For example, although Samsung has a huge gap in the wafer foundry market, it still has the ambition of anti TSMC; risc-v also wants to break the embedded river that arm has consolidated in recent years by virtue of its open-source feature; AMD is also taking advantage of Intel’s obstruction to pursue bravely. In fact, the above is just the tip of the iceberg of the whole market. In the chip industry, there are a lot of counter attacks going on.
Wafer foundry, Samsung’s counter attack desire
As we all know, TSMC is the largest wafer foundry company, far ahead of other companies with half of the global market share. It is also the first to introduce advanced technology such as 7Nm, and establish a huge “circle of friends”. Recently, TSMC also announced its next step plan. According to economic daily news, Tainan Economic Development Bureau said on the 12th that TSMC Nanke 5-nanometer factory has entered trial production, and is expected to start production next year. Construction of the new 3-nanometer plant is expected to start next year, with a total investment of 1.15 trillion yuan for 5-nanometer and 3-nanometer. In addition, TSMC also stressed that not only Nanke 18 plant will be the flagship plant with the largest investment amount over the years, but 5nm and 3nm will also become the most advanced process of TSMC with more than 50% revenue in the future, and it will be a very important production base for Taiwan and the global semiconductor industry.
According to the latest news, TSMC’s board of directors has approved about US $6.62 billion in capital allocation for the construction of new wafer plants, the installation and upgrading of equipment with advanced technology capabilities and advanced packaging capabilities, as well as R & D investment and continuous capital expenditure in the first quarter of 2020. TSMC’s board also approved a capital allocation of about US $106.1 million for the capitalization of leased assets in the first half of 2020.
At a time when TSMC’s market share was far ahead and its technology was not backward, most people thought that there were not too many opportunities, but Samsung was coming.
As a catch-up, Samsung has been trying to capture TSMC’s market share since it made its wafer foundry independent in 2017. Samsung plans to build a mass production system with EUV as its core. The plan is expected to last for 10 years and strive to catch up with TSMC in OEM business. In response, Samsung vice president Li Zairong said that the company will invest 1 trillion yen in production equipment and R & D in this field every year. As the revenue pillar of semiconductor memory business, Samsung plans to make the OEM production business No.1 in the world.
In Samsung’s conference call in the third quarter, Samsung also mentioned its progress in wafer foundry – Samsung said its 7Nm EUV process would be mass produced in Q4. In addition, according to Korean media reports, Samsung will start mass production of 6-nanometer products in the second half of this year and 5-nanometer products in the first half of next year. It is reported that the 5nm EUV process has been finished and new customer orders have been obtained, but Samsung has not released specific information. After 5nm, Samsung also has an improved version of 4nm process. Samsung mentioned that it is building the infrastructure of 4nm wafer plant. In addition, at Samsung’s OEM forum this year, they also announced the mass production roadmap of 3-nanometer products in 2021, which will introduce the gate all around (GAA) process and adopt a new generation of transistor architecture. It is reported that Samsung has sent the 3 nm gate all around early process design kit to customers.
According to Samsung’s Roadmap, they are trying to compete with TSMC in wafer foundry by using EUV technology. In addition, some people in the industry have analyzed that Samsung has adopted a new transistor architecture in the 3nm process. With the support of EUV technology, 3nm will become the key node of Samsung’s anti TSMC.
At the same time, in the third quarter’s financial report, Li Zairong also stressed his goal in Wafer Foundry: “after memory, in the system semiconductor business including OEM production, it will become the world’s first.”. Earlier this year, according to the system semiconductor development plan “semiconductor vision 2030” released by Samsung Electronics in April this year, Samsung plans to invest 73 trillion won in South Korea by 2030, invest 60 trillion won in production equipment, a total of 133 trillion won, and plan to recruit 15000 system Semiconductor R & D and manufacturing professionals, looking forward to improving the competitiveness of system chip business.
However, according to the market survey agency, TSMC and Samsung Electronics accounted for 50.5% and 18.5% of the market in the third quarter of 2019 respectively, which further widened the gap by 3% compared with the previous quarter. From this point of view, it is not an easy thing for Samsung to achieve counter attack in the field of wafer foundry.
Challenge Sony, Samsung plans CIS
As the industrial responsibility of South Korea, Samsung’s semiconductor ambition is not only wafer foundry, but also a target they want to attack. As we all know, in this field, Sony, the leader, also takes the position of Diaoyutai with more than 50% market share. But Samsung, second in the market, is also trying to shake Sony’s position in the field.
Samsung’s breakthrough point in the CIS field is the demand of smart phones for cameras. According to the South Korean media “Korea business”, LG’s v50s synq front camera lens uses Samsung’s 32 million pixel image sensor. The report also pointed out that Samsung’s systems business unit has identified LG as one of its long-term stable customers. In addition, thanks to the growing demand of Chinese partners such as Xiaomi, vivo and oppo, Samsung’s image sensor business has a foundation to challenge Sony.
In terms of products, in May 2019, Samsung officially released the world’s first 64 megapixel mobile camera CMOS – isocell bright GW1. In addition, Samsung also launched the isocell bright GM2 sensor, 48 million pixels. Both use 0.8 μ m unit pixel sizes and will go into mass production in the second half of the year, according to Lei technology. In addition, in the automotive image sensor market, Samsung also has some plans. According to relevant reports, Samsung has started mass production of the industry’s first 18 million pixel image sensor in August, and the industry’s first 0.7 micron ultra compact mobile image sensor was developed in September 2018, and has achieved mass production in 2019.
Samsung also stressed the importance of CIS when it released the third quarter results at the end of last month. According to Samsung Electronics in the third quarter earnings conference call, in order to meet the growing demand for CIS, Samsung reviewed the overall optimization of semiconductor production capacity, and planned to increase CIS production capacity from the first quarter of 2020. As a result of this optimization, Samsung expects given inventories to return to normal in the first half of 2020. At the same time, Samsung also said it would continue to manage capital expenditure to respond flexibly to changing market conditions, while remaining committed to medium and long-term investment.
But Sony won’t wait to die.
According to Nikkei, Sony’s image sensor market share has exceeded 50% since 2018, and the company has not built a new factory for 12 years. But Sony said that with the popularity of 5g networks in the future, the demand for image sensors in unmanned vehicles and factory automation will only be higher. To this end, the company announced that it would invest 100 billion yen (about 918 million US dollars) in the next fiscal year to build a plant and expand production, and it will be put into production as soon as possible in the accounting year starting from April 2021.
At present, Sony has a capacity of 100000 per month, and Sony hopes to increase the capacity to 130000 by March 2021. On the one hand, of course, this is related to the market demand that the company sees; on the other hand, it is believed that it has a certain relationship with Samsung’s catching up.
AMD: the counter attack of the second man of ten thousand years
AMD and Intel, this is a long-standing story. But today, it seems that the status quo has changed.
According to the October CPU processor sales statistics released by mindfactory, one of Germany’s largest PC retailers, amd processors accounted for as much as 78% of mindfactroy that month, 3.5 times more than Intel. In addition, statistics from Korean channel providers show that in the third quarter of this year, for the first time in history, AMD processor sales exceeded Intel’s, with the shares of 51.3% and 48.7% respectively.
At the same time, there are data in this report that the three generations of sharp dragon account for 74% of AMD’s total revenue. It can be seen that this generation of Intel is the magic weapon of AMD. According to Netease technology news, some analysts said that for the first time, the sharp dragon 3000 processor has outperformed the Intel processor in terms of technology and performance, especially in the single core performance. According to the data released by AMD, the single core performance of the sharp dragon 3000 processor is better than that of the core i9-9900k, although it is only about 1% higher.
In addition, in the third quarter, amd also launched the second generation AMD epyc (Xiao Long) processor. According to its official data, AMD is the first manufacturer to launch an x86 architecture server processor based on 7Nm technology. The second generation AMD epyc processor has doubled core density and IPC optimization, twice the integer performance results, and 4x the floating-point performance of the previous amd server family. Based on the hybrid multi chip architecture, this second generation AMD epyc (Xiaolong) CPU decouples the processor core from the input / output (I / O) chip, provides an independent manufacturing timeline, and ensures that both can develop at their own pace without restricting each other. According to its financial report, Google, twitter, Dell, HPE and Lenovo have adopted the second generation Xiaolong processor.
In September this year, amd updated its CPU, data center CPU and GPU architecture roadmap. According to its roadmap, amd released the 7Nm zen2 architecture this year, and AMD will release the zen3 architecture with 7Nm + in 2020. It is reported that the design has been completed.
On the CPU of the data center, the first generation of 32 core Naples processor has come into the market. At present, it is a 64 core Roman processor, zen2 architecture, and zen3 architecture is Milan processor. The design of zen4 architecture Genoa processor has been completed.
On Intel’s side, third quarter results showed that Intel’s PC business fell 5% in the quarter. But at the same time, we also need to note that Intel is in the process of strategic transformation from PC to data center, and its diversified product portfolio means that it can have great market potential. According to the news, Intel said in its third quarter earnings call that 18 advanced production systems based on its 10th generation core (ice lake) processor have been officially launched so far, and another 12 are expected to be launched in 2019. Starting next quarter, Intel expects that 10 nanometer chips will also be shipped from its Arizona plant, which will increase the supply of ice lake processors. Who is better between the two is far from the time when the hammer is set.
In addition to the competition from AMD, Intel also faces the threat of another “a” prefix enterprise on the infrastructure. At present, Intel’s x86 architecture and arm’s arm architecture are two important parts of the global CPU architecture. For a long time, arm is good at designing low-power processor, so it is favored by mobile terminal. X86 is in the server market with nearly 90% of the market share, established an unparalleled advantage. In recent years, some enterprises have invested in the arms of arm architecture server, which also let the industry smell the breath of breaking through barriers.
Arm server chip will shake the tree?
As mentioned above, arm server chip manufacturers have long coveted Intel’s near monopoly market. According to the manufacturer’s introduction, arm architecture has the ability to enter the server chip. Many chip R & D enterprises begin to use arm architecture to develop server chips. According to some previous reports, Amazon and Huawei have successively released arm based server chips.
Recently, a new message came from the server based on ARM architecture. According to Marvell’s official news, Marvell has worked with Microsoft to successfully apply arm based thunder x2 server to Microsoft’s internal workload. It is learned that the second generation of arvelio is based on the maximum capacity of the server, which is based on the server’s dual channel configuration.
According to the news from the Japanese media on November 13, arm is promoting the next generation of arm architecture that supports servers. Arm is gradually expanding the current armv8 architecture, and the ultimate expansion vertex is regarded as a bridge to armv9: for armv9, people in the CPU industry have said that security expansion is the most important key. Arm focuses on security in the phase expansion after armv 8.3 and is seen as a step towards armv9.
In addition to upgrading the architecture, arm released the new neoverse brand at the Techcon conference last year, and formulated a clear roadmap for the future development of IOT infrastructure and server chips. According to previous reports by semiconductor industry observer, arm’s current neoverse series is based on Cosmos platform with A72 and A75 cores and adopts 16nm technology. AWS graviton processor is developed based on this platform. Arm promises 30% performance improvement per year in the next three years. This year’s main platform is code named Ares, Zeus in 2020 and Poseidon in 2021. The Neover se processor is designed to hold up to 256 cores, with faster memory interface, inter chip interconnection bandwidth, security features, and virtual machine and other high-performance computing technologies.
In April this year, it was also reported that arm invested in the server chip company ampere, which further reflected their attention to the market. Ampere is a company founded and operated by Renee James, former president of Intel, with the support of Carlyle Group, which is committed to the field of arm server chips. Ampere builds its own kernel and its own design on the basis of arm64. Its products can be used for ultra large scale cloud processor of storage and server. It is reported that the new company has launched the first arm server chip. Recently, apple chip executives founded nuvia, I believe, also wants to use arm server chips to steal food from Intel.
Although the arm architecture has been constantly updated, and server vendors continue to adopt this architecture. But it’s still a long way to go. According to IDC statistics, worldwide revenue of X86 servers will reach $18.5 billion in 2019, while non x86 servers will drop 13.7% to $1.3 billion year-on-year. It can be seen that the dominant position of X86 servers in the server market is still strong.
Rise of risc-v, arm like walking on thin ice?
When arm was trying to move from the mobile to the server, it seemed that there was a fire in arm’s backyard. Risc-v, an open-source architecture, is setting off a whirlwind in the world.
As a chip with open source as its selling point, risc-v has great advantages, which not only attracts the attention of many domestic manufacturers such as pingtouge and xinlai technology. Zhaoyi innovation also released the MCU based on RISC – V architecture earlier. At the same time, TSMC and Samsung have also carried out OEM business related to RISC – V architecture. Ni Guangnan, academician of the Chinese Academy of engineering and chairman of the China open instruction Ecology (risc-v) Alliance (crva), also said that with the rise of 5g in China, the Internet of things, big data, edge computing and blockchain are also booming. These are very good application fields for open-source chips. In the new field, several architectures are new, risc-v will have great advantages.
Under the fierce attack of risc-v, arm is also seeking change.
Not only are they changing their licensing model. At arm Techcon 2019 in October this year, arm announced that it would introduce custom instruction capabilities into some CPU cores, that is, customers can write their own custom instructions to accelerate their specific use cases, embedded and Internet of things applications. In the custom instruction function, arm allows customers to create their own acceleration algorithms and run them directly on the CPU. At present, arm has entered into cooperation with many partners to support custom instructions, including IAR systems, NXP, core technology, St STM, etc.
In the past few decades, the reason why technology has been able to achieve earth shaking changes has a lot to do with the upstream efforts of these chip manufacturers. But since the invention of integrated circuit in 1958, we have seen many changes in the whole industry. Whether it is market transformation, or technological innovation, or the merger and acquisition of competitors, or even low-cost competition. The practitioners of this industry are facing all kinds of challenges all the time. Especially in this era of rapid change, if you don’t always pay close attention to the market and keep alert, maybe when you see the change, it will no longer be the “world” you are familiar with.
But for these manufacturers, how to deal with them while knowing where their challengers are is also the fundamental to ensure the sustainable development of enterprises.
Editor in charge: CT