Top of the big era

Han Yan and Chen Yan only stayed in the capital for two days. Zhou Buqi accompanied them to watch a Beijing opera, listen to a cross talk, and went to China World Trade Center to buy a lot of things, spending a total of more than 400,000 yuan.

In particular, the two LV bags with a retail price of 160,000 made Chen Yan dumbfounded, and Teacher Han Yan was so frightened that she kept putting off and refused to take it.

With Zhou Buqi's insistence several times, she accepted it beautifully, and the smile on her face never disappeared all day long.

This morning, Zhou Buqi arranged a plane to take the two of them back to the Northeast.

Go back to the company and deal with something important.

Shen Xiangyang's report on AI chips has been typed. It is a total of 13 pages of A4 paper. The various data and technical directions listed on it also clearly explain the cutting-edge ideas and future expectations of AI chips.

A lot of content can't be read very well.

But the general content is clear.

The so-called AI chip is actually an AI accelerator, a module specially designed to handle a large number of computing tasks in artificial intelligence applications. It can exist as a separate chip, and when the technology becomes stronger, it can also be integrated into a general-purpose chip.

According to a case of foresight, Zhou Buqi has a deeper understanding of AI chips - face recognition.

Face recognition is not a cutting-edge technology, it appeared 20 years ago. It has been widely used since the early 1990s. At the beginning of the 20th century, my country's public security departments were able to arrest criminals based on face recognition technology.

But why didn't smartphones really start using face recognition technology until iPhoneX?

Because until this time, Apple's A-series chips finally had bionic functions, that is, the technology of AI chips was integrated into the A-series chips.

The so-called face recognition is a kind of artificial intelligence, a kind of judgment of computer bionics, which requires a huge amount of calculation, which is difficult for general-purpose chips on the market.

Even if there are some Android phones with face recognition, they are fake, it is 2D recognition, and it does not need a real face, you can unlock it with a photo of your face.

3D structured light for face recognition is not too difficult, and many companies around the world can do it.

The hard part is the bionic chip.

Apple is the only one in the world.

With the strong support of AI processing technology, the computing power and processing power of Apple's A-series chips have been greatly improved, and the 4-core chip can easily beat the 8-core chip of the Android machine.

However, in 2008, AI chips are still at the forefront, a new track, and technology companies all over the world are groping forward.

According to Shen Xiangyang's report, the industry has not been able to reach a consensus on the future technical direction of AI chips, and it has been divided into several major genres. One is that the research and development of algorithm models such as deep learning is not yet mature, and the other is that there is still a big gap in the basic theory of AI.

In other words, the AI ​​chip is a real "integration of software and hardware".

The soft level is the AI ​​algorithm at the code level; the hard level is through the chip design of the integrated circuit. The method of chip design and the application of embedded language are based on the AI ​​algorithm at the software level.

The two must be deeply combined and highly unified, and the technology on the software can be displayed through the hardware to develop the most powerful AI chip.

Going deeper... AI chips will involve a wider and more complex field, not only the integrated circuit semiconductor industry, but also deeply integrated with Internet technology and software technology!

For top chip companies such as Qualcomm, Intel, Samsung, and IBM, the most powerful strength is reflected in the "hard" aspect, while the "soft" aspect is lacking.

They are not Internet companies, nor are they software companies, and they have technical shortcomings.

The core of an AI chip is AI. AI technology must be available first, and then related chip designs can be carried out based on AI technology. The most powerful AI chip must have the best AI technology and the best chip design.

The same example, and cloud computing.

Server giants such as IBM, Cisco, and Hewlett-Packard cannot develop a cloud computing platform, and this market is divided by Internet giants. Because the core of the cloud computing platform is the virtualization and scheduling capabilities of the cloud operating system, it is a "soft" technology, and the distributed connections of millions of servers are just supporting facilities.

AI chips, like cloud computing, are a field in which the Internet dominates hardware. It should not be chip companies that dominate, but Internet companies.

For Zhou Buqi and Ziweixing, now is an opportunity!

Shen Xiangyang is in charge of the artificial intelligence laboratory of Ziweixing Research Institute, and the research is on AI technology. It is completely possible to expand technology and combine AI technology with chip design.

Although the two technologies need to be combined, this project is not necessarily more difficult than the 4G general-purpose chip that Li Yinan is going to do.

Because general-purpose chips are general-purpose, they have more functions, more precision, and more complex designs. The AI ​​chip has a single function, and the complexity of chip design is not so high.

What is most lacking in China is the talent in the chip field. In comparison, there are still a lot of AI-related talents. From this perspective, making AI chips is more likely to succeed than making general-purpose chips.

And this is a new track, as long as it is done, it will be the best in the world.

Zhou Buqi's thoughts immediately jumped out of Shen Xiangyang's report.

The first step is a general-purpose chip; the second step is an AI chip; the third step is the combination of a general-purpose chip and an AI chip to launch a bionic chip similar to Apple's A-series chips.

Obviously, for most of the current chip giants, what is lacking is the ability of AI chips.

Ziweixing has a good layout in the field of artificial intelligence. It has Shen Xiangyang, a top scientist, and Lu Qi has been dug from Yahoo. It can be said that it is full of opportunities.

It was almost 6 o'clock in the afternoon, Zhou Buqi called Shen Xiangyang, "Old Shen? Come and chat if you are free."

This is a big deal, Shen Xiangyang rejected the meeting in his hand, and ran over immediately, "Is it about the AI ​​chip?"

"Yes, I have decided, this project must be done!"

"Really?"

"Well, it has to be done, this is our chance."

Shen Xiangyang took a deep look at him, "This direction is more risky than cloud computing, and there is a lack of relevant supporting facilities and enough talents in China. If you want to make AI chips, you need someone who is proficient in both soft and hard technologies." Top talents must understand both artificial intelligence and chip design.”

"And you?"

"I can't," Shen Xiangyang shook his head, "I lack research on the direction of chip design. But Li Yinan is okay, his Internet knowledge is not top-notch, but it is pretty good."

Zhou Buqi waved his hand, "Anyway, this matter is settled, and it will be launched immediately. Ziweixing will make AI chips; Astar will make 4G general-purpose chips. Gemini, one is indispensable."

Shen Xiangyang had already guessed the result.

This is in line with his perception of Boss Zhou.

After a moment of silence, he said slowly: "This will be a big project. The country lacks not only relevant industries, but also relevant talents. I think it should be divided into three steps."

"you say!"

"First, let Li Yinan join Ziweixing's technical committee and serve as the vice president of Ziweixing Research Institute to coordinate and improve the technical cooperation between the two parties."

"good!"

This is of course no problem.

Although they changed their names, Ziweixing and Asda are a real family. Although it is not a group company, it is a holding relationship.

Just like Ali and the split Ant, they are two companies in terms of ownership structure, but in terms of management system, Ant is still in Ali's organizational system.

Shen Xiangyang said: "Second, cooperate with Tsinghua University. I talked to Li Yinan, and he also went to the Institute of Microelectronics for investigation. Making chips is a long process, and it is good to have results in five or ten years. If there is a shortage of talents, we will cultivate them from scratch. We can design professional courses according to our needs."

Zhou Buqi twitched the corner of his mouth.

Depend on!

Let's assign tasks again.

For this kind of matter, Boss Zhou must come forward to coordinate. Not only must he communicate with the school, but he must also negotiate with the central government and the competent department of education.

To start a new major in Tsinghua, and it is this level of high technology... at least it will cost tens of billions of dollars. To invite top professors at home and abroad, you must rely on national power to promote it.

Shen Xiangyang also knew that it was very difficult, and said: "If you can't set up undergraduate majors, you can set up master's and doctoral majors. Attract undergraduate students majoring in computer science and integrated circuits."

This is much simpler.

It doesn’t even need the state’s promotion, Ziweixing can do it with a little money, Zhou Buqi’s face looks much better, and he nodded slightly, “Well, we will organize a technical delegation later to contact Tsinghua University.”

Shen Xiangyang said: "The third point, which is the most important point, is to go to Silicon Valley to develop, to establish a biocomputing laboratory and an AI chip laboratory in Silicon Valley. Bring back the research results there, and cooperate with the domestic ones. Research docking can also be applied in teaching.”

To put it bluntly, it is money.

Ordinary engineers can only do some ordinary technical positions, and they are all top engineers working in scientific research laboratories, which means that the per capita salary of this type of laboratory is very high.

According to a preliminary assessment, a scientific research laboratory with a team of 50-60 people in Silicon Valley will have annual salary and welfare expenses of 40-50 million US dollars.

Previously, it had been planned to establish four laboratories in Ziweixing American Research Institute: Cognitive Computing Laboratory, Artificial Intelligence Laboratory, Deep Learning Laboratory, and Big Data Laboratory. Now there are more biological computing laboratories and AI chip laboratories.

Annual expenditures for the institute may reach $300 million.

Zhou Buqi has become numb to money figures, and doesn't feel the difference between 200 million US dollars and 300 million US dollars.

As long as the trend of the mobile Internet is captured, making money will be faster than robbing a bank.

This research and development funding is nothing at all.

Artificial intelligence is the next wave of huge opportunities after the mobile Internet.

Get in early.

Zhou Buqi waved his hand, "That's it! Let's get started!"