Science Technology

At the time when AI vision exploded, who injected a strong “core” agent into China?

With the breakthrough and promotion of deep learning technologies such as convolutional neural networks, AI has been quickly introduced into machine vision applications, and has shown explosive growth momentum in recent years. It has become the largest technology and application in the 100 billion-level artificial intelligence market. Among them, the potential market size of AI vision chips is as high as 22.5 billion yuan. With the continuous promotion and implementation of AI technology and applications, and the blessing of emerging technologies such as 5G/Internet of Things/cloud computing/big data, the potential of AI vision is constantly increasing. Being excavated, the market detonated at the right time.

All enterprises enter the game and sail the blue ocean

The new generation of artificial intelligence technology has endowed the AI vision chip with a new role, no longer limited to command control, but more image structuring and big data analysis, which also makes traditional chip architectures such as CPU no longer meet the needs of AI applications ; At the same time, in a wider range of 3C, industrial weak current and other application fields, the capabilities of the original chips are increasingly tight. Low power consumption, high computing power chips are the correct solution for AI computing.

It can be said that the arrival of artificial intelligence has ushered in a new era, and at the same time it is also restructuring the chip system, and AI vision, which occupies the head position of AI applications, has attracted many companies to enter the game, including Nvidia, Xilinx, Intel, etc. Veteran chip giants include local IC design leaders such as Huawei HiSilicon and Zhanrui, security leaders such as Hikvision and Dahua, and AI algorithm companies such as YITU and Yuntian Lifei, but the most notable one is the birth of Shanghai Qigan Electronic Information Technology Co., Ltd. is an important member of a large number of AI vision chip start-ups.

At the time when AI vision exploded, who injected a strong “core” agent into China?

For a time, the AI vision chip industry showed a good situation in which a hundred flowers bloomed. Due to the strong scene adaptability of AI vision chips, the chips developed by different companies often have unique industry attributes. Therefore, we see that most of the entrants continue to cultivate and develop in their fields of expertise, hoping to make AI vision better. The effect of specialization and landing is more obvious.

Different from many AI chip start-ups, Qigan Technology has a mature business model, clear landing scenarios and a complete industrial ecology. The chip shipments in 2020 have reached tens of millions. Dong Boming, CEO of Qigan Technology, recently accepted OFweek Veken revealed in an interview: “Our products are not only used in cameras, but also in a wider range of fields such as doorbells, door locks, access control, cloud shelves, smart screens, etc. It is estimated that next year’s shipments are expected to exceed 20kk .”

For multiple scenarios, full-stack services have become a trend

With the deepening of the industry, AI vision chip companies began to expand to related scenarios.

The first is horizontal development, based on AI vision, it will advance to a wider range of application scenarios such as smart homes, commercial buildings, smart security, smart retail, autonomous driving, robots, and data centers.

As a big manufacturing country, some of my country’s industrial output value occupies a pivotal position in the world. Among them, the video surveillance equipment market accounts for half of the world, robot shipments account for one-third of the world, smart home market accounts for one-quarter of the world, and data centers The market size accounts for more than one-fifth of the world… What is important is that compared with developed countries, my country’s penetration rate in many fields is still very low. For example, in the field of smart home, the penetration rate in the United States is above 15%, and the global average is 7% .7%, while my country is only 1%; as the world’s largest populous country, the market potential to be tapped is huge.

It is undeniable that different scenarios require different application algorithms, but in fact they have a strong correlation. AI vision has certain versatility, such as smart home, security and service robots, all of which have human body recognition, face recognition, and object recognition. The general requirements for recognition and even behavior analysis provide the possibility for the horizontal expansion of AI vision chips.

Take the QG2101 series chip launched by Qigan Technology as an example. It is developed based on a multi-scenario algorithm model and has successfully landed in multiple scenarios such as smart home, smart security, smart retail, and smart hospital, supporting human body recognition, face recognition, vehicle Identification, pet identification, object identification and other multi-algorithm applications.

At the time when AI vision exploded, who injected a strong “core” agent into China?

The second is vertical development. With the emergence of more and more AI vision products, the manufacturing industry has also put forward higher requirements for the chip supply chain, which requires a complete R&D package. In terms of functions, low power consumption and strong computing power are required. , Good compatibility; In terms of development, it is required to be simple and easy to use, which can greatly shorten the development cycle; In terms of price, it is required to be sufficiently cost-effective; In terms of application, it is required to have sufficient performance, long mean time between failures, and high calculation accuracy.

The actual needs of the industry are forcing AI vision chip providers to continuously strengthen their vertical product lines; but the AI industry has very high industry attributes, and there are not many chip suppliers that can meet the many needs of the market at the same time.

Although Qigan Technology has not been established for a long time, it has developed rapidly in the past five years, and has completed:

① Rich scene algorithm model: covering various visual algorithms such as face, human shape, behavior, and objects, combined with hardware-level neural network algorithm acceleration, it provides intelligent security, smart home, smart retail, gate security inspection, witness comparison, Various scenarios such as robots provide rich algorithm model support;

②Multi-series AI chips and modules for large-scale mass production: NNA (Neural Network Accelerator) and NNU (Neural Network Unit) can greatly improve the performance of computing power; integrated ISP and H. 264/H. 265 compression algorithm, which can not only adapt to the current mainstream encoding and decoding environment, but also greatly improve the efficiency of information flow processing; achieve better and stronger computing power support under the same computing power, and meet the application needs of a wider range of scenarios;

③ Low-threshold software and tool chain development platform: One-stop automatic algorithm transplantation tools can not only greatly reduce the user’s development threshold, but also allow software and hardware to play the best performance;

④Stable and efficient Turnkey solution: It supports real-time detection and recognition of faces at 30fps and human figures at a high frame rate of 300fps, and can complete the operation and start within 300ms, achieving millisecond-level ultra-low-latency response, greatly shortening the terminal, The time-consuming computing of edge and cloud allows AI vision to truly realize the rapid application of “what you see is what you get”.

At present, in the field of AI vision, whether it is chips, algorithms, development tools, or modules, they are basically from different suppliers, and then integrated and developed into products suitable for various scenarios. There are generally long development/debugging cycles, matching It has disadvantages such as low accuracy, poor compatibility, difficulty in exerting computing power to the extreme, and insufficient accuracy of algorithm analysis.

The layout of Qigan Technology has formed a full-stack technology matrix of cloud, edge, terminal, sense, algorithm model, software, and tool chain, and has realized a cost-effective solution for multi-scenario AI vision applications. It is quite rare among visual chip companies.

In addition to providing its own products and technical services, Qigan Technology feels that China has encountered difficulties in the chip supply chain in recent years. Especially during this period, more and more chip devices are in short supply. In order to better meet the market To meet the full-stack service needs, Qigan Technology has integrated its industrial resources accumulated over the years to form an industry alliance covering various components. It has sufficient stocking for various components and partners, and can provide customers with one-stop WiFi chips, NB-IoT, capacitors, resistors, inductors, MEMS and other device products, as well as advanced manufacturing services such as PCB boards and PCBA boards; in order to ensure a stable supply of devices, this model is extremely rare among domestic chip suppliers.

Dong Boming, CEO of Qigan Technology, said: “Our ecological alliance is open, and all excellent suppliers at home and abroad are welcome to join; during this period, we have discovered many excellent local product and technology providers, and we have already ensured that the supply will not be interrupted by foreign countries. supply.” Dong Boming also said that the openness of the alliance is not only reflected in cooperation, but also fully demonstrated in product supply. Customers can customize and select the required components according to their needs, or deliver them in the form of modules. , the alliance will exert its greatest momentum to provide the highest quality products and the most cost-effective services.

Who is in charge of the ups and downs of China that lacks “cores”?

China is a big manufacturing country, but it has not been able to become a manufacturing power. The fundamental reason is that my country’s high-tech industries such as chips and core devices have long relied on international suppliers. Among them, the semiconductor field has not yet built its own strong chip supply chain system. Such an industrial structure can easily get stuck in my country’s manufacturing industry.

With the vigorous development of emerging industrial technologies such as machine vision, artificial intelligence, 5G, and the Internet of Things, it becomes more urgent to solve the problem of core shortage. The arrival of the AI ​​era has opened a window for my country’s “chip” future.

Just when everything was going well, on April 16, 2018, the United States began to comprehensively suppress the development of China’s high-tech industry. Unfortunately, the semiconductor industry became the hardest hit area. A shade.

In October 2019, the second phase of the National Fund began to leverage trillions of capital to increase the size of the semiconductor industry; in August 2020, the State Council issued “Several Policies for Promoting the High-quality Development of the Integrated Circuit Industry and Software Industry in the New Era”, which has become a guide for my country’s technology industry An important platform for change…

So far, the international environment has strengthened China’s determination to develop the semiconductor industry by self-reliance; the domestic atmosphere has revived the vitality of the local semiconductor industry chain.

Qigan Technology, which has just grown up, is fortunate to be one of them. It not only perfectly avoids the risk of US sanctions, but also completes the layout of end-side AI vision chips with the help of the accumulation of the past five years; edge reasoning chips will be launched in the next two years , cloud reasoning chip, and realize the full stack supply of cloud, edge and terminal.

At the time when AI vision exploded, who injected a strong “core” agent into China?

According to Dong Boming, at present, Qigan Technology adopts 28nm manufacturing process, which avoids the problem of being “stuck” in the manufacturing process; facing the complex situation at home and abroad, the demand for localization of chips is increasingly urgent, and Qigan Technology is targeting domestic chips The leader, Huawei HiSilicon, hopes to solve some of the needs of the huge market based on its own strong AI technology strength, chip supply capabilities, full-stack technology matrix, and alliance resource integration capabilities. Dong Boming further explained: “The (domestic chip demand) market is huge, and it is impossible for us to achieve full coverage and only cut a piece of cake.”

It is particularly worth mentioning that Qigan Technology’s AI vision product line will no longer be limited to chips, but will extend to sensors, automotive-level core devices and other fields, so as to expand the application scenarios to autonomous driving, smart cities and other needs. In AI application scenarios with higher computing power and efficiency, we are constantly attacking the goal of being the only provider of AI vision solutions with a complete layout in the world.