Deep analysis of the industry: processor nuclear fission

RM and IBM have opened up the processor cores in the terminal and server markets one after another. Just as neutrons break up the atomic nucleus and undergo a nuclear fission reaction and release tremendous energy, the opening of the processor core will strongly impact the existing integrated circuit industry.

The system vendors aggregate multiple IP cores on one chip, and the nuclear fusion reaction also generates higher levels of energy, which will have a profound impact on the entire IT industry.

“The first faint morning light appeared in the East. At this moment, it seemed that under the earth's crust a light that was not in this world was rising. It was a sunrise that the world had never seen before. At this moment, immortal deeds appeared. As time has stagnated, space becomes a dot, and it seems that it is falling apart. People feel that they have the privilege of witnessing the birth of the world."

The only New York Times reporter for science and technology, Lawrence, who was authorized to interview the US atomic bomb development project, the Manhattan Project, wrote on July 16, 1945, after witnessing the test of the world’s first atomic bomb in the desert of New Mexico. Lawrence won the Pulitzer Prize for his report on the development of the atomic bomb and the nuclear attack on Japan.

The power of the atomic bomb comes from a nuclear fission reaction in which the atomic nucleus of a heavy metal element is destroyed by neutrons and then split into two or three light atomic nuclei and releases energy.

Although "software definition" has become a hot word in the industry, and the proportion of software in the IT industry is also increasing, but the software is running on the processor, so it can be said that the entire IT industry is built on the processor.

For a long time, the processors that people see on the consumer market as well as on the enterprise-level market have been modeled and manufacturer's chips. The processor, as the smallest computing module, cannot be subdivided. The processor is produced by processor manufacturers such as Intel, AMD, and IBM.

Until Apple's smart phone iPhone got great success, people discovered that the processor can continue to be subdivided, and Apple can also produce its own processor. As a result, the behind-the-scenes hero of the smartphone, ARM, came to the stage and it was the business model of ARM's open processor core that made the smart mobile terminal market such as smart phones and tablet computers flourish.

On August 7, IBM, Google, NVIDIA, Tyan Computers and Mellanox announced the establishment of the OpenPOWER Alliance. The alliance will provide advanced server, network, storage and graphics processor (GPU) acceleration technologies for the next generation of ultra-large-scale cloud computing data. The developers of the center provide more choices, stronger controls and better flexibility.

To put it plainly, the goal of these high-end server chips, search engines, GPU acceleration, server motherboards and network vendors Taoyuan is to "cradle" ARM's business model in the server market, intended to reproduce ARM's consumer electronics market in the server market. Success.

As we all know, contrary to nuclear fission, nuclear fusion reaction is the polymerization of multiple lighter nuclei into heavier nuclei and release of energy. Hydrogen bombs based on nuclear fusion reaction are more powerful than nuclear bombs based on nuclear fission reaction. However, it is little known that nuclear fusion needs to be performed at extremely high temperatures and pressures. Therefore, it is necessary to use an atomic bomb as a trigger to detonate hydrogen bombs. In other words, there is nuclear fission before nuclear fusion.

Similarly, the energy generated by the opening of the processor core will profoundly affect the industrial structure of the integrated circuit (IC). The energy released by system vendors such as Apple Inc. that “opens up” multiple third-party open intellectual property (IP) cores on one chip will affect the entire IT industry.

Part I: Industry Evolution and Platform Leadership

Through the process of the evolution of the industrial form, one can feel how the SoC flows from the undercurrent to the industrial trend. Looking back at the struggle for the platform leadership by the manufacturers, the impact of the processor core on the market structure of the chip will be more profoundly understood.

IC industry segmentation evolution

As with the development of other industries, the development of the integrated circuit industry has also undergone a process of deepening professional division of labor.

Chip production is usually divided into three links: design, manufacturing, and packaging and testing. The early IC industry was dominated by machine manufacturers. Motorola, a communications company, owned a large-scale semiconductor business. In the 1970s, integrated device manufacturers such as Intel (IDM) began to dominate chip manufacturing and packaging testing; by the 1980s, foundries such as TSMC had emerged in the manufacturing sector; in the 1990s, IC design was fine. Separated IP vendors represented by ARM.

The independence of OEM and IP design in the chip industry has made it possible for third-party processors such as Qualcomm and system vendors such as Apple to use SoC (system on chip) to customize their own processors. According to the semiconductor market research company IC Insight's statistics on the global semiconductor market from 1999 to 2012, the compound growth rate of Fabless (non-product line) manufacturers' sales was 16%, while the compound growth rate of IDM manufacturers' sales was only 3% over the same period. In 2012, Fabless sales accounted for 27.7% of the semiconductor market, which strongly impacted the IDM model.

One of the unique driving factors in the IC industry segmentation is the increasingly high investment and increasingly fierce competition in the semiconductor market.

Chip manufacturing is a knowledge-intensive and capital-intensive industry. As the width of IC processing lines continues to shrink, the investment in building a 45nm semiconductor production line has reached as much as $3 billion. The high market threshold allows small and medium-sized semiconductor manufacturers to sigh.

The continuous narrowing of the line width also means that the process latitude of different production lines is reduced. In 2002, Samsung’s StrongARM processor had been clocked at 1.2GHz, while ARM’s processor IP provided only 400MHz. This is because Samsung has optimized each other in the design and manufacturing process, which in turn significantly improves the chip clock speed. ARM's strength at that time was still weak and it was difficult to optimize the production lines of different manufacturers. Therefore, the main frequency was not going to go.

Intel's core competitiveness lies in having the world's strongest design team and the best production facilities at the same time, making it easier for both to optimize each other and then produce the best performing chips. Intel's pursuit of process consistency is nearly exacting. The "precise copy" program led by former CEO Barrett is designed to allow Intel to have a consistent, high-level manufacturing environment at its factories in many parts of the world.

Intel competes with chipmaker AMD with high investment and uses design and manufacturing to optimize each other to compete with the foundry's foundry TSMC.

In the competition with AMD, Intel won the bet. In 2008, AMD was finally unable to bear the burden of investment, and it divested its chip manufacturing business and became a Fabless company. In the competition with TSMC, TSMC not only did not lose, but also cooperated with Chipless (chipless) companies such as ARM to achieve mutual optimization of chip design and manufacturing, and similar to Intel's time in introducing new processes. This ensures the high performance of the foundry chip.

At the same time, ARM is also growing strong enough to optimize for a specific production line of a specific foundry company. Today, if customers want to build on TSMC, ARM can provide hard IP cores that package different production processes based on TSMC's performance optimization or power optimization to meet user's preference for performance or power consumption.

Competing for platform leadership

"Standard is king" is reflected in the control of the platform to a large extent on the whole product. Controlling the platform is tantamount to dominating the market, which means generous returns. Therefore, the battle for control of the platform is particularly important.

The battle for control of the platform can be described as brilliant in the personal computing market.

In 1981, when the PC came out, IBM defeated Apple with its open PC architecture and established its position as the big brother in the personal computing market. However, the open architecture of the PC makes it difficult for system vendors to control. So when IBM introduced the 286 processor-based PC in 1985, it introduced a proprietary micro-channel bus on the motherboard, trying to control the PC through an exclusive bus architecture.

As a result, Intel and Compaq joined forces to introduce an open PCI bus on a 386 processor-based PC introduced in 1986, thus ensuring the openness of the PC. Since the PCI bus is supported by the chipset, Intel also realized for the first time the benefits of platform leadership.

In 1991, IBM provided POWERPC processor technology for RISC (Reduced Instruction Set) architecture, Motorola's responsibility for manufacturing, and Apple's POWERPC alliance, which is responsible for the production of complete machines and providing compatible machine licenses. The goal is to target the personal computing market. All members of the alliance were outstanding. At that time, the RISC architecture had certain technological advantages over the x86 CISC (Complex Instruction Set) architecture. Motorola Semiconductor Division received the first National Quality Award from the US President Reagan in 1988, and Apple Computer Macintosh. And its operating system also makes PC and Windows look good. However, such a dream team, because Motorola in the manufacture of POWERPC clock frequency delays, coupled with Apple's withdrawal of the compatibility of the authorized aircraft, eventually collapsed.

After escaping this robbery, Intel and Microsoft joined forces to "edit" the PC system design specifications in 1997, 1998, 1999, and 2001, while other PC makers only had to participate. At this point, Intel and Microsoft have completely established their leading position in the PC market. The two companies received a market return of 90% of the profits in the PC market.

In 2003, Intel released the Centrino chipset via the WiFi wireless communication platform, which greatly weakened Microsoft's leadership position.

In contrast, ARM, although the ARM architecture had more than 90% of the market share in the smart phone, but the smart phone market right to speak in the operating system manufacturers Apple and Valley singer, even Qualcomm, MediaTek speak more weight than ARM.

The middle ground is very simple. The control of the PC platform is almost completely fixed in the processor and the chipset. Therefore, it is natural for Intel to have the final say. However, ARM, which only provides IP core licenses, can hardly control the platform control power in the chip.

In addition, the premise of accepting this business model of ARM is to be able to tolerate very low profit returns. In 2012, the global shipments of processors based on the ARM architecture reached a record high of 8.7 billion, but ARM's annual revenue was only 910 million US dollars and profit before tax was only 410 million US dollars. In contrast, Intel’s annual revenue for 2012 was 53.34 billion U.S. dollars and profit before tax was 14.87 billion U.S. dollars. ARM's annual revenue and pre-tax profits are only 1/58 and 1/36 of Intel respectively.

Middle article: IBM scrambling and Intel tangled

In today's IBM transition to the road to software, it is a big mistake to think that IBM is opening its POWER processor core to make big money. The OpenPOWER Alliance will undoubtedly become a magic weapon for IBM to attack and retreat. Intel had to accept the move and was forced into a dilemma.

IBM open kernel for what

Starting from punching machines to dominate the mainframe, from hardware companies to services, and from service to software in the past two years, it is these continuous transformations that have made IBM's dream of a century-old store. And "towards high-margin realm" has become the compass of IBM's transformation.

From May 2005, the PC business was sold to Lenovo, and in June of that year, Apple announced that it abandoned IBM's POWERPC processor and switched to Intel processors. IBM has already withdrawn from the personal computing market. In April of this year, it was reported that IBM is negotiating with Lenovo to sell x86 server business. This is an open secret in the industry. This move shows that IBM has decided to retreat from the x86 market. As Microsoft announced in June of this year that the next generation of Xbox will use AMD processors, the POWERPC architecture that once unified the Microsoft Xbox, Sony PS3, and Nintendo Wii consoles will only be left to Nintendo.

Various signs in the market confirm that IBM will continue to peel off the low-margin hardware business in the transition. The lower profit margin after the opening of the processor core seems to be contrary to IBM's transition to the high-margin field.

In fact, the POWER processor is equivalent to the foundation of IBM's "Empire State Building." Despite being buried underground, the importance is beyond doubt. The most effective way to consolidate this foundation is to expand the user base of the POWER architecture. After POWER is forced to give up more than half of the console market, expanding the user base becomes even more urgent.

IBM has introduced "special value" hosts and servers such as Linux dedicated hosts and Linux servers. Although it is full of price tempts, it is unrealistic to allow a large number of enterprise users on other platforms to migrate to the POWER platform, and extends the POWER architecture to other areas. The field can play the role of double-edged swords, which not only expands itself, but also beats competitors.

Look at Google, another important member of the OpenPOWER Alliance. In 2006, this reporter interviewed the JavaOne conference in San Francisco and ate with the head of the Sun company responsible for media reception. At that time, Google was the hottest topic, so he also talked about one thing related to Google. Google Chairman and Chief Executive Officer Schmidt once discussed with Sun Chairman McNealy about Google's adoption of the Sun UltraSPARC processor architecture. Because Schmidt had previously served as Sun's CTO, McNealy used the old tone to talk to Schmidt. , Everyone does not know, at this time Sun is already falling flat, and Google is just like the sun. The result was that Schmidt was very upset and this big deal was soaked.

Although Google's x86-based servers and storage are designed to provide high-density, low-power, and economical advantages, the cost and power consumption benefits of building SoC chips based on third-party licensed processor cores are high. If you go through Google's millions of servers and storage, the economic benefits will be very significant. Therefore, when IBM has the idea of ​​opening up POWER, Google will naturally respond positively.

IBM's opening of the POWER processor core allows the POWER architecture to take over the cloud computing high ground, and it also let Intel's Intel move toward the SoC in full swing at 14nm last year.

Intel is very tangled

"Dilution" is an important key word for the chip industry. In 2011, for example, Intel’s R&D investment was US$8.35 billion and its production facilities invested US$10.76 billion. This, together with other expenses such as operations and labor, constituted a huge amount of cost. Most of these costs were allocated to chips. Therefore, The larger the chip sales, the more the cost can be spread even more.

General-purpose processors are widely used as an ideal carrier for the dilution, and x86 processors are far ahead in the sales of general-purpose processors, and therefore become the best carrier for dilution. Every year, Intel dares to spend money on R&D and production facilities. If it is not from the external stimulus, I believe Intel has no reason to give up this most profitable business model.

This stimulus first came from ARM's IP-licensed business model in the consumer electronics market. Although Intel put forward the idea of ​​expanding Moore's Law as early as the Intel Developer Conference in the spring of 2002, and wanted to use this to recapture the glory of the PC industry in the communications field, Intel has only seen the harvest in the consumer electronics market for years.

The big reason for this is that it is difficult for general-purpose chips to meet consumer electronics requirements for small size, full functionality, low power consumption, low cost, high reliability, and high speed.

As a result, Intel began to carefully explore the consumer electronics market. In 2009, TSMC reached an agreement with TSMC to undertake the OEM of some Atom processors to add third-party IP cores to Atom processors. This is Intel's first time in more than 40 years to put the product to OEMs from other vendors, and is Intel's attempt to open Atom processors.

The reason why Intel chose Atom processor to try is because the consumer electronics market has broad prospects. Following smart phones and tablet computers, digital TVs and wearable computers will greatly expand consumer electronics market space. The consumer electronics market has thus become Intel's battleground. Moreover, with the huge profits brought by Core Duo and Xeon processors, Intel can also allow Atom to let go in the consumer electronics market and ARM.

However, ARM quickly used TSMC to get TSMC, making Atom's attempt to open the kernel had to "taste it."

When SoC took the lead in the field of consumer electronics and achieved sweeping momentum in the industry, Intel announced in 2012 that it would start a full transition to SoC from 14nm and will no longer produce general-purpose processors.

Compared with the ARM open-core business model, Intel's move is a compromise approach, both to optimize the processor for a specific application area, but also ensure the integrity of the processor, and the integrity of the processor is to ensure that the Intel platform leader The premise of status.

Although there is no open processor core, Intel's move toward SoC is still leading the trend in the desktop and server markets.

However, the emergence of the OpenPOWER alliance made the industry's focus on Intel's attention shift from whether to SoC, quickly transferred to open or not on the processor core.

After the advantages of processor core development have been validated in the consumer electronics field, POWER's primary task is to expand the user base rather than count on POWER to make big money. Therefore, IBM can tolerate the problem of low profitability accompanied by the openness of processors. Grab the high-end market of cloud computing while attacking the x86 architecture to attack the high-end server market. It should be IBM's higher expectations. The most unlucky one is to use Wei Wei to save Zhao's way to ease the x86 server by "mixing" the x86 server chip's pricing system. Market pressure on RISC servers.

This move makes Intel face difficult choices. Without opening the processor core, in the event that IBM proves that ARM's business model is feasible in the server market, Intel will suffer a disruptive blow. The opening of the processor core means that its “living standard” has fallen from “big fish” to “boiled rice”, and the most critical thing is to lose the right to speak on the platform. This is probably Intel’s most difficult to accept.

Next: Meeting the Industrial Revolution

The x86 server chip market is calm before the storm. With ARM's low-power intrusion from the low-end and IBM's support of the OpenPOWER Alliance from the high-end market, the x86 server chip market is bound to become a storm. Perhaps, as Intel, the "leader" of the x86 chip market, it's time to battle.

As far as the IT industry is concerned, the reason why PCs subverted minicomputers is because PCs can spread or even extend the use of computing technology. The new generation of leading technology, "mobile Internet + cloud computing", will spread cloud computing's powerful computing capabilities through intelligent terminals to areas covered by wireless networks, and thus has a PC's incomparable diffusion capabilities. It is not surprising that tablets and smartphones are replacing PCs to dominate the personal computing market.

The business model of ARM's open core coincides with the industry development trend of "mobile Internet + cloud computing", which has contributed to the explosive growth of smartphones and tablets. To a large extent, the opening of the processor core only acts as a booster for industrial change.

Therefore, before discussing the impact of processor core opening on the enterprise-class market, we should focus on the impact of cloud computing. In Gartner's excellent situation in which the server revenue increased by 19.5% in the second quarter of 2011, the author wrote a thousand-word review on the crisis of the brand server market. It mainly analyzes the cluster architecture as the mainstream hardware architecture of cloud computing, and raises the availability of computing systems to the same level as that of brand servers, thus making the brand of servers become dispensable in cloud computing centers. The several times increase in server utilization by virtualization means a significant reduction in server market demand. Therefore, when the cloud computing is popular, it is the time of the brand server crisis.

In a sense, Google’s success is based on the economics of Google’s own design of servers and storage based on x86 chips. The establishment of the OpenPOWER Alliance has allowed Google to increase its performance while reducing costs. It can be said that the opening of the processor core will significantly reduce the operating costs of the cloud computing center, thereby effectively reducing the application threshold of the cloud computing platform, the most important innovation platform today.

The server chip market will be particularly exciting. First, after ARM had entered the server market for several years, its 64-bit architecture Cotex-A57 will finally come out next year; AMD completed the cross-border business to the server market by acquiring micro-server vendor SeaMicro in 2012. Brigade; now, IBM has also challenged Intel with OpenPOWER.

The server chip market is destined to be difficult to calm, and only Oracle will be able to wait and see.