Focus analysis｜It is rumored that "AMD" will acquire the world's boss "Xilinx" of FPGA, or it may take a fancy to the 600 billion yuan market

 On October 9, according to foreign media citing people familiar with the matter, the processor manufacturer Advanced Micro Devices Inc. (Advanced Micro Devices, hereinafter referred to as AMD) is acquiring Xilinx Inc, a leading manufacturer of field programmable gate array chip (FPGA), a rival. . (Hereinafter referred to as Xilinx) for advanced negotiations.

People familiar with the matter said that the value of the deal may exceed US$30 billion, and the two companies are discussing an agreement. The results may come out as early as next week, but there may be no guarantee of reaching an agreement.

In this regard, AMD did not respond. Xilinx told the media: "It does not comment on any market rumors."

Affected by this news, as of the close of US stocks on October 9, Xilinx shares rose 14.11% to US$120.94.

If the merger is reached, it will be the second largest semiconductor M&A transaction this year. In September, NVIDIA announced that it had agreed to acquire ARM, a British chip design company, for US$40 billion. This is also AMD's largest M&A transaction since AMD acquired ATi for US$5.4 billion on July 24, 2006.

The acquisition is also regarded by the outside world as AMD's imitating Intel. In 2015, Intel acquired FPGA manufacturer Altera at a price of US$16.7 billion to help Intel further grow in new markets. But the market does not seem to be optimistic about AMD's move. On October 9th, US stocks closed, AMD stock price fell 3.41 US dollars, a decrease of 3.94%, to close at 83.1 US dollars.

Some analysts say that Xilinx may say no to AMD, and Intel’s acquisition of Altera has not formed the expected synergy effect, so it is not optimistic about this similar move.

However, if this transaction is reached, AMD will combine the product line layout of CPU+GPU+FPGA, which is also an important step for AMD to move towards next-generation heterogeneous computing. It also means that AMD will have the opportunity to further penetrate into AI chips, The Internet of Things, aviation, automobiles, 5G communications, artificial intelligence and other fields have opened the third advantage track, and the capital to compete with Intel has increased.

"Returning" to the high-end arena, is AMD's pockets bulging?

AMD was founded in Silicon Valley in 1969, only one year after Intel was founded. AMD founder Sanders and Intel founders Noyce and Moore both worked at Fairchild Semiconductor. The two companies can be described as having the same origin and different fate. After more than 50 years of development, ADM has become the world's second largest central processing unit (CPU) manufacturer after Intel. It is also one of the few independent graphics processing unit (GPU) suppliers in the world besides NVIDIA.

In the past 40 years or so, Intel, which is a technological leader, has been firmly seated as the world's leading processor despite occasional turmoil. AMD has gradually moved from being a “second supplier” to an independent self-research, and after ups and downs, it finally has the ability to “break the wrist” with the boss:

• In 1999, AMD was the first to cross the 1GHz CPU frequency. In 2003, it launched a 64-bit Athlon chip compatible with X86 pre-products, breaking the image of followers and imitators and ushering in a bright moment.

• In 2005, Intel launched the "Pendulum Plan" to start the "Golden Decade", and AMD fell into the "Lost Decade." By 2015, AMD under Su Zifeng's helm will be repositioned with a long-term focus on the high-performance computing market. Key markets include data centers, personal computers, and game consoles.

• In 2017, AMD launched the first-generation Zen micro-architecture processor to open up the data center business and battle Intel in the six-core and eight-core era of traditional PCs.

• On October 9, 2020, AMD officially announced the new Zen 3 CPU architecture with TSMC’s 7nm process and the latest generation of Ryzen 5000 series desktop processors. At the same time, it revealed some details of the Zen 4 architecture CPU based on the 5nm process, Zen 4. The framework is planned to be launched before 2022.

In the capital market, AMD's stock price once fell below US$2 per share in 2015, and its market value fell to only US$1.9 billion. As of the close of trading on October 8, 2020, AMD's stock price has risen to US$86.51 per share, with a total market value of approximately US$101.568 billion. This year, the company's stock price has soared by 90%. According to the second fiscal quarter, AMD's revenue increased by 26% to 1.93 billion US dollars, and its net profit was 157 million US dollars, an increase of 349%, thanks to record sales of notebook computers and server processors.

But in terms of cash flow, the latest financial report shows AMD's cash flow is only about 1.8 billion US dollars. Bloomberg also analyzed that if it wants to successfully acquire Xilinx, AMD is likely to adopt Nvidia's acquisition of ARM, that is, cash plus stocks.

Why Xilinx?

To answer this question, we must first clarify what Xilinx does.

According to its official website, Xilinx is the inventor of FPGA, programmable SoC and ACAP. Its highly flexible programmable chip is supported by a series of advanced software and tools, and can be used in consumer electronics, automotive electronics and the cloud.

Compared with CPU and GPU, another important member of the chip family FPGA is little known. However, with the rapid development of 5G communications, big data, and the Internet of Things, the importance of FPGAs is getting more and more attention from the market.

The biggest feature of FPGA is that it is programmable and can be customized based on hardware. Generally speaking, this chip allows users to program on it, and users can flexibly change the calculation tasks of the chip. There is no need to re-customize the chip when updating tasks. Compared with the development process of ASIC (application-specific integrated circuit), FPGA development can be quickly tested. Wrong iteration shortens development time. Its flexibility is comparable to general-purpose processors, and is superior to general-purpose processors in terms of computing speed and power consumption, but inferior to ASICs.

These features make FPGA play a key role in video image processing, communications, digital signal processing, and embedded fields. In addition, FPGA programming has a very high threshold, and Xilinx, Intel and other giants have more than 6000 patents surrounding FPGA, and a very high construction threshold has been established.

Looking at the global FPGA market, Frost & Sullivan's data shows that Xilinx, Intel, Lattice and Microsemi have almost monopolized the FPGA market. Among them, Xilinx occupies the first place, with a market share of 50%.

As of the close of US stocks on October 8, Xilinx reported $105.99 per share, with a total market value of $25.89 billion, which is about 1/4 of AMD's market value.

Last year, Xilinx’s key customer Huawei and a number of important customers were restricted by the US government. Xilinx lost many important customers’ business in China and suffered setbacks. Earlier, there were reports that Xilinx’s FPGA chip business was hit after the US government banned Huawei from purchasing US components and software, and layoffs were necessary. This move by the US government cut off Xilinx’s quarterly revenue of $50 million. This also gave AMD an opportunity to acquire Xilinx.

At the same time, Xilinx and AMD have also had a good cooperation before. Xilinx has provided solutions for the EPYC (Xiaolong) series data center processors released by AMD, providing a series of solutions from NVMe HA, NVMe TC and Embedded RDMA. The IP of the storage system enables it to realize the efficient storage acceleration function of FPGA.

Or sword refers to heterogeneous computing

There is no doubt that AMD's acquisition of Xilinx is the FPGA in its hands. "Heterogeneous computing will become the mainstream of future processors", this view has gradually become the consensus in the industry.

Data, algorithms, and computing power are the three key elements of information technology. With the rapid development of AI, big data, cloud computing, 5G and other technologies, data is growing at an unprecedented rate. IDC has issued a report saying that by 2025, the global data circle is expected to grow to 175 ZB . At the same time, the algorithms used to process and utilize data have become more and more complex. Explosive growth of data, continuous complexity and optimization of algorithms all put forward new requirements on the underlying computing power. General-purpose computing represented by CPUs is becoming more and more difficult, and the demand for heterogeneous computing is increasing.

For example, if the relationship between the three is compared to a working kitchen, then the data is the ingredients, the algorithm is the recipe and the chef, the computing power is the kitchenware, and the various applications are the final finished food. As customers want more and more food types, tastes, and quantities (ie more and more applications), there are more ingredients in the kitchen, better chefs, richer recipes, and better kitchenware. , That is, I hope that kitchenware has more detailed division of labor, greater firepower (computing power), more energy saving (lower power consumption), and affordable prices.

And heterogeneous computing is this "better kitchenware". The so-called heterogeneous computing refers to the calculation method of a system that uses computing units of different types of instruction sets and architectures. Common types of computing units include CPU, GPU, ASIC, FPGA, etc. The CPU here is equivalent to commonly used equipment such as pots and stoves. The GPU is like an automatic peeler that does repetitive tasks. ASICs and FPGAs are like cooking machines. The former can only cook special food, while the latter can. Dismantling and changing various foods flexibly, combining them together is a more efficient kitchen system.

Back to reality, cloud service providers such as Amazon, Google, Alibaba, Tencent and other cloud service providers are increasingly demanding computing. At the same time, they are also competing to use artificial intelligence to enhance their service capabilities. They invest a lot of money in new data centers to meet these requirements. demand.

IDC has tracked the IT spending of 4,800 companies around the world. The total IT spending in 2018 is expected to reach 1.1 trillion U.S. dollars, of which Amazon ranks first with 13.6 billion U.S. dollars, and Google parent company Alphabet ranks second with 12 billion U.S. dollars. , Servers and other data center infrastructure account for most of Amazon and Google’s parent company Alphabet.

On the other hand, due to the limitations of physical conditions, it has become more and more difficult for semiconductors to follow Moore's Law to move forward. Heterogeneous computing may be one of the ways to push computing power forward. These have stimulated the development of heterogeneous computing.

Including Intel, AMD, Nvidia, Xilinx, and Alibaba are all conducting research on heterogeneous computing. For example, Xilinx released a new 16nm FinFET 3D transistor-based FPGA product-"Zynq UltraScale MPSoC" in early 2017, which integrates multiple modules such as CPU, GPU, and FPGA; Alibaba Cloud also has a related heterogeneous computing platform launched; in the data In the field of central server computing power acceleration, both Intel and AMD have launched solutions using CPU+FPGA core architecture.

As early as 2012, AMD, ARM, Imagination Technologies, MediaTek and Texas Instruments (TI) jointly established the "Heterogeneous System Architecture Foundation" (HSA Foundation) and became the first group of members. Now this foundation has more than 40 Member companies.

In addition to traditional PC and embedded businesses, AMD’s data center processor business has also been growing in recent years, and it has become increasingly competitive with Intel, which has long dominated the field. It is reported that AMD is processing data centers The market share of the device is expected to reach 10%. According to the latest data from the survey agency PassMark, AMD’s current total CPU market share is 37.5%, the highest level in the past 14 years. The data center processor business is more profitable, and the emergence of heterogeneous computing needs also gives players outside of Intel a window to enter this field. In the future, this business will be more likely to become a key point for AMD's growth.

"IDC Global Server Quarterly Tracking Report" shows that in the fourth quarter of 2019, the global server market vendor revenue increased by 7.5% year-on-year to US$25.4 billion. In combination with this, the data of the Foresight Institute shows that the overall market size of the global server industry in 2020 will be nearly US$90 billion (approximately 602.5 billion renminbi).

Once Xilinx's FPGA business is integrated into AMD, AMD will have more capital to compete with Intel, and it will be more hopeful to occupy a larger share in the rapidly growing telecommunications, AI, and defense markets.

Source: Investigation Agency PassMark

From the perspective of the overall semiconductor market, AMD is not the only one targeting heterogeneous computing. In March 2019, Nvidia agreed to acquire Mellanox Technologies, a chip manufacturer, for US$6.9 billion, whose chips are mainly used to provide high-speed network connections for servers. In September of this year, Nvidia announced that it would acquire ARM, the world's largest chip IP supplier, for US$40 billion and include the CPU in the bag.

If the AMD acquisition of Xilinx can be successful, AMD will integrate the three “king Kongs” of CPU, GPU, and FPGA before NVIDIA. What dynamics will NVIDIA have afterwards? In the arena of heterogeneous computing, the "Three Kingdoms" of Intel, NVIDIA, and AMD may have just begun.