domeet webmaster

This article is from WeChat official account:Semiconductor industry observation (ID: icbank), author: Du Qin DQ, from the head of FIG: pixabay

No one wants to be the second in the millennium. In the semiconductor industry, only the boss is nourished. The second is very difficult to survive. But the second child has the advantage of having a goal, always thinking about headwinds. As a second-best AMD in CPU and GPU, recently, the chip industry has reported that it will swallow Xilinx at a price of more than 30 billion US dollars. It must be noted that at this stage AMD wants to make any acquisitions, and we have no reason to question (after all, rich). Even if hit by Xilinx transaction rumors, AMD’s market value is still US$101.6 billion, and it is no longer the “brother” that hit a low of US$1.45 billion in August 2015.

In recent years, AMD’s Epyc processor product line has gained approximately 10% of the server CPU market; on the PC side, they have also achieved good performance in contending with Intel; at the same time, they are also sitting on second-generation game console chips. business. From a market perspective, AMD is definitely at the highest level in the local area. So why does AMD, which is gaining market share, want to acquire a company that has a stable market share in its core FPGA business for the past six years?

“Nirvana Rebirth” AMD

In the semiconductor industry, AMD can be said to be one of the typical representatives of “turning over the serf to sing.” In 2015, AMD’s stock price once remained at around US$2.6, but now it is the “big brother” that you can’t afford to climb at US$84.29 next door. After years of insolvency and losses, AMD is struggling tenaciously as the second child in the semiconductor CPU and GPU field.

Source: Yahoo Finance

In the past two years, thanks to its own design advantages and the support of TSMC’s advanced manufacturing process, AMD’s position in the processor has increased day by day, and it has gradually gained the strength to challenge Intel. And this is inseparable from its self-developed Zen architecture. What if AMD’s biggest move in this decade? The Zen architecture must be on the list. Another is to embrace the turnaround brought by TSMC. Before that, Lisa Su shared on Hotchips that 60% of the performance improvement of AMD CPU products comes from the fab.

In a survey report of electronic devices running PassMark, AMD has captured 40% of the CPU market share for the first time in 14 years. In fact, since the introduction of processors based on the Zen architecture in 2017, AMD has been grabbing market share and continuously increasing. At present, x86 is mainly divided into two parts, desktop and mobile in the consumer market, among which AMD’s share of the desktop market has reached 48%. Since the release of the Ryzen 3000 series processors last year, AMD has continued to conquer the market with 7nm Zen2, and its market share has also risen from 40% to 48%, which is almost half the world.

In June 2017, AMD returned to the server chip market. Its first-generation EPYC server Naples(Naples) was born and broke Intel’s monopoly in the PC market for many years also shows its determination to re-initiate an impact on the data center processor market. And its second-generation EPYC Rome is called “the strongest x86 server chip in history”. Compared with Intel Xeno series processors, EPYC doubles its performance while at half the price.

Su Zifeng previously stated in an interview that AMD’s share of the X86 server processor market at the end of last year was about 5 to 6%, and by the end of 2020 this share will increase to 10%. This 10% share The goal is also on average, Su Zifeng said that they may get a higher share in certain areas.

In the GPU field, there has always been a virtuous and eternal battle between AMD and Nvidia. AMD has the upper hand in the mid-range market, while Nvidia excels in the high-end segment. AMD has some advantages in price, but Nvidia is expensive but there are many products with good performance. In the final analysis, both companies rely on competition to thrive. According to the GAMINGSCAN report, in the GPU field, there is no real winner, and both have their areas of expertise and areas for improvement. In most cases, AMD will be better in terms of budget and mid-range graphics cards, while Nvidia is a good high-end graphics card.

AMD & Intel’s arms acquisition race is in full swing

Actually, in the final analysis, the battle between AMD and Intel ultimately focused on PCs and data centers. Over the years, Intel has been expanding its network product line and adding products for data centers.

In July 2011, Intel acquired Fulcrum Microsystems, a manufacturer of low-latency Ethernet switch chips, at an undisclosed price. Fulcrum’s technology can complement Intel processors and Ethernet controllers and help Intel become a comprehensive data center provider.

In January 2012, they acquired QLogic’s InfiniBand network business for US$125 million. This acquisition will help expand Intel’s network product line and provide technologies to increase the internal bandwidth of the system.

In April 2012, Intel acquired the “Gemini” XT and “Aries” XC supercomputing interconnection business from Cray for US$140 million, of which QLogic InfiniBand and Cray Aries technology constitute the Omni-Path interconnection Foundation. This interconnection is mainly paired with Intel and “Knights” Xeon Phi parallel processors, but Xeon Phi failed to proceed as planned, and the production line was officially discontinued in July 2018. Last week, the Omni-Path network business, including QLogic and Cray intellectual property rights, was separated from Intel and established a company calledThe new company of Cornelis Networks, the company was co-founded by QLogic and Intel staff.

The largest acquisition was in June 2015. Intel announced that it would acquire Altera, the world’s second largest FPGA manufacturer, at a price of $16.7 billion, making it the most expensive acquisition time in the company’s history. The acquisition of Altera helps Intel defend and expand its most profitable business: the supply of server chips used in data centers. As more and more users use tablet computers and smart phones to surf the Internet, PC chip sales are also declining. Data centers that require a large amount of production information and services for mobile devices have also promoted the number of orders for Intel’s high-end processors. Today, Intel announced that its FGPA has been officially applied to mainstream data center OEM manufacturers.

In the field of artificial intelligence, which has been hot in the past few years, Intel has naturally not let go. In recent years, Intel has been active in this field. The current Intel executive vice president Navin Shenoy once said at an event at Intel that as Intel transforms into a data-centric company, Intel’s AI chip business is of strategic importance to the company.

In 2016, Intel acquired Movidius, a start-up visual processing company, aiming to create a general in the field of computer vision applications; in 2018, Intel acquired Vertex.A, a company focused on the development of deep learning compilation tools and supporting technologies Seattle startup company. Intel stated in a statement that with this transaction, Intel has obtained an experienced team and intellectual property rights to further realize flexible deep learning of edge computing; in 2019, Intel acquired Habana Labs for 2 billion US dollars, Habana is a company The company that produces programmable deep learning accelerators, it is understood that this acquisition will bring faster and more accurate artificial intelligence services to functions such as photo and voice recognition.

In addition, autonomous driving is also regarded by Intel as a major development area. In March 2017, Intel spent $15.3 billion to include Mobileye under its command. Prior to this, Intel also bought Yogitech, Arynga, Itseez, Nervana Systems, Movidius, and many other companies from various countries that are involved in autonomous driving.

In general, Intel’s combination of AI products can be said to cover edge computing, data centers, and various business support from CPU to ASIC. AMD does not seem to be so aggressive, and not many companies have been acquired over the years.

AMD received 3.The purchase of the microserver manufacturer SeaMicro for $3.4 billion was a huge sum of money for AMD at the time, and this was done mainly to obtain the 3D torus interconnect “Freedom” and composability middleware created by SeaMicro , To create a micro server system with a certain degree of scalability. This strategy was not properly combined. AMD found that it was unable to sell the “Freedom” interconnect with Opteron CPUs to partners. For this reason, they tried to sell SeaMicro systems with partners. After Lisa Su took charge of the company, they decided that this would not work, so they wrote off all their assets in early 2015.

In April 2017, AMD announced the acquisition of Nitero, the company focused on wireless VR technology, so that VR helmets can be worn wirelessly. AMD clearly sees a long-term market opportunity for Nitero in VR/AR wireless solutions. Headquartered in Austin, USA, Nitero mainly provides wireless VR/AR transmission solutions. There are two core technologies: one is a wireless 60 GHz transmission solution, which uses millimeter wave communication technology; the other is a coding technology for low-latency head-mounted display to obtain image data. But recently, Nitero has also begun to switch to providing chips for PC-end headsets.

What is the biggest benefit of AMD’s acquisition of Xilinx?

In 1984, Ross Freeman and Bernard Vonderschmitt founded Xilinx to develop more affordable programmable chips. This has changed the semiconductor industry in many ways. Fast forward to 2020, FPGA pioneers have invested themselves in the operations of chip giants, and innovations around highly programmable solutions continue.

According to seekingalpha’s analysis, Figure 1 shows the breakdown of AMD’s market share and Intel’s desktops, laptops and servers during the same period. AMD’s share of desktop computers is close to 15%, due to AMD’s launch of 32-core 3970X and 24-core 3960X Threadripper processors at the end of 2019. Compared with Intel’s latest i9-10980X CPU, they have higher performance in video and 3D rendering.

Comparison of AMD and Intel’s desktops, laptops and servers share (Source: seekingalpha)

In the core business, AMD has been in the weak MPU market in recent years (the compound annual growth rate from 2013 to 2019 is 5.2%)Be ahead of Intel in the market. Xilinx’s FPGA market share of 52% in 2019 is ahead of Intel’s 35% market share. FPGA revenue accounts for approximately 12% of MPU’s share. With the acquisition of Xilinx, AMD may usher in an increasing market share of MPU.

Let us take a look at Xilinx’s FPGA and AMD’s MPU revenue. As shown in the table listed in seekingalpha, in the past six years, Xilinx’s FPGA market share has been hovering around 50%, but Xilinx CAGR is 0%. In contrast, AMD’s share of the entire MPU market is small, but it has since increased. Multiplying the market share by the overall market yields the revenue of the two companies. In this case, we can see that the revenues of the two companies in the two markets are roughly the same. But AMD’s revenue has been growing at a compound annual growth rate of 7.4%, while Xilinx’s FPGA revenue has only grown at a compound annual growth rate of 4.8%.

Comparison of revenue of FPGA and MPU between Xilinx and AMD (Source: seekingalpha)

For AMD, another big advantage of the acquisition is that Xilinx began to produce smaller size devices. Doing so will increase the number of chips on the wafer, reduce costs, and improve chip performance. Xilinx puts its 28nm, 20nProducts produced at m and 16nm nodes are called “advanced products”, while all other products are classified as “core products”.

In 2014, only 24.7% of Xilinx’s FPGA revenue was advanced products, and by 2019, advanced products accounted for 70.8% of revenue. From 2014 to 2019, the compound annual growth rate of advanced products was 30.7%, while the compound annual growth rate of core products was -12.4%.

Xilinx core product VS advanced product (Source: seekingalpha)

However, as pointed out above, Xilinx’s market share has hardly increased in recent years. Obviously, the growth of the FPGA market is not a reason AMD is willing to spend $30 billion on acquisitions. Even if FPGAs will replace GPU AI accelerators in the data center and edge, the market is currently small but has huge growth potential. However, it also has no reason to spend $30 billion on acquisitions.

For AMD, the most important thing is Xilinx’s ACAP solution. Xilinx has been committed to the development of its adaptive computing acceleration platform (ACAP) for many years. ACAP is the name of the company, its combined “adaptive hardware” (programmable logic), scalar engine (CPU), intelligent engine (AI or DSP accelerator) and hard IP block connection and security. The company spent more than $1 billion and four years to develop this new chip.

In early 2020, Xilinx released the details of its Versal Premium series,Compared with linx’s previous generation FPGA, it provides three times the bandwidth and doubles the computational density of accelerated workloads in the data center. The 7nm Versal ACAP series products are the successors of the 16nm Virtex UltraScale+ fpga launched by Xilinx in 2015. Most importantly, Xilinx often points out that Versal ACAP solutions are considered a separate product category, not FPGAs, and reports that ACAP far exceeds the capabilities of FPGAs.

The target markets of Versal ACAPs include data centers, wireless 5G communications, aviation and defense radars, automotive driver assistance (ADAS) and wired communications . These technologies represent the next generation of growth in the semiconductor market, requiring significant performance improvements and greater power efficiency. According to Xilinx, in the emerging era of big data and artificial intelligence, Versal ACAP is an ideal choice for accelerating widespread applications.

Conclusion

When AMD walks the way of Intel, will the voice of “AMD YES!” be louder? AMD is currently in the strongest position in its 51-year history. Although AMD is not at its peak, it may be better. A healthy and competitive AMD can fully face Intel and Nvidia, which will bring more benefits to the entire semiconductor market and even users.

This article is from WeChat official account:Semiconductor industry observation (ID: icbank), author: Du Qin DQ