Tech Hammer

At the Huaxia Future Intelligence Research Center, Ning Wei is leading his students to do the final verification in a specialized laboratory.

It is said that because the complexity of three-dimensional silicon silicon tube chip technology has basically reached the extreme of modern chip design, chip verification even accounts for 70% of the entire chip design process.

Especially four months ago, after the first tape-out failed, Ning Wei took a group of students to the Ning Si Laboratory in Jiangcheng and held meetings for three days. And I witnessed with my own eyes that the big guys in charge of each link actively carried out criticism and self-criticism...

Ningsi Laboratory moved from Shenzhen to Jiangcheng two years ago. It's not just Ning Wei's suggestion, but also the conditions offered by Jiangcheng are really good, and it seems that Huawei can't hold on any longer.

Although Ning Wei has never cared about how much money he spent on designing a three-dimensional silicon silicon tube CPU, it is certain that the investment is huge.

As a result, this project only lasted a year, and Huawei had to introduce third-party strategic investors. As a result, Ningsi Laboratory was separated from the 2012 Laboratory as a whole. Later, Jitu Group and Huaxia Precision invested in Ningsi and occupied 37% of the laboratory's shares.

When Mr. Yan called him to report this matter, Ning Wei also sighed and asked a few questions about how to introduce strategic investors at this time. After that, Yan Ming's explanation made Ning Wei feel a little ashamed.

Probably the group has made calculations. If the chip under design can be put into production in five years, then based on this year's investment, without affecting the normal investment in research and development of other businesses, Huawei can probably only sustain one year. In the middle of the year, if other business research and development are suspended and the research of Ningsi Laboratory is fully guaranteed, and most of the credits granted by major banks are used up, it will probably only be enough to last another two and a half years...

And it has to be guaranteed that it must be mass-produced in two and a half years.

This does not include the various systemic impacts caused by the suspension of R\u0026D investment in other businesses and the impact on performance.

You must know that it is not a simple matter for a laboratory to recover after it is stopped, let alone the research and development of its main projects.

As a company decision-maker, it is no choice to introduce strategic investors in advance.

At that time, Ning Wei was stunned by these words...

Well, at that time, Ning Wei really didn’t realize that Ning Si had grown into a large R\u0026D team with more than 12,000 people around the world in one year. Almost all of these teams were cooperating with his understanding of the chip, outlining and verifying the chip. The structure and function of three-dimensional chips.

In addition, in order to better cooperate with Ningsi Lab, investment in minimalist EDA has also begun to increase exponentially.

There is no way, except for minimalist EDA, there is no EDA software that can support circuit design on three-dimensional structures, and everything is explored from scratch, and most of the past experience and design composition are not available.

In other words, everyone has spent a lot of money, and the pressure can be imagined.

In particular, the first tape-out not only failed, but also exposed many problems. For example, the finished chip had serious stability problems in the high-frequency state, and if the EM failed, there was a risk of unstable chip life, and the LVS stress test failed. Stable, the pin test is always open...

A severe blow to everyone's morale.

This also forced Ningwei to devote all his experience to the chip in the past few months and began to examine the reasons for so many problems. The final summary is also very appropriate. The chip testing team still lacks understanding of the three-dimensional structure. All tests are based on past experience. This leads to the fact that although the test on the two-dimensional layout can pass, the three-dimensional structure fails. Various problems will arise.

For example, in the eyes of LVS, the reason why the test is unstable is that a test port layer is directly connected to the operation layer, but not connected to the interface inside the chip and chip packaging. Under normal circumstances, it should be a signal inside the chip. For example, the 1.2V signal should be converted from 1.2V to 3.3V after passing through the interface, then enter the PAD opening, and then connected to the package through the bonding gold wire, and finally reach the chip and be soldered on the PCB. of pins.

During design inspection, because the gold wire connection in the three-dimensional design was too complicated, this error was not discovered according to the conventional chip inspection method, which led to problems during the final tape-out.

The solution was that Ning Wei personally re-drafted the contents of the check-list, directly increasing it from the previous more than 300 items to 1,207 items. This time, I have thought of all the problems exposed during the first tape-out, and also listed some problems that may be hidden in advance.

There is no way, this thing is too expensive to tape out.

Intel may only need a few hundred million yuan for a tape-out. After all, the technology was very mature in the past, but new technologies are different. The total cost for a tape-out is about 3.5 billion, nearly 4 billion.

Don’t be too expensive. After all, for Huaxin Precision, which is responsible for the tape-out, the investment in cooperating with the tape-out is also huge. It is said that in order to cooperate with the tape-out, the engineers who adjusted and modified the equipment parameters also did not take a break for more than half a year. Everyone is working hard to meet the supporting technical parameters and ensure that progress will not be delayed due to equipment problems.

Of course, if you tape out multiple times, the price of a single tapeout can probably be evened out, for example, if you tape out eight times out of ten, but no one can afford such a hassle.

After all, the company cannot live by its ideals. The more than 10,000 people in Ningsi Laboratory, whose average monthly salary exceeds 40K, cannot be fed by the northwest wind. Although Ningwei himself has no pressure, he cannot watch his partners really because The chip design and production he led were completely wiped out...

The net investment of more than 50 billion a year, with no immediate return on sight, is enough to make the helmsman suffocated by the pressure. The last time Ning Wei saw Yan Ming, he discovered that the vice president had started to go bald recently. Yes, the hairline has moved back at least two inches. After the CPU design is successful, I am afraid that I will have to have a hair transplant.

At this time, Ning Wei had re-uploaded the overall design that had been revised after four months to the server and began to do the final integrity verification. The three students he had just brought had also been busy for ten years in accordance with his requirements. For many hours, all the parameter standards were set, and March wasted a huge amount of computing power to simultaneously start the digital simulation of the three-dimensional simulation circuit...

Now the master and apprentice finally breathed a sigh of relief, eating pastries and chatting in the lounge.

No one wanted to rest. Although they knew that the entire verification process in the final stage would last more than thirty hours, the dominant emotion at this moment was excitement.

The first true desktop-level CPU with completely independent property rights in Chinese history, the first true three-dimensional structure desktop-level CPU in human history, was completed by myself. As long as it is successful, it will definitely be recorded in the history books in the future.

From this point of view, we can see the benefits of choosing a great mentor. If you are lucky, if you work on a certain project, you may be able to leave your name in history. Of course, the odds are low.

"Director Ning, have you read the news? Intel's 12th-generation and 13th-generation CPUs have completely cut prices. This time it's real. Even the motherboards have also been reduced in price. It is said that the 14th-generation CPU is about to be released. Warm up the 14th-generation CPU in advance. !”

"Oh!" Ning Wei, who was drinking milk tea, responded casually.

Milk tea is a loving milk tea made by Jiang at home.

Because I would rather drink it, I don’t feel comfortable buying it in a milk tea shop, so I just make it myself. I don’t want to ask too much about the specific method. It probably means boiling fresh milk, adding sugar, tea leaves, starch, etc., and finally filtering the tea residue and drinking it.

How to put it this way, because the little Ning Ke in the family likes to drink very much, he can’t drink much at a time and is always greedy. It’s impossible for Jiang to satisfy the little Ning Ke’s requirements all the time, so Ning Wei simply made a Tetra Pak packaging craft in the backyard. When it comes to machines and sterilization equipment, Classmate Jiang can fill twenty or thirty boxes at a time. Ning always brings a few boxes with him when he goes to the laboratory. If he misses the little one, he will bring one box.

Of course, the main reason is because of being rich and willful. Just as Liu Wei predicted before, after the birth of little Ning Ke, Ning Wei finally enjoyed the fun of being rich...

Although little Ningke is only over two years old now, he already has a toy room of his own, with various educational and non-educational toys. Basically, if Ningwei likes Ningwei, he will buy it back.

In addition, there is a big screen in the toy room. There is a cat resident in the big screen. As long as it prefers to be in the room, the cat is always ready to interact with the little princess, even if it means "beep" at the screen. , the kitten will also blink its eyes and meow twice, giving the little princess a perfect feedback.

More often than not, this kitten also takes on the responsibility of teaching when the children of Ning Ke express their curiosity. For example, the little princess Ning Ke can now call out the names of several animals and recite several ancient poems. I can count from 1 to 20 thanks to this cat.

Ever since it was discovered that this cat’s patience and effectiveness in teaching children are far better than those taught by adults, both grandparents, mothers, and grandmothers have given up the task of teaching children to read. As for dad, when he is free at home, he He only focused on playing with Xiao Ningke. After all, Ningwei didn't have much time to rest or play, so how could he think about teaching his children knowledge?

"Director Ning, you can't underestimate people too much. Look, the Intel i9 13900 has dropped by another 3,000 yuan. Now you can buy it for only 6,200 yuan. The i7 13700K only costs 1,800 yuan, and the i5 13400 has dropped to 999 yuan. I don’t believe it if they say it’s to make way for the 14th generation. Did Intel receive some rumors that our CPUs may be in mass production soon, so they will directly start the price reduction mode and harvest them first?” Zhou Zhenghao couldn’t help but emphasize.

As one of the first batch of direct Ph.D. students recruited by Ning Wei, after studying with Ning Wei for a year, Zhou Zhenghao directly chose the direction of smart chips when writing his thesis, so he was more concerned about industry trends. Of course, it was also because he had been following Ning Wei for a long time. After signing the contract with Huawei, he will still settle in Ningsi Lab after getting his doctorate. This also means that the market response to the first-generation CPU directly determines his future career development plan, so he naturally pays more attention to market dynamics.

This is also the reason why Ningwei always brings three students, Meng Runze, Lu Ying, and Zhou Zhenghao, to make CNMD architecture CPUs. It is not because of preference, or because these three students study best among all the students, Meng Runze is the main one. Both Lu Ying and Lu Ying also chose the development direction of intelligent hardware.

Others' main topics are software, algorithm and language oriented, and they are still deeply involved in theory, and their research directions are not related.

"Why are you so anxious? Isn't this normal? We are making so much noise, it would be strange if Intel didn't know about it. Price cuts of this scale are of course normal, and the intention of suppressing them must be there. Now consumers are changing CPUs, and in the future Our CPU is out, and whether we should replace it in the personal application market is indeed a problem. Our CPU pricing is also an issue. If I were to replace it, Pat Gelsinger would also choose to reduce the price at this juncture. It happens to be the Spring Festival. , and it can also disrupt the holiday market. This fully demonstrates that Intel's market analysis and marketing department is very qualified."

Ning Wei put down the milk tea, moved his eyes away from the monitor, and helped Zhou Zhenghao analyze it carefully.

"What should I do? If my CPU is launched together with Intel's 14th generation, do you think they will also set a super low price for 14th generation?" Zhou Zhenghao asked worriedly.

This time, without waiting for Ning Wei to answer, Meng Runze next to him answered naturally: "Zhou, isn't this normal? It is certain that the price of 14 is lower than previous new products. I took a look at the recent Intel In the patent library applied for, there have been no major technological breakthroughs in the past two years. 12 is so amazing, and 13 is so impressive, as you know, I guess 14 will not be that eye-catching. This is why when 13 was released, our CPU was still The design has not been completed, otherwise the price reduction would not be possible until now.”

"Second the proposal!" Lu Ying, a man of few words, also expressed his opinion.

"Actually, I know that, but I just feel a little scared. The money that has been poured in can at least build four or five aircraft carriers. Now we have reached the final stage. If the market does not perform well in the end, I don't know if Ning Si will still insist on it. If you don't persist," Zhou Zhenghao shrugged.

This is probably a typical case of concern and confusion.

Compared to others, Zhou Zhenghao has followed the CPU the most closely in the past three years, and Ning Wei is indeed training him in this direction.

The current first-generation CNMD structure CPU still uses 64NM process technology. The goal Zhou Zhenghao has set for himself is also very simple. In his lifetime, he will advance the CNMD structure production process to the 1nm era. According to the extremely scientific and rigorous mathematics passed in March The model predicts that this goal is completely achievable.

Of course, the prerequisite for this is that this generation of products can sell well. In other words, the market for CNMD structure CPUs can support everyone's confidence in continuing to research and improve the process. If it really fails to meet market expectations, it is difficult to say that this kind of technology regardless of cost is The investment will continue.

This does not mean that the project will be abandoned.

Zhou Zhenghao, who has been working on chip projects for two years, naturally understands that CPUs with completely independent property rights will definitely not be given up, and there will definitely be orders. However, if there is too great a disparity between effort and reward, this technology may also be used as a strategic reserve. technology.

In other words, as long as the first-generation CNMD structure general-purpose CPU is produced, it is confirmed that the performance can indeed be comparable to the world-class general-purpose CPU, such as the Core architecture CPU, but the product will not be profitable, or the profit will not be enough to support the future. With the subsequent product iterations and innovations, it is impossible for Ningsi to continue operating at full capacity as it is now. It is estimated that it will be lucky to retain a team of 1,000 people in the end. The vast majority of people will be forced to move to other positions to do profitable things. project, or have to choose to leave.

In this case, you will definitely not be able to get much R\u0026D funding support, and technology iterations will probably have to be left to chance.

Of course, with the technical reserves, if one day Intel loses its mind and feels that the threat is no longer great, and dares to increase prices arbitrarily or implement some restrictive policies, there will probably be a large amount of capital injection, and the project can be revived immediately.

But for a person who regards this matter as his lifelong career, it is absolutely intolerable to waste time because he has no money. After all, most of these people have no idea about money and do not regard money as their master.

Meng Runze said with a smile: "Zhou, why are you panicking? Our CPU focuses on differentiated competition. It must be matched with the operating system. If you are really worried, call Chen Yuru and Lu Yuanzheng every day and ask them How are Huawei's preparations for the Hongmeng upgrade? As long as the system is not stretched, our chip sales should not be too bad."

Ning Wei smiled as he listened to the conversation between the three students. Naturally, he knew Zhou Zhenghao's worries very well.

It's just that it's hard for him to express his position now.

To be honest, although he was in charge of all the work on the first-generation CNMD structure CPU from instruction set to chip design, and Hongmeng system cooperated with the CPU system upgrade work, he has been following up, but in the end, whether the market can pay for it, he I don’t dare to pat my chest and guarantee that it will work.

This is a mutually reinforcing issue, and even the most critical point is not on the chip side.

Ningwei is still confident about the performance of the chip, but he is not sure about the acceptance of the upgraded desktop system. No way, this generation of CPU failed to get the support of Microsoft in the end, and the matching Hongmeng system made many extremely bold updates, which involves a problem of operating habits.

What Ningwei can guarantee is only some orders in key positions, and there will definitely be no shortage of them. But without popular market support, he can't guarantee whether this generation of CPUs will be able to recover costs or even sell well under Intel's full suppression.

What's more important is that even if the CPU is successfully taped out and enters mass production, the early products will definitely not have any price advantage.

Not to mention that Intel has decisively lowered the price now, even the selling price before the price reduction is unacceptable even when the cost is factored in.

According to the recommended selling price given in March, the most perfect CNMD structure CPU, sixteen-core design, supports up to 128G memory, and the main frequency can reach up to 3.6GHZ. If the first batch of 20 million pieces is mass-produced, each piece will leave the factory. The price is about as high as the retail price of the 12th generation ShenU. If you add various distribution costs, taxes, and warranty costs, the price will probably double, and you can make a profit.

This is calculated based on the mass production yield rate reaching 80%. If the yield rate cannot reach this price, the price will have to be increased to give everyone the motivation to continue doing it.

Under such circumstances, it is indeed difficult for Ning Wei to talk about the market... After all, this cannot be decided by technology alone...