Canaan Technology's Zhang Nangeng: The future global chip center will be in China

At present, mainland China is the world's largest chip market, and will definitely be the center of chip design, manufacturing, and use in the future. Domestic integrated circuit talents will be the best in the world.

Text｜Reporter Lian Ran from China Entrepreneurs

In 2011, when Zhang Nangeng first came into contact with blockchain, he believed that it had a 1% chance of changing the world. At that time, he was still a master's student at the Beijing University of Aeronautics and Astronautics. He did not hesitate to establish the research direction of blockchain ASIC (application-specific integrated circuit) chips.

A year later, Zhang Nangeng, who had not yet graduated, decided to drop out of school to start his own business. He judged that the possibility of blockchain changing the world had risen to 5%, and he could not wait any longer. In April 2013, Zhang Nangeng and his friends invested a total of 100,000 yuan to establish Canaan Technology in Beijing, deciding to make blockchain chips, with Zhang Nangeng serving as chairman and CEO.

"However, people are no longer discussing how much blockchain has the potential to change the world. This is a consensus," Zhang Nangeng told China Entrepreneur. "After so many years from the Internet to mobile Internet, now is the time to inject fresh blood. In the AI ​​(artificial intelligence) computing platform, Canaan Technology has accumulated advantages in providing super computing power. Ten years later, we hope to provide efficient and low-cost super computing power to achieve the 'silent' effect of edge-side artificial intelligence."

From starting a business in 2012 to developing high-performance dedicated computing chips and releasing the world's first blockchain computing device based on ASIC chips, to the team reaching the 16-nanometer process node in semiconductor technology in 2016 and being able to mass-produce corresponding chips, Canaan Technology has found its own position. Now, Zhang Nangeng and his team are beginning to think about the future direction.

Mobile phone chips? Accelerator chips for genetic engineering and the pharmaceutical industry? After a long discussion, the team finally determined that AI is a direction worth working on for at least the next decade. Among them, how to obtain computing power efficiently and at a low price will be a key issue in the industry. Computing power is exactly the advantage that the team has accumulated before, so they decided to start promoting the development of AI chips.

Zhang Nangeng believes that China has become the largest consumer market for chips, and now is a great time for the development of domestic chips. Policies, funds, and technologies are all moving towards this product. In the future, mainland China will definitely become the center of chip design, manufacturing, and use; domestic integrated circuit talents will be the best in the world.

Iteration

Since the development of the first generation of chips began in 2013, Canaan Technology has made continuous breakthroughs in chip investment records in the past six years: from 110 nanometers to 55 nanometers, 40 nanometers, 28 nanometers, 16 nanometers, and then to 7 nanometers, and 28 nanometer AI chips.

"There is no shortcut to accumulating teams and technologies in the chip industry. It is built up generation after generation and wafer by wafer," Zhang Nangeng emphasized.

In 2016, Zhang Nangeng and his team began to design the first generation of AI chips, which was equivalent to applying the computing technology accumulated in the blockchain field to the development of AI chips. Zhang Nangeng believes that in the past few years, their research and development in the field of blockchain chips has had a great impact on the design methodology of the entire semiconductor design industry, and even on the back-end supply chain, but it is not known to the public.

From the perspective of design methodology, when developing blockchain computing chips, the pursuit of computing energy efficiency and low cost is extremely extreme. The results achieved are orders of magnitude higher than traditional design methodologies. Based on this premise, the focus of the chip competition will be mainly on computing power.

From the perspective of the supply chain, for example, more advanced packaging technology was rarely used in the mainland before. It is a higher-performance technology, but it requires slightly higher equipment investment and one-time costs. It was rarely used in China a few years ago. After the blockchain industry took off, the usage was so large that the production capacity was increased, and the price of packaging technology was reduced. Other types of chip designs also benefited from it.

Zhang Nangeng told "China Entrepreneur" that for AI, there may have been more cloud chips in the past few years, and everyone was pursuing replacements for foreign GPUs (graphics processing units) and so on. Since about 2017, the industry has generally turned its attention to the end side, but the intelligent features of the end-side devices have not actually kept up. There are two reasons for this. One is the problem of algorithms and data; the other is that there is currently no good AI computing platform on the end side.

Why do we do edge AI? Zhang Nangeng explained that there are many things that do not need to be sent to the cloud for processing, or it is unsafe to send them to the cloud for processing. How can this problem be solved? Some people think that strengthening data encryption or strengthening server security management and control can solve this problem, but in fact this solution has limitations.

Zhang Nangeng's opinion is to solve the problem from the root. For example, in the scenario of monitoring the home, for security reasons, an end-side AI camera should be used to determine whether there is a problem at home. If there is no problem, the video data should not be transmitted to the cloud. This is a means of protecting privacy on the end side.

However, one fact is that edge AI is currently in a very early stage, and Canaan Technology's AI chips are now competing more with traditional chips. "In the future, when edge application scenarios are gradually fixed and algorithms and architectures are relatively stable, the key to AI chip competition will be lower energy consumption and higher computing efficiency," Zhang Nangeng predicted.

In September 2018, Canaan Technology released its first-generation chip, K210, which is a system-on-chip (SoC) that integrates machine vision and auditory functions. It is positioned in the two fields of AIoT and focuses on high cost performance, low energy consumption and excellent localized computing capabilities. It is mainly used in the AIoT market, providing more options for solution providers in the small smart home and smart education industries.

After the chip is released, mass production capability is critical. Because the supply chain management system is very complete, from the front and back ends, packaging and testing to the bottom layer and wafer fab cooperation, process improvement and circuit design and even deeper things, there are corresponding teams responsible, Canaan Technology has never encountered problems in chip mass production.

"To put it simply, Canaan Technology can complete all aspects of chip design within the company. This also shows that technology is our core advantage," Zhang Nangeng concluded.

Innovation

Technology is always accompanied by innovation. Innovation is easy to say, but it is not easy to carry it through. Zhang Nangeng emphasized that everything in Canaan Technology is centered around innovation.

Zhang Nangeng admitted that in the past few years, perhaps because they were young and had less experience, he and his team were more likely to think outside the box and use innovative methods to solve problems. This would be risky, but it could indeed achieve designs that were far superior to those of competitors and bring better products.

In 2016, Zhang Nangeng and his R&D staff tried various traditional methods to improve the chip's performance-to-power ratio, but none of them worked. Zhang Nangeng recalled: "We reached a turning point - the factory provided a set of libraries for us to design chips, and the voltage of the library was fixed. In order to achieve lower power consumption, we needed a lower voltage, but the library no longer provided it. So our team completed all the processes ourselves."

"In the past few years of entrepreneurship, we have been moving forward step by step, breaking through the original boundaries in everything we do," Zhang Nangeng emphasized. "The final result is the opposite. When we make AI chips and other computing-centric designs, these previous efforts have become our technological barriers, making it difficult for others to catch up."

The new opportunities facing Canaan Technology are: the overall computing architecture is undergoing major changes. Moore's Law is failing. After the failure, innovative computing architecture is needed to solve the problem. Just like putting AI algorithms into chips with innovative architecture, you will get unbelievable performance. Why can a computer be solved with such a chip and a battery? Because the architecture is different.

But Zhang Nangeng also pointed out that the risks involved are very high. We cannot follow the path others have taken, so we have to forge our own path. And we need sufficient funds and technical strength to do this. "However, this transformation will last for the next 20 to 30 years, and in a sense, everyone's starting line is relatively close."

Zhang Nangeng also pointed out that since IC design is a computing-intensive industry, traditional chip companies are very weak in IT, and many companies still use computer room structures from the 1980s and 1990s. Canaan Technology is one of the first companies to put IC design on the cloud. He believes that when a company needs more computing resources, it does not actually need to build a computer room because the utilization rate is not high. Canaan Technology now uses an automatically scalable cloud computing architecture, which is very rare in the IC industry.

Canaan Technology is also making breakthroughs and innovations in its business model. Chips belong to the hardware industry. Zhang Nangeng believes that the reason why the hardware industry has not received a high valuation in the capital market is that hardware is a one-time transaction. Once it is sold, it is sold, and there is no follow-up. "It is not like the Internet industry that can continuously obtain income from customers. But I think starting a business in the hardware or chip field requires changing this business model."

Zhang Nangeng analyzed that chips are the foundation of computing platforms, and the software and algorithms on the upper layer belong to sustainable services. Going further up, there are customers. Customers may have more demands, such as platforms that provide cloud services, equipment for passenger flow analysis, data statistics platforms, etc. These are actually very much inclined towards the Internet. With such a whole set of services, there is no need to make money from selling hardware, but to provide Internet-based services around customers.

In order to meet customer needs, Canaan Technology has a variety of service methods. For example, in terms of chips and chip-related software algorithms, Canaan Technology mainly provides services to customers; in terms of cloud services, Canaan Technology will cooperate with AWS, Alibaba Cloud, etc.; as for specific businesses, such as how to analyze these data and how to make decision-making recommendations to customers, this is a highly professional matter, and Canaan Technology is less involved. Canaan Technology's philosophy of serving customers is to give full play to the advantages of chips and computing, and integrate this part well, and ultimately deliver to customers a service ecosystem based on AI chips.

"At first we thought selling chips was a simple matter, but later we found that the software and algorithms for AI chips were difficult," Zhang Nangeng revealed to China Entrepreneur. "AI chips are a brand new product, and no engineers can develop them, so we have to do it ourselves. We not only provide chips, but also software, algorithms and solutions. Later we found that since we already had all of this, it would be better to package it directly into hardware modules for sale, which is more convenient to use. Later we also developed a SaaS platform, and our product line is more complete. Of course, the key is to provide customers with a complete solution."

The diversification of products has made Canaan's customers more diversified: some buy chips, such as many developers; some buy hardware modules, such as door lock manufacturers that want to upgrade to facial door locks, which require facial recognition modules; there are system distributors, such as Softcom Wisdom, and we need to provide complete products during the cooperation process, and they will build smart cities or community properties; in the developer community, we cooperate with Baidu AIG, mainly on PaddlePaddle (a deep learning platform), and customize hardware modules for them; some customers, such as Alibaba, are also using Canaan's chip K210.

era

China is currently the world's largest chip market, with annual chip imports reaching hundreds of billions of US dollars. According to data released by the World Semiconductor Trade Association, in 2018, China and the Americas have become the world's top two semiconductor consumer markets, accounting for 32% and 22% of the market respectively.

Zhang Nangeng said that the domestic chip environment is relatively good now, and many policies are tilted, including national industrial plans, which have invested a lot in the chip industry. In addition, mainland China is the largest chip consumption market, and production capacity is bound to exist around the market. The center of chip manufacturing and technology is gradually shifting to mainland China, which is a very obvious trend. However, it is worth noting that mainland China's investment in talent training in the field of chip design is a bit lacking, and now more attention is paid to manufacturing.

"In terms of chip design talent, we can compare China with the United States," Zhang Nangeng analyzed. "The former is obviously lacking in talent quantity, quality and experience, while the latter is experienced but also faces a lack of successors. As far as I know, integrated circuit and chip design are not particularly attractive industries in the United States. In the future, mainland China will definitely be the center of chip design, manufacturing and use. In the next 5 to 10 years, domestic integrated circuit talents will be the best in the world."

However, opportunities always bring challenges.

Zhang Nangeng said that after the country began to pay attention to the chip industry, especially in the field of AI+chips, too much capital poured in, and the cost of talent was pushed up a lot. In the past three years, the salary of IC and AI talents has more than doubled, and the cost of enterprises has risen. In this case, people will also be impetuous, which will affect the stability of the company. Therefore, Canaan Technology is more about training talents by itself, "the master leads the newcomers to make products", and the company pays tuition for the newcomers to learn. Although the timeline is extended and the cost is increased, it is very beneficial to the stability of the team.

Zhang Nangeng has always been very concerned about the underlying technology. In addition to managing the company, he usually focuses more on products. "Ever since I was in school, I have considered myself a product person, but at that time there was no team, and I was responsible for everything from R&D to supply chain, from hardware to software, design and structure. Now that the team is complete, I still have to go back to product development. Accordingly, in terms of technology pursuit, I prefer application rather than geek route."

"In essence, Canaan Technology is a company that provides super computing power solutions. Our next-generation chip, Canaan K510, is expected to be released at the end of this year. The computing power will be greatly improved, and the application scenarios will also extend to smart retail, smart driving and other fields." Zhang Nangeng finally revealed.