China's First AI Chip Unicorn Cambricon Bags USD100 Million in A-Round Financing
Qian Tongxin
DATE:  Aug 21 2017
/ SOURCE:  Yicai
China's First AI Chip Unicorn Cambricon Bags USD100 Million in A-Round Financing China's First AI Chip Unicorn Cambricon Bags USD100 Million in A-Round Financing

(Yicai Global) Aug. 21 -- Artificial intelligence chip startup Cambricon Technologies Co. has raised CNY667 million (USD100 million) in its A-round funding led by SDIC Venture Capital with support from Alibaba Capital Partners, CAS Investment Management Co., Turing and two angel-round investors Suzhou Oriza Holdings Co. and Shanghai Yonghua Investment Management Co.

The startup, founded by two brothers who are researchers at the Chinese Academy of Sciences' Institute of Computing Technology, is now valued at USD1 billion.

Cambricon released China's first AI chip, the 1A, last year, stating that it was the world's first commercialized neural network processor chip. 1A is designed for smartphones, security surveillance, unmanned aerial vehicles, wearables and autonomous driving devices, and offers significantly enhanced performance per watt compared to traditional processors when performing mainstream AI algorithms.

"There are a large number of deep learning applications today, but all of them are based on traditional generic processors such as Central Processing Unit (CPU) and Graphics Processing Unit (GPU)," said chief executive Chen Tianshi.

"For example, a few years ago Google used more than 10,000 CPUs to train a cat face recognition model, but neither CPUs nor GPUs are suitable for developing neural networks of a size similar to the human brain."

"Cambrian is a processor specifically designed for AI deep learning, and outperforms traditional processors in graphic and voice recognition by at least two orders of magnitude," Chen added.

"It also has a high integration density several times that of traditional processors, making it possible to install AI chips on mobile devices."

For Chen Yunqi, co-founder and chief scientist at the company, to compare generic chips with specialized processors is like comparing Swiss army knives with kitchen knives. Generic processors are the 'Swiss army knives.' They are designed for various purposes but not for any specific use, which results in a waste of resources. On the other hand, deep learning processors are purpose-built for specific applications, like kitchen knives.

The 1A processor offers two advantages, said Chen. It can substantially enhance AI computing performance by two orders of magnitude compared with traditional CPUs and GPUs, and its offline intelligence feature also ensures optimal information security by removing the need to upload user data.

The firm has a number of processor models, and funding will be used to commercialize their devices and cloud computing platforms. The funds will also aid the development of high-performance, low energy-consumption smart cloud solutions for users.

The devices include smartphones, smart glasses and smart wristbands that require better processors for image, audiovisual and text recognition. On the cloud computing front, Cambricon has attracted many leading players in the field such as iFlytek Co. [SHE:002230] and Dawning Information Industry Co. [SHA:603019], better known as Sugon.

"According to the national strategy for the development of new-generation artificial intelligence technology, AI processors will have wide applications in the security and military industries and video AI algorithms, where demand is stable," said Sheng Linghai, vice president and chip industry analyst at market research agency Gartner.

"There is no need for Cambricon to worry about obtaining orders. As long as the products possess solid performance, they're completely capable of replacing Nvidia Corp.'s [NASDAQ:NVDA] Jetson AI development board."

However, a professional at Nvidia rebutted this claim. "It doesn't make much sense to compare Cambricon's processors with Nvidia's, because the former is an application-specific integrated circuit (ASIC), like Google's tensor processing unit, which was developed specifically for Tensor-flow applications. ASICs have limitations. By contrast, Nvidia's GPU is more scalable, and can be programmed for different applications and be optimized for different algorithms. It can also be adapted to technology upgrades," he said.

Total revenue in the AI-related deep learning chipset market will skyrocket from USD500 million in 2016 to USD12.2 billion in 2025 at a compound annual growth rate of over 40 percent, AI market intelligence firm Tractica forecasts.

Follow Yicai Global on
Keywords:   MSCI,Institute of Computing Technology,Chinese Academy,Cambricon,NVIDIA