'Insert Alpha Dogs into Mobile Phones' may only require one step: Designing a unique chip for artificial intelligence (AI). Starting in 2017, China is setting off an unprecedented AI chip startup craze, 'smart core',... Become China's new engine for occupying an intelligent era.
'Smart Core' : Accelerating the Intelligent Times
The chip is an artificial intelligence engine. 'No chip, no AI.' Wei Shaojun, director of the Institute of Microelectronics at Tsinghua University, said that the chip is the carrier of artificial intelligence. Whether CPU, GPU or other chip platforms, it is inseparable from the chip. Doing AI must have chips. Chips are an essential part of artificial intelligence.
For most smart demands, traditional processors based on general-purpose processors have high costs, high power consumption, large size, and low speed. The smart dedicated 'core' that is more suitable for deep learning becomes a must for industrial chain competition. Track.
Chen Tianshi, the founder of Cambrian Technology, an artificial intelligence chip 'Unicorn', told reporters that the Swiss army knife and kitchen knife can be used to describe the relationship between a general-purpose processor and a deep learning processor: Although the Swiss army knife has many functions, it does At the same time, the kitchen knife is more suitable. In terms of intelligent processing, a more appropriate kitchen knife is needed.
Nvidia's processor GPU, which is used for image rendering, is more suitable for hyperscale mathematic operations because of accidental hits on deep learning. It allows computer vision and other accuracy to exceed the human eye. It can be said that if there is no GPU This wave of AI waves has accumulated even more data. It also made the stock of NVIDIA, which was still around $30 two years ago, behave like a rocket and rise to nearly $250 today.
Nvidia is also not a matter of peace of mind. Google also released two generations of TPU, an AI chip optimized for the deep learning framework Tensor Flow, because Alpha Dog's debut is stunning. Microsoft, Intel, AMD and other veteran technology giants are all in this section. Active layout on the road, including vertical links.
The same starting line: The 'China Core' rush opportunity
At present, the chip is still China's largest import commodity, and the annual cost is 200 billion US dollars, which exceeds the cost of petroleum. In recent years, with the continuous development of the integrated circuit industry, the domestic chip replacement rate is gradually increasing. For smart chips, the good news is that there is no universal smart ecology in the world, and China is standing on the same starting line with other countries.
In 2016, the Cambrian developed the Cambrian 1A processor, the world's first commercially available deep learning processor IP product. In September 2017, Huawei Hass introduced the world's first mobile AI chip, the Unicorn 970, with its powerful local AI processing capabilities. It is from the integrated Cambrian 1A processor. Compared with the traditional 4-core chip, this chip achieves 25 times more performance in key artificial intelligence tasks such as computer vision, speech recognition, and natural language processing. And 50 times more energy efficiency.
"For example, after optimization of power, volume, speed, etc., one day we can put Alpha dogs into the cell phone." Chen Tianshi said that to do AI chips is to do things like building a house or building roads. There is no house, no road, no top-level application can go backwards.
From research and development to application, the AI chip is showing a flourishing trend here: In November 2017, the Cambrian released a new generation of smart processor products; in December, Horizon released the independently researched and developed AI chip 'Rising Sun' and 'Jour'; 2018 In March, Deep Thought released its medical imaging dedicated AI chip 'M-DPU'.
From the Cloud to the End: A Hundred Flowers Raceway
However, at present, there is no algorithm architecture that can cover all applications. Is there a general-purpose AI chip such as CPU in the end? The industry is still groping, and China's entrepreneurial enterprises are flourishing.
Similar to the Cambrian, etc., are the 'supporters' of the general AI chip, and Shenjian Technology, Horizon, etc., are focusing on the application terminal. Shen Song, CEO of Shenzhen Jianye Technology believes that every AI chip company must become a software company. & Systems Corporation. Because in the AI chip, the chip itself only 1%, the remaining 99% is a complete system and software.
'One day, the mobile phone may no longer be used as a major Internet device, but a variety of Internet of Things devices. Whoever will master the Internet of Things in the future, who is the creator of key data. A dedicated chip based on sound and images will become very Important. 'Huang Zhisheng, the founder of speech recognition company Huang Wei, said that he is very much looking forward to the multiple layouts of industry funds that can have a certain amount of shipments of terminal AI chips in the Internet of Things field, and the formation of 'AI cloud chips - terminal applications - government procurement' Ecological closed loop.
Where should an AI killer application be explored? Will there be a general-purpose AI processor that exists as a stand-alone CPU? If so, what is the architecture? If it doesn't exist, where are so many startups going now? Wei Shaojun believes that this issue has not yet been answered. Most of the AI chip founders in the market will become 'martial', and this will become the most admirable and most moving event in AI development.