Chapter 485 Indescribable Shock

After seeing the two of them off, Xu Chuan returned to the second floor to pack his luggage.

If he followed his past habits, he would usually stay in his hometown until the 12th day of the new year before setting off.

After all, it is rare to come back a few times throughout the year. Spending time with parents and visiting relatives is something that must be done during the New Year.

However, this year's situation is special, so he is planning to leave on the sixth or eighth day of the lunar month.

Although I don’t know how far HiSilicon and Huaxin have advanced chip research, the international situation and the country’s desire for chips are indeed not optimistic.

When chatting with the two in the afternoon, Ding Jingguo, the leader of development and reform, mentioned these things more than once, especially after the realization of controllable nuclear fusion technology, the related situation and development have become more serious.

So Xu Chuan planned to end his vacation early and go to HiSilicon and Huaxin to check on the situation.

If he can solve these mathematical problems and the country can make a breakthrough in chips, he will still be very happy.

After all, the development of science and technology has never been a matter of one person.

Just like the breakthrough of controllable nuclear fusion, it would not be possible without the efforts of other scientific researchers, engineers, and even security, construction and other grassroots workers.

In those areas that are easily overlooked by ordinary people, there are countless people struggling silently.

After saying goodbye to his parents, Xu Chuan took the high-speed train back to Jinling.

Outside the villa at the foot of Purple Mountain, under the leadership of Zheng Hai, a middle-aged man passed the security check with some anxiety and anticipation, and entered the villa that did not look outstanding.

"Academician Xu, the people from Huaxin have brought them here." Zheng Haihui reported in the living room.

Xu Chuan nodded and said with a smile: "Sorry to trouble you."

Zheng Hai: "I'm going out first. If anything happens, you can just call me at any time."

After saluting, Zheng Hai turned and left. In the living room, only the middle-aged man arranged by Xu Chuan and Hua Xin was left.

Seeing the figure sitting in the living room, the middle-aged man felt a little nervous and apprehensive, but also looked forward to and admired him.

After all, this is the legendary figure. In terms of achievements, he is no different from the famous figures who are admired in history.

Xu Chuan didn't pay much attention. He glanced at the middle-aged man standing there nervously. He smiled and said, "Sit down first. Have you brought the data problem?"

"Hello, Academician Xu, my name is Mao Shun. I am the leader of the third group of Huaxin Technology R&D Department. I am mainly responsible for mathematical design. This is the information you requested."

Hearing this, Mao Shun trembled, walked forward quickly, took out a small box from the backpack he carried with him, entered the password, took out a hard drive, and handed it to Xu Chuan respectfully.

Xu Chuan reached out and took the hard drive, looked at it twice, and said with a smile: "Do you understand all the mathematical problems in the hard drive?"

Mao Shun: “I don’t dare to say that I understand, but I understand most of them.”

This is also the main reason why the leader arranged for him to come here this time. His level is not high. In theory, if he wants to connect with this big boss, he should be a vice president-level figure at the end.

However, during the meeting yesterday, Mr. Ren from Warwick said that it would be best to arrange for someone who has knowledge of chip mathematical design to come over.

But when it comes to understanding mathematics, I am afraid that no one in the world dares to say that they understand mathematics in front of this big boss.

Looking at the entire history of mathematics, it is difficult to say where this expert ranks. But in this era, he is undoubtedly the number one person in mathematics.

"Then don't waste time and come with me." Xu Chuan nodded, got up and walked towards the study.

Solving the mathematical problems in chip design is something that should be done sooner rather than later.

Since the person in front of me, Mao Shun, knows something about it, let’s read it over before talking about it. If you have any questions, you can also ask him to explain it on the spot.

In the study room, Xu Chuan connected the hard drive to his computer.

Instead of a laptop, he used a small supercomputer that Zheng Hai arranged for someone to build in the villa next door.

If his laptop might be hacked because it connects to the Internet to query papers and the like, then this specially built small supercomputer has basically no chance of being hacked.

This kind of thing related to the development of national chips should be paid attention to.

Opening the hard drive and entering the password, Xu Chuan found the question file that Mao Shun mentioned and read through it carefully.

Although he does not understand the circuit design and other fields in the chip, he still understands the basics of the operation of the chip or computer.

The foundation of computers is inseparable from three people. Leibniz invented binary, George Boole introduced Boolean algebra, and Shannon described switching circuits.

The combination of these three methods laid the foundation for the development of computers.

Binary is an algorithmic change of decimal. It is the basis for machine operation and the embodiment of the introduction of mathematics into life.

It is undoubtedly the lowest level of computers, because whether it is a simple or complex problem, it ultimately needs to be simplified into arithmetic operations to perform operations through logical switches.

The core is to use Boolean algebra to simplify all operations into AND or NOT operations of 0 and 1.

For example: when adding a+b, the carry is a&b, and the 0 bit after the addition is (a&!b) | (b&!a). That is, AND, OR, and NOT can be used to express addition and subtraction.

Binary, combined with Boolean algebra, can represent all mathematical operations.

Claude Elwood Shannon brought "Symbolic Analysis of Relays and Switching Circuits", which brought binary and Boolean operations into reality through relays, laying the foundation for today's digital circuits.

Today, regardless of the increasing integration of chips, the latest CPUs, GPUs, mobile phone processors and other devices are released, and even the number of transistors has exceeded 10 billion.

But the mathematical logic running inside is still the binary proposed by Leibniz, the Boolean calculation invented by Boole, and the switching circuit described by Shannon.

It has never changed!

Of course, this is just the most basic foundation.

The real operation and design in computer chips today are much more complicated than this.

Including the information that Xu Chuan will read this time.

Although he doesn't know much about chip design, he knows mathematics and modeling.

So he didn't look through those difficult math problems at the first time, but first learned some information through documents.

After all, if he doesn't even understand the situation, he can't solve the problem.

It's like to solve a difficult math problem, you always have to understand where the core of the problem is.

Through the information brought by Mao Shun, he roughly understood the ideas and progress of HiSilicon and Huaxin.

According to the data, the seven-nanometer chips designed by these two companies are based on the ARM architecture.

The so-called Arm architecture is implemented by ARM for various micro-architectures to provide a software-compatible architecture with various power consumption, performance and area combinations. It is one of the three strongest architectures for processors (CPUs) at present.

Purchasing other people's infrastructure and then developing it is a very common thing in the chip industry.

Just like most scholars who study a difficult mathematical problem need to step on the shoulders of their predecessors to move forward.

It is still a very rare thing to construct your own method from beginning to end.

And HiSilicon and Huaxin did this.

They conducted in-depth research on the ARM architecture, continued to optimize and transform it, and established a set of their own NPU neural network architecture.

The so-called NPU (embedded neural network processor/network processor) is a processor specifically used for network application data packets, and it is also a type of integrated circuit.

But unlike the single function of special-purpose integrated circuits (ASICs), neural network devices are more flexible in handling more complex problems.

Generally, software or hardware can be specially programmed according to the characteristics of network operations to achieve special uses of the network, and many different functions can be realized on a chip to be applied to a variety of different network devices and products.

The architecture designed by HiSilicon and Huaxin is better at processing massive multimedia data such as videos and images than traditional methods by adopting the method of "data-driven parallel computing".

It can be said that the younger generation is better than the older generation.

However, the current architecture involves many systematic mathematical problems when constructing it, such as activation functions, two-dimensional data operations, decompression modules, etc.

If it is in a conventional CPU or GPU chip, these problems are not difficult. However, the NPU neural network architecture is different from the previous two in nature, and its mathematical compatibility is completely different.

This results in the performance of the designed chip being even worse than that of 10nm or even 14nm chips.

At present, the difficulty of HiSilicon and Huaxin lies here.

Neural network and machine learning processing are still in the early stage of demand explosion. There are much fewer mathematicians in this field than in traditional CPU/GPU architecture mathematicians. There are not many talents studying this field abroad, let alone in China.

So under no other circumstances, the Development and Reform Commission came to him, hoping that he could help solve the mathematical problems in this area.

After understanding the overall situation, Xu Chuan opened the mathematical file and read it.

"Large-scale sparse matrix eigenvalue calculation"

"Overall solution of initial value problem of nonlinear ordinary differential equation system"

"Initial value problem of first-order nonlinear time-delay differential equation under neural network architecture"

"Parallel eigenvalue of polynomial matrix based on NPU"

"."

The questions passed through his eyes one by one. There were not many problems, but not few. There were about a dozen in total, most of which were difficulties arising from the NPU neural network architecture.

Xu Chuan first opened the questions one by one and looked at them. From the mathematical basis, these problems were not difficult for him, but he did not understand them very well when it came to the NPU neural network architecture.

If he wanted to study, he had to learn the NPU neural network architecture first, understand the general situation, and then do it.

But it was not all these. There were two or three mathematical problems that he should be able to solve now.

No matter how complicated these mathematical problems involved in chip design were, they were not that difficult for him.

No matter how complex the calculation method, model structure and value selection are, they can never be more complex than the high-temperature and high-pressure plasma turbulence model in controlled nuclear fusion, let alone be compared with the seven major millennium problems.

In his opinion, as he has solved two of the seven major millennium problems, it may be a bit exaggerated to describe these problems with elementary school students' mathematics, but at most they are at the level of college students.

Anyway, when he saw these problems, he already had ideas and methods to solve them in his mind.

After spending more than half an hour going through the entire document that Mao Shun brought over, Xu Chuan sat up straight, stretched, breathed a sigh of relief, and leaned on the chair to think.

Across the way, Mao Shun, who had been sitting in the study for a long time with his butt next to half of the chair, immediately cast an expectant look.

But before he could speak, Xu Chuan across from him took out manuscript paper and pen from the drawer, lowered his head again and began to get busy.

Seeing this, he forcefully swallowed the words that had reached his throat.

Across the way, Xu Chuan ignored Mao Shun who was still sitting in the study. He quickly wrote down titles on the manuscript paper with the black pen in his hand.

[Stability analysis of stochastic time-delay differential equations driven by fractional Brownian motion]

"Some definitions of notation and function spaces. Let (Ω, F, P) be a complete probability space with a non-descending σ algebraic family {Ft}t≥0 that satisfies the usual conditions, that is, {Ft}t≥o is It is right continuous and Fo contains all P zero measurement sets. For all t<0, let FL=Fo. The symbol |·| means "

"Then the time-fractional Sobolev space can be defined as its norm: Hs0 (D) = {U∈Hs (d), supp uD}"

"."

[High-precision calculation of eigenvalues ​​and singular values ​​of Vandermonde matrix]

"First given."

"."

In the study, Xu Chuan was busy solving problems.

Sitting at the desk opposite him, although it was not clear what the boss was studying, Mao Shun did not dare to disturb him and could only sit silently and wait.

Time passed like this bit by bit. In front of the desk, Xu Chuan had forgotten that there was another person in the room.

After writing the last line of calculations, he looked at the manuscript paper on the desk and touched his chin.

Three questions, that’s about it.

After clearing away the manuscript papers on the table, Xu Chuan followed his own process to check the answers to the questions.

Although problems of this level are not difficult for him, rigor is still essential.

After reviewing the answers on the manuscript paper, he nodded with satisfaction.

Okay, in theory there should be no problem.

On the opposite side, Mao Shun, who finally waited for Xu Chuan to stop his calculations, thought for a moment, hesitated for a while, and finally plucked up the courage to ask: "Academician Xu, how do you feel?"

Xu Chuan sorted out the manuscript papers on the table and said: "I just took a quick look at it. Most of the problems in it involve your NPU neural network architecture, so I can't give you accurate answers for the time being. reply."

"But I have already solved the purely mathematical problems for you."

On the other side, after hearing this answer, Mao Shun was immediately confused and stammered: "What do you mean it has been solved for us?"

Xu Chuan smiled, stood up and walked to him holding the manuscript paper, handed him the manuscript paper, and said with a smile: "Literally, these manuscript papers are the process and answers. Although I haven't had time to sort and print them out, I think It’s almost the same, you can just take it back later.”

"As for other issues, I just took a look. Most of them are difficulties based on the NPU neural network architecture. I may need to study them."

Mao Shun took the manuscript paper with a confused look and turned it over. The white words were filled with various mathematical calculations.

Looking at the stack of manuscript paper in his hand, he couldn't help but swallow.

It seems that it hasn't been two hours since he walked into this villa, right?

I looked at the time on my watch, it was two thirty-four in the afternoon.

He could clearly hear the sound of his own swallowing. Mao Shun recalled that if he remembered correctly, he should have arrived on time at one o'clock.

Putting aside the time spent on security check, communication, and academician Xu’s reading of documents, it should only be about an hour, right?

It was really hard for him to describe this shock.

As the team leader in charge of mathematical design, he has come into contact with many domestic mathematicians in the past year, including many academicians of the Academy of Sciences.

But this problem that has not been solved for a year can be solved in one hour by this person?

Even if only one problem is solved, isn't this a bit too exaggerated?