Rewrite the Technological Landscape

After the news that AMD gave Dafeng Semiconductor an order and guaranteed production came out, the news of AMD's formal cooperation with Dafeng Group also came.

Kelsinger came to Huaxia again, attended the cooperation conference with Meng Qian, and announced the in-depth cooperation between the two companies at the conference.

This cooperation will involve many fields, including computer chips, mobile phone chips, artificial intelligence, and service platforms.

In terms of cooperation, Dafeng Semiconductor's guarantee of AMD's recent output can only be regarded as a gift of cooperation with Dafeng Group. What Dafeng Group will really bring to AMD is something more important.

That is: AMD YES!

A well-known stalk in the chip industry.

AMD's rebirth from the ashes was inseparable from Dr. Su's three major principles after taking office, creating great products, deepening partnerships, and simplifying operations.

And these three points are actually achieving one goal, product priority.

In the chip market, operations are very important, but in the final analysis, the core still depends on the product, especially for American companies, basically do not have to worry about being suppressed by their own country, and the most important thing is to do a good job in product quality.

Dr. Su obviously knew this very well, so he put the product first, and Dr. Su, who knew AMD's weakness, bet on TSMC, and it turned out that she was right.

And now, Gelsinger has chosen Gale for AMD.

Dafeng Group, the global pioneer of 3D transistors, is only one month earlier than Intel, but it is also the elder brother one day earlier and the younger brother one minute later.

Dafeng Group is one of the world's two leading lithography machine companies, one of the world's three largest chip IDM companies, one of the world's four largest chip architecture companies, and one of the world's top five chip patent companies.

In this cooperation, Dafeng Group will first help AMD solve the two major problems of compatibility and power consumption.

AMD’s compatibility has never been as good as Intel’s. There is no way around it. Many games and software will be adapted to Intel platforms and Nvidia graphics cards after they are released. AMD can’t change anything, especially in this life. AMD did not acquire ATI, although It is more focused on the CPU, but there is no graphics card or chipset, and the compatibility problem is even more obvious.

But now Gale graphics cards are overwhelming Nvidia, and Gale Group has its own ecology. AMD will come to build a platform to solve the compatibility problem very well.

AMD's power consumption problem is also an old problem, and this happens to be an advantageous technology of Dafeng Semiconductor. If there is no accident, obvious optimization can be seen within a year or two.

However, what Kilsinger is most looking forward to is another thing. Because of the comprehensiveness of Dafeng Group, although there is still a clear distance from Intel in the design and manufacture of computer chips, the future development blueprint of Dafeng Semiconductor is very interesting. .

As a partner, Kelsinger can see the outline of the semiconductor development blueprint of Dafeng Group in the next ten years. Of course, don’t think about the details. Everyone is also an opponent, but for AMD, there is no absolute choice for both, and it is also a gamble.

As for Dafeng Semiconductor’s semiconductor development blueprint for the next ten years, it was actually finalized at the beginning of the year, because there was a problem that everyone struggled with for a long time, and finally Meng Qian made the decision, that is, whether to use DUV or EUV for the 10nm process.

As we all know, Intel has been skipping tickets since 14nm, cheating a lot of customers. Samsung and TSMC have started to develop 5nm technology, and Intel is still competing with 10nm. Even Intel handed over 7nm orders to TSMC for the first time because it was uncertain about 7nm.

Then many people said that Intel has declined, and Intel technology is no longer good. In fact, this matter is not so simple to evaluate, and the things inside are a bit complicated.

In terms of density and performance, Intel’s 10nm++ is definitely comparable to TSMC’s 7nm. It doesn’t mean that 7nm is necessarily better than 10nm. There is no such statement. The key to this matter lies in the semiconductor development of various companies around 2010. Blueprints are formulated differently.

Chip companies will have five-year and ten-year development blueprints for how to develop chips in the future. Instead of thinking about next year this year, this year is already thinking about things five or even ten years later.

In 2011, the chip began to enter the 14nm breakthrough. At this time, a problem appeared. If it is pushed forward, it will be 10nm. If the process is to be advanced, EUV is definitely the best choice based on the current technical theory, but No one knows how many years to wait for EUV.

Moreover, after the EUV lithography machine comes out, it does not mean that it can be used directly. The conversion process in the middle is very complicated. No matter how important the EUV lithography machine is, it is only a part of the entire chip manufacturing. After using the EUV lithography machine, it needs The supporting overall technology usually takes two or three years to mature, which leads to differences in the planning of different companies.

Samsung is the most radical one. It was the first to propose Samsung’s EUV plan. Intel is obviously the most conservative on this issue. It has repeatedly stated that the maturity of EUV should be after 2020, and Intel’s use of EUV must be at least 2020. After that, it will be the 7nm process.

TSMC was stuck in the middle, and in the end it was all-or-nothing and became the king in this field.

Therefore, many people at Intel ask why Intel does not use EUV for 10nm. In fact, the reason is very pure, because it was set in the original plan.

But Intel may indeed be out of luck. One of the most important reasons why Intel is the most conservative is that it is really awesome. Intel can completely squeeze out DUV and research thoroughly.

On the contrary, TSMC and Samsung, which are backward in technology, must rush forward. According to normal market predictions, it is impossible for Samsung and TSMC to fail to see that EUV will not mature in the short term. From 2016 to 2017, TSMC wandered back and forth three times. It can be seen that TSMC almost didn't use EUV to make 10nm.

But the problem is that Samsung and TSMC have just gone to Intel for DUV technology, but it’s just not enough.

Therefore, from the perspective of that era, Intel’s decision is not wrong. It may be that Intel has invested too much new technology in the 14nm and 10nm process. It is miraculous to quickly adapt and digest EUV technology within 2 years.

The reason why TSMC was able to complete EUV adaptation earlier than expected is due to smart mobile devices.

TSMC is a pure foundry. The rapid development of smart mobile devices has brought a huge amount of foundry experience to TSMC. The problems of yield rate and leakage rate are based on the experience of the previous generation process and the huge amount of foundry. The process can be continuously improved, and the advantages are too great.

This natural advantage enables TSMC to grow rapidly when it is growing steadily, and the R\u0026D and cost of advanced manufacturing processes are allocated to the company. That's why it is said that TSMC's success is supported by Qualcomm, Apple and Huawei.

So the facts actually prove that Intel's original prediction was correct, because Intel does not have Apple, Qualcomm, and Huawei to share costs for him and provide a huge amount of foundry experience.

If Intel's original blueprint was to use EUV at 10nm, Intel would probably only be cooler.

And Samsung is also aware of this, so in 2017 it announced that the chip foundry was independent.

Intel is indeed accurate in its predictions, and its technology is indeed awesome. The 10nm process level achieved by Intel with DUV is not bad at all. It can even be said that only Intel in the world can use DUV to achieve this level of 10nm process. This is Intel. absolute strength.

But TSMC relies on smart mobile devices to make good use of EUV technology, which may be Intel's fate.

Your technology is awesome, but you can't stand the change of the times. Smart mobile devices are dismantling the monopoly kingdom established by Intel time and time again.

However, EUV has entered Intel's plan after 2020. It is hard to say whether Intel will change after starting to use EUV.

Therefore, it is very important for Gale Semiconductor to formulate its 10-year semiconductor development blueprint. This plan will have a great impact on subsequent development.

Judging from the current technological development, there is a high probability that mass production will be achieved before 2015, so although it is impossible to be 100% sure, Meng Qian still decided to start using EUV at 10nm.

After making this decision, in the past few years, the company can start experimenting with immature EUV prototypes in the process of continuing to attack 14nm. Therefore, if there is a small probability accident, the impact will be great. In the future, in the computer It's a bit of a hang-up if you want to compete with Intel on the chip.

However, it is also difficult for Dafeng Semiconductor to compete with Intel on DUV. This cannot be solved by spending money. We can only hope that fate will not deliberately make things difficult for Meng Qian this time.

At the same time, another important thing is to develop the OEM of Dafeng Semiconductor, which is why Meng Qian is so active in finding OEM orders this year.

According to the ten-year development blueprint of Gale Semiconductor, the foundry business department of the company will be independent by the end of 2012 at the latest, and the subsequent plan is to mass produce 14nm before 2014, to produce EUV lithography machines in 2015, and to adopt EUV light before 2017. Engraving machines mass-produce 10nm, challenge 7nm in 2018, and become a leading enterprise in 5nm process by 2020.