Canaan Creative has completed its pre-roadshow and is expected to list on Nasdaq on November 20

According to exclusive sources from Unicorn Early News (ID: iponews), Canaan Creative has completed its pre-roadshow and is expected to be listed on the Nasdaq on November 20.

On November 12, news that Canaan Creative Co-Chairman Kong Jianping had started sending invitations spread widely on the Internet. People related to Canaan Creative said that the company absolutely did not send invitations and "focused on research and development."

Kong Jianping himself also posted on his WeChat Moments, "What does this invitation look like? If anyone has one, please send it to me so I can study it."

The edge AI chip K210 independently developed by Canaan has already achieved sales scale. As of September 30, 2019, the company has delivered more than 53,000 chips and development modules to AI product developers, and cooperated with more than 30 AI algorithm companies to develop overall AI solutions for end consumers.

The prospectus shows that in September 2018, Canaan officially released its first-generation AI chip, the K210, and officially started the commercialization of the chip in March this year. The release of the chip also means that Canaan is the world's first company to master the independent intellectual property rights of commercial edge AI chips based on the RISC-V architecture.

From R&D in 2016 to large-scale commercial use this year, Canaan has finally reaped the fruits of its three-year self-developed chip road. At the same time, Canaan has also clarified the AI ​​chip R&D route. The second-generation chip K510 is expected to be mass-produced in the first quarter of next year, and the third-generation chip is suitable for both cloud and edge scenarios.

RISC-V architecture is fully applicable to edge scenarios

Since 2016, breakthroughs in deep learning have pushed artificial intelligence into the industry, thus kicking off the battle for AI chips.

AI chips are mainly used in two application scenarios. One is the cloud, where the training of algorithm models requires the throughput of massive data and is usually deployed in data centers to obtain sufficient computing power. The other is the edge scenario, where devices need to perform data inference based on trained algorithm models, mainly aimed at improving computing efficiency.

Edge computing refers to the side close to the source of objects or data, using an open platform that integrates network, computing, storage and application core capabilities to provide the nearest end service. Its applications are initiated on the edge side, resulting in faster network service response, meeting the basic needs of the industry in terms of real-time business, application intelligence, security and privacy protection, etc.

From a technical point of view, due to the technical advantages of RISC-V such as simplicity, low power consumption, modularity and scalability, it is particularly suitable for edge computing fields such as closed or semi-closed applications with relatively low ecological dependence, embedded artificial intelligence, etc., bringing significant flexibility and cost advantages to the Internet of Things industry.

Anchoring the edge market and embarking on the road of self-developed AI chips

Considering the power consumption and computing power restrictions of edge scenarios, Canaan has been developing chips based on the RISC-V architecture from the beginning, so that its chips can flexibly expand instructions according to actual scenario requirements and enhance the programmability of products. At the same time, thanks to the lightweight characteristics of the RISC-V architecture, the K210 can process data at lower power consumption.

Thanks to the open source nature of the RISC-V architecture, Canaan is able to conduct research and development of independent intellectual property rights on this basis.

Among them, the core network accelerator KPU of the Kanzhi series chips is completely based on self-development, which can not only accelerate the reasoning of CNN, but also accelerate the convolution-based operators of traditional image algorithms. At the same time, KPU supports algorithm models such as TinyYOLOv2, and can provide an exploration platform for cutting-edge technology directions such as TinyML.

Canaan stated in its prospectus that the development of artificial intelligence technology, especially technology related to edge computing, and the acceptance of ASICs in artificial intelligence applications are crucial to the company's future success in product diversification.

Canaan chose the edge market as the entry point for AI chips mainly due to two reasons: On the one hand, the cloud chip market is overheated, and giants such as Nvidia, Google, and Amazon have launched their own cloud AI development platforms and acceleration solutions, and have also begun to extend to the end side, such as the TPU released by Google. On the other hand, compared with the cloud, the edge market is still a blue ocean, especially in China, where many so-called end-side smart devices have not yet been able to perform tasks based on AI algorithms.

Kanzhi K210's landing capabilities are beginning to show

Regarding the implementation of AI chips, Zhang Nangeng, chairman and CEO of Canaan, once said, "We must make profits when we make things. Canaan is not doing it for fun. Companies and enterprises must make money, and they must make money from the business itself, not from government subsidies. The same is true for AI. Canaan is also exploring a profit closed loop. Our first-generation chips are trying their best to fit the application scenarios and help customers solve practical problems."

The commercialization of Canaan AI chips was officially launched in March this year. As of September 30 this year, the company has shipped more than 53,000 chips and development kits to AI product developers within half a year. Its first-generation AI chip, K210, has been deployed in smart buildings, smart homes, smart energy consumption, and smart agriculture and forestry.

In the field of smart buildings, the sensorless access control system developed by Canaan fills the gap in domestic technology in this field to a large extent. The access control system is based on the first-generation edge AI chip K210 independently developed by Canaan, supports offline data storage and processing, can store 10,000 facial information and 20,000 door opening records on the edge, and directly uploads text results such as facial feature values ​​in a local computing mode, without the need to build a backend server cluster and private cloud.

Image source: Zhihu

In the smart home scenario, user privacy is a challenge currently faced by AI devices. The smart door lock module provided by Canaan can extract facial image feature values ​​directly on the client side without uploading to the cloud. Since the power consumption of the K210 is only 0.3W, the battery replacement frequency is about one year, so it can also be implemented on a large scale in scenarios such as hotels.

In Asia's largest community, Nanming Huaguoyuan in Guiyang, high-rise buildings are densely distributed, the service area is large, and there are many floating populations. The traditional meter reading method cannot fully grasp the community's energy consumption data. In order to solve the problems of "high cost, low efficiency and difficulty in accessing households" in the traditional meter reading mode, Canaan Technology deployed the Smart Things Run-in, which uses an external edge computing device to process electricity/gas data in real time through the energy consumption meter CNN algorithm.

In the business scenario of controlling forest pests and diseases, Canaan cooperated with Baidu and Beijing Forestry University to insert audio processing hardware equipped with an 8-channel high-performance microphone array into trees, using the sound of insects biting trees as the sound source to determine the location of pests.

At the same time, Canaan makes full use of the visual capabilities of the K210 chip, placing the chip in a 40mm square smart box, and uses image classification and detection methods to determine whether there are pests in the field of view. This audio-visual comprehensive judgment method effectively improves the efficiency and accuracy of judgment, and has a wide range of application scenarios in forestry and farmland.

Clarify AI chip R&D planning

Regarding future AI chip research and development, Canaan disclosed its roadmap in the field of artificial intelligence chips in its prospectus.

The prospectus shows that Canaan is currently developing the second-generation 28nm AI chip. Compared with the previous generation of products, the new chip is developed for 5G scenarios and has greatly improved computing power and energy efficiency. It will be used to explore new scenarios such as smart retail and smart driving. The mass production of the second-generation chip is expected to start in the first quarter of 2020. In addition, the company also plans to launch the third-generation 12nm AI chip in the second half of 2020, which is expected to be suitable for edge and cloud computing.

Canaan said the company's mission is to enable everyone to enjoy the benefits of supercomputing and to enable widespread adoption of blockchain and AI technologies to improve people's lives.

This article is from "Unicorn Early Knowledge".