporation Limited, a Chinese AI chip company known as the world's first successful streamer and has a mature product. Cambrian has two product lines: terminal AI processor IP and cloud high-performance AI chip. The Cambrian 1A processor, which was released in 2016, is the first commercial deep learning processor for various types of terminal equipment such as smart phones, security surveillance, drones, wearable devices, and smart driving. The current Cambrian terminal processor IP products have been derived from multiple models such as 1A and 1H.

Shen Jian Technology

In October 2017, the investment department of Ant Financial, an important investment arm of Ali, invested in Shen Jian Technology. It is understood that Shenzhen Science and Technology was founded in 2016 by Tsinghua University and Stanford University's top deep learning hardware acceleration researcher. In 2017, it launched a deep neural network development kit "DNNDK" designed to simplify and accelerate deep learning. Application development greatly shortens the cycle of algorithm deployment to hardware.


In November 2017, Ali also led Kneron. The positioning of endurance is the terminal artificial intelligence technology provider, and the core competitiveness lies in the lightweight NPU, focusing on the terminal market. Relevant data shows that the power consumption of its M4 and M5 chips can be reduced to milliwatts, and the required volume can be reduced to 1/40 of the mainstream neural network chips. Currently, we have provided customized solutions for smart homes, smart security, and mobile phones.

Self-developed Ali-NPU neural network chip

On April 19, 2018, Alibaba announced that Dharma was developing a neural network chip called Ali-NPU. According to the design, Ali-NPU performance of Alibaba will be the mainstream CPU and GPU architecture AI chips on the market. Times, manufacturing costs and power consumption is only half, cost-effective over 40 times. The Ali-NPU chip is mainly used to solve AI reasoning and calculation problems in commercial scenes such as image, video recognition, and cloud computing to improve computational efficiency and reduce costs.

Overall, from Alibaba’s AI investment map, it can be found that its AI chip company covers many areas, including mobile phones, security, smart home and so on. As for the reasons behind this, it can be understood that Ali hopes to open up and guarantee through the investment of the chip. "Operating system (yunOS) + processor chip or application chip + algorithm + terminal + application + cloud" This channel is smooth and unobstructed.

Regarding self-developed chips, Ali-NPU, independently developed by Alibaba, focuses on AI reasoning in both video and image processing scenarios. The establishment of the Dharma College last year was a strong support. Song Hao, head of the Alibaba Strategic Investment Department's Zhongtian Micro Project, stated that Ali will continue to invest in the chip field continuously.

Baidu loves FPGAslayoutIntelligent Voice and Autopilot


DuerOS Smart Chip

Compared with Ali, Baidu released AI chips earlier through cooperation with other manufacturers.

In March 2017, Baidu released its DuerOS smart chip. The DuerOS smart chip integrates DuerOS into the Violet Zonui chip, which enables the chip to have low power consumption and low cost. It also provides a rich IO interface and supports Wi-Fi/Bluetooth connection modes. At the same time, the chip adopts the ARM mbed OS kernel and its secure network protocol stack to achieve secure connection with the cloud and reduce the entry threshold for device vendors. IoT solution provider Han Feng Technology also launched the WiFi module based on the chip, integrating the DuerOS intelligent voice interaction function.


At the California Hot Chips Conference in 2017, Baidu released XPU, a 256-core, FPGA-based cloud acceleration chip partnered with Xilinx. According to Lei Feng Network, Baidu designed the chip architecture to highlight diversity, focusing on computationally intensive, rule-based tasks while ensuring maximum efficiency, performance, and flexibility.

Baiyang researcher Ouyang Jian said: “In Baidu, we have been using FPGAs for many years. We have a large number of FPGAs in our data centers, cloud computing platforms, and automated driving programs. We understand the advantages and disadvantages of FPGAs and how to optimize them. With XPU's large core, we focus on a variety of computing tasks. If XPU is proven to be used for AI, data analysis, cloud computing, and driverless, then Baidu may need to use ASIC technology to develop XPU. The XPU's goal is to balance performance and efficiency and handle diverse computing tasks. ”


In February 2018, Baidu Ventures and a consortium of US semiconductor executives led the US startup Lightelligence. Lightelligence mainly uses the new optics-based technology to accelerate artificial intelligence workloads, and accelerates information processing through emerging technologies called photonic circuits. Photonic circuits are more efficient alternatives to electronic circuits, not electrons but photons for calculations.

Unlike Alibaba, which prefers to invest in AI chip companies and has not yet officially released its own AI chips, Baidu has released an FPGA acceleration chip that DuelOS's smart chip volume cooperates with Sai Si Ling. AI chip salaries for investment include Lightelligence. Although there are not many AI chip companies to invest in, Baidu set up a special organization to invest in AI projects in September 2016. It was named Baidu Ventures (BV) and BV invested more than 30 projects, including virtual modeling technology companies. 8i, FPGA acceleration technology company Falcon, drone flight control and map platform Airmap, intelligent retail store solution provider YI Tunnel, and smart education company job box, including many companies.

From Baidu’s AI layout, it’s not hard to find that Baidu seems to be more interested in FPGAs, and is now focusing on autopilot and smart voice.

Tencent's AI chip layout is at its lowest, and investment is mainly injected into smart medical care


Tencent is the latest in the deployment of artificial intelligence in the BAT echelon. In addition to the investment in Barefoot Networks chips mentioned above with Alibaba, Tencent does not currently have any other investment in open AI chips, nor does Tencent independently develop AI chips. Message exposure.

However, in terms of AI investment, Tencent prefers industry-wide coverage of strategic approaches such as VoxelCloud and carbon cloud intelligence as representatives of the smart medical field; real-time technology mainly involves motion algorithms; Weilai cars directly point to unmanned driving ;Leju Smart is a research and development company for home entertainment humanoid robots. At present, Tencent Capital's injection efforts are mainly in the smart medical field.

Through the arrangement of Lei Fengwang, it is not difficult to discover Ali's dedication to investing in AI chip companies, and announced that it is developing its own Ali-NPU chip during the US Department of Commerce’s ban on ZTE. From the public information, Alibaba is also an AI chip in BAT. The layout of the field is more comprehensive and the fields involved are also very rich. Baidu mainly cooperates with AI chips and prefers FPGAs. The focus of attention is mainly on intelligent voice and smart driving. Tencent is the third of the AI ​​chip layout, of course, through the investment in the AI ​​field layout is not limited.

So, what do you think of China's three major technology giants BAT who are most likely to save the core of China?

China IT News APP

Download China IT News APP

Please rate this news

The average score will be displayed after you score.

Post comment

Do not see clearly? Click for a new code.

User comments