Download Chapter 654 of the champion finally caught up

Next chapter

Ordinary semiconductor circuits use metal wires to connect electronic components according to circuit diagrams. By adjusting the circuit diagram and replacing new components, the circuit can be upgraded, and if any component is broken, it can also be detected and then removed and replaced.

For integrated circuits, except for relatively niche programmable circuits, most mainstream products are not suitable for this method, which is based on a single silicon wafer through repeated doping and multiple photo/chemical etching methods "integral molding".

To produce chips, the wafer factory needs to remelt the high-purity (nine nines, only one billionth of impurity) polysilicon, prepare it into monocrystalline silicon ingots with a diameter of tens of centimeters through guò recrystallization technology, and then finely grind it into cylinders and cut it into thin slices with a thickness of one millimeter, which is the so-called silicon wafer. The processing and manufacturing of the core of the chip is carried out on this thin sheet, which is smoother and smoother than a mirror, and the name of the wafer factory is also derived from this.

The silicon wafer is round, and the chip core is rectangular or square, so the core number of CPUs and GPUs is an even number of 2468, and the three-core processor is actually a weaker core with a quad-core shield. The same chip area, the larger the wafer, the less offcuts are wasted, which can lead to better benefits.

However, the larger the wafer, the higher the requirements for silicon ingot preparation, and even the strongest toothpaste factory in the universe can only achieve a diameter of 300 mm (the theoretical upper limit is 450 mm).

On the other hand, when the chip area remains the same, the process technology can be improved, which can reduce the volume and distance of the transistor, so that more quantities can be stacked to bring higher performance, or the area with the same performance can be reduced, bringing lower material costs. At the same time, the reduced distance can also reduce the resistance of the connection circuit, reducing the heat generation and power consumption of the chip.

Therefore, whenever the process is upgraded, chip manufacturers will always talk about the three highlights of area reduction, performance improvement, and power consumption reduction.

It's just the same way, every time the process is upgraded, there will be a long period of running-in, raw material impurities, process fluctuations, and other imaginable and unexpected reasons may lead to defects in the product.

There are hundreds of dies (chip cores) on a wafer, and there are billions of transistors on each die, and the defects are completely unavoidable, and can only be slightly compensated for by guò repair measures, which is why the "yield rate" is often mentioned in the news such as chips.

Chips are tested before and after packaging, and those that are completely unqualified can only be discarded, and then the cost is added to other qualified products. If there are obvious defects, they can be rescued by shielding the fault area and reducing the technical indicators, and finally the ratio of all qualified products to raw materials is the yield rate.

This is also why the same batch of CPUs, some of them have a "good physique" and can be overclocked for a long time, while some can only be overclocked a little, and some simply can't even hold up the default frequency, so they can only be labeled as "Celeron" or "triple core" and sold out.

Mobile phone processors, memory flash memory and other chip products also have similar "physique" problems, memory chips have different main frequencies, flash memory chips have different transmission speeds, and the maximum frequency of the same batch of CPUs is also different. It's just that because the mobile phone battery is not strong, no one will overclock the phone's CPU and memory, and sometimes it will be downclocked.

For the wafer factory, with the maturity of the production line, the yield rate is getting higher and higher, and the product is naturally getting better and better.

Therefore, there is often such a phenomenon in the chip industry: as long as there are enough orders, the old production line will not be closed.

They seem to be backward, but in fact, they are low cost because of mature technology and high yield, and are deeply loved by manufacturers who do not have high requirements for product performance, but value cost and stability.

For example, Intel has both the world's leading 14nm factories and many factories with backward manufacturing processes, and its 200mm and 130nm factories in Massachusetts were closed until this year. These "backward" factories, of course, cannot be used for the production of processors, the main products of which are the north-south bridge chips of the motherboard.

And for Intel, keeping them can also help save money, although the "toothpaste factory" relies on the world's unique x86CPU to make a lot of money, but its investment in technology development is also very huge, the old factory only needs to be open to make money, naturally will not be eliminated for the first time.

This is also the reason for the long-term losses of domestic wafer enterprises, because of the technical blockade, they can only introduce backward technology, but they have insufficient production experience when they first arrive, and the products produced with backward technology are still not as good as foreign counterparts, and they have more experience and less depreciation pressure on equipment. When it was finally time to get through, the backward technology has become a sunset technology, to the point of complete elimination.

Fortunately, the mobile phone chips are simple and repetitive Japanese SC streamlined architecture, and the production process requirements are much lower than x86, and domestic wafer factories such as SMIC have finally caught up with the international mainstream level at the 28nm stage, and the advanced level at this time is 14/16nm.

Bee has always maintained a high degree of attention to SMIC's 28nm production capacity, and cooperated with it in the trial production stage to produce bee heart series chips in small batches, but it is the 41x series desktop processors that have been officially signed and commissioned to be foundry, as well as a small number of 42x low-end tablet processors, the main reason is that the latter still has some missing classes to make up for in terms of high-performance processors.

Although under the kind care of the superiors, SMIC also obtained the production materials of the top processor of 436, but their pole process technology is not as good as TSMC after all, and was finally abandoned by bees, and finally used a stepping stone for BP6 performance counterattack.

Of course, such caution will not affect the cooperative relationship between the two parties, after all, Bee Heart needs SMIC's tariff-free and low cost, and SMIC also needs Bee's orders to ensure cash flow. After all, it exclusively occupies the three major customers of Bee mobile phones, tablets and PT with sales exceeding 10 million, and the chip shipment of Bee Heart is light and has achieved the first place in China.

It is precisely by virtue of his copycat micro-innovation technology, which is more powerful than Chen Jin, that he can overtake in the corners and take only a few years to complete the development path of others for more than ten years or even decades.

Of course, there is also the credit of the fabless fabless model, which is too suitable for Atletico Madrid.

If every chip manufacturer needs its own FAB (wafer factory), the bee heart can't play at all, after all, the current wafer factory is a tens of billions of heavy asset projects. Not only chips, but also the scale of investment in the screen field is also rising, and the butterfly screen that bees are proud of is that they provide technology and funds to entrust other companies to produce on their behalf.

This is also the advanced experience that bees have learned from Apple, and Apple's success is the success of the foundry model. Because Apple's orders are stable and the bids are high, the foundries are also happy to do it, although the profit margin is simply coolie compared to the former, but it is still much better than some factories that are eager to sell blood.

However, this also brings a risk that if their technological upgrades do not keep up with Apple's pace, they will be ruthlessly eliminated by the latter.

However, most of the time, as long as the leader does not make mistakes, he rarely falls behind, after all, they have the excess profit brought by the lead, as long as they are willing to spend money to ensure the advantage, it is no problem.

Three years ago, Mark Bohr, a senior executive in Intel's process technology department, once said in a high-profile manner that "the fabless model is at the end of the road", when the toothpaste factory released the third generation of Core i-series processors, which are manufactured using 22nm FinFET (FinFET (FinFET) technology, and the performance power meter Xiàn is quite good.

At that time, those wafer foundries, such as TSMC and GLOBALFOUNDRIES, were still in the 28nm stage, and Samsung was still in the 32nm stage, which was backward and would be abandoned by the world.

As a result, three years later, Intel is still the world's leader in the field of x86 PCs and servers, but the fear that PC growth has peaked and is gradually declining has become a reality, and their attempt to open a second battlefield into mobile computing has completely failed.

Because of the complex structure and high cost of their ATOM processor due to the x86 instruction set, they could only rely on subsidies to sell out, and finally lost $7 billion and still did not see any hope of recovering their costs, so they had to suspend the subsidy market plan.

On the contrary, Samsung and TSMC, which are despised by them, have successively overcome the difficulties of FinFET technology, Samsung skipped 22/20nm and directly launched the 14nm process, TSMC was poached by Samsung because of the core talents, and now it has its own 16nmFinFET process

FinFETs are considered to be the key technology for chip manufacturing to enter below 20 nanometers, because as the process reaches the nanometer level, the insulation layer separating each transistor is only a dozen atoms thick (silicon atoms are 0.22 nanometers in diameter), which will become weaker and weaker in the face of the increasingly strong quantum tunneling effect, and the former can improve this phenomenon.

According to quantum theory, the motion of microscopic particles such as electrons is random, and when the barrier is weak enough, electrons may randomly move to the other side, and the macroscopic table is that the current breaks through the insulator, so it is also called the quantum tunneling effect.

In the past few years, chip manufacturers and wafer factories have begun to emphasize various electrode processes and even 3D transistor technology.

FinFET was invented by Chinese scientist Hu Zhengming in 2000, and TSMC originally poached Hu's student Liang Mengsong as the director of the R&D department, but because he did not use his technology on the 20nm node, the latter was poached by the more ambitious Samsung. Although TSMC later won the non-compete lawsuit, it lost Lizi.

In addition, the reason why TSMC is jokingly called Taiwan leakage is because the process was not mature enough when it first entered the 28nm process, and sometimes unexpected leakage between transistors led to additional power loss, making the product more power-intensive than expected. Fortunately, after several small process upgrades, its 28nm process has entered the mature stage, and the power consumption has improved significantly.

It is with the blessing of their HPC process that the bee BP6 staged a good show of breaking through the glasses, which not only sent the bee into the top ten of the global smartphone sales list, but also made the bee system sit firmly in the position of the third largest mobile phone system.

Of course, Qualcomm's good friends also contributed to the hard work, and each BP6 gave people 60 patent fees, which is simply worth it!

However, Gao Baobao took the money, but his heart was bitter, he also had a patent share for selling 810 himself, and there was a price commensurate with his high-end flagship status, and he had to earn more.

Speaking of which,In the 810 heating problem,Qualcomm itself has a lot of responsibility,In order to launch an eight-core 64-bit flagship in front of MediaTek,They directly used Arm's public version of the high-performance A57 core,The performance is really a bunker,Power consumption is also heheda。

After all, as an IP licensor, Arm needs to consider the needs of partners in different fields when designing CPU cores, and the public version of the guide solution often seems too modest, after all, they are only a "small" company with more than 1,000 people, and there is no way to achieve everything.

Qualcomm is now working hard to promote the next generation of products, 820, using its own kryo core, and abandoning the many useless eight-core architecture, changed to 22 quad-cores, and the foundry also abandoned TSMC and switched to Samsung's 14nmlpp (low power enhancement) process.

TSMC's 20nm process itself is customized for the big customer Apple factory, Radeon (AMD's independent display department) and female IDIA would rather continue to use 28nm than use it, Qualcomm did, but it was a loss, so it was angry and bricked Samsung Semiconductor's arms.

Interestingly, TSMC has once again recently demonstrated Apple's talent for optimization, and the evaluation after the release of the iPhone 6S mobile phone shows that the TSMC OEM A9 is better than the Samsung 14nm process version. Qualcomm's 820 has not yet been officially launched, and some people have already ridiculed them for choosing the wrong partner again, and all kinds of "I have made an appointment, and I have to finish it with tears" jokes emerge in endlessly.

However, Bee Heart did not have this problem, and directly contacted TSMC to inquire about the remaining production capacity.

Relying on Atletico Madrid's brainwave 3D printing technology, Bee has already prepared the design scheme of the next generation of processors, and when the three 14/16 process SoCs are introduced to the market, Atletico Madrid has borrowed a lot of ideas and experience from them, and the next generation of 536 processors has been "handmade" by him.

Although after having the finished product, Bee Heart still needs to cooperate with TSMC to prepare dozens of GB or even more process numbers, but after having a clear goal, it is still a lot easier in general, and 536 has great hope of becoming the second 16nm process mobile phone chip after A9.