Server chips are competing, a proven semiconductor track

Yesterday, according to the first financial report, it was rumored that Huawei’s x86 server business sale has made substantial progress. The industrial and commercial registration shows that the shareholder of Huawei-controlled Hyperfusion Digital Technology Co., Ltd. has been changed to Henan Hyperfusion Energy.

As the main force of Huawei’s servers, X86 has brought a lot of revenue to Huawei. The choice to sell x86 is more due to the helplessness of the US restrictions. In addition to X86, Huawei also has ARM architecture in the field of cloud computing, such as Kunpeng series and Ascend computing. Huawei’s experience has become an accelerator for domestic cloud computing self-developed chips.

On October 22, Shenzhen Yuxian Microelectronics received over 100 million yuan in angel financing. Yuxian Microelectronics is a high-performance cloud computing CPU company. The start-up company established last year was co-founded by former Intel and ARM executives. They are developing the first domestically produced high-end server CPU based on the ARM architecture, 4nm process, and 160 cores.

Cloud computing, as the nerve center of the Internet, maintains the stable transmission of information, and the attention of its chips is increasing day by day. Domestic cloud computing manufacturers are bringing forth new ideas, and cloud computing chips have become the next explosion point.

Cloud computing is growing against the trend, and Chinese manufacturers have an empty window of opportunity

In the 2021 China Academy of Information and Communications Technology’s “Cloud Computing White Paper”, the data shows that in 2020, my country’s cloud computing market will grow explosively, with the overall market size reaching 209.1 billion yuan, a growth rate of 56.6%. Among them, the size of the public cloud market reached 127.7 billion yuan, a year-on-year increase of 85.2%; the private cloud market reached 81.4 billion yuan, a year-on-year increase of 26.1%. Cloud computing gradually appears siphon effect, and forms a new service mode in security and network edge. JP Morgan reports that by 2025, the chip market for data centers will be worth nearly $18 billion.

For cloud computing to develop rapidly, it is necessary to develop reliable core hardware. On the chip, server chips and memory chips are used as the two main positions.

China’s server market recovered quickly during the epidemic, maintaining a year-on-year growth of 17.6% last year. With the advancement of the national 14th Five-Year Plan and the investment in new infrastructure, China’s server market will maintain healthy and stable growth in the next five years. From 2021 to 2025, the size of China’s server market will increase from US$25.731 billion to US$41.029 billion, maintaining a compound annual growth rate of 12.5%, and server chips will also usher in a broad market.

At present, the architecture of the global server chip is still dominated by X86, and the share of ARM and other architectures is small. According to data from JP Morgan, there is no clear sign that a large number of data centers are starting to move from the x86 architecture to the ARM architecture. It is expected that by 2025, X86 will still be the main architecture of the data center, and it is expected that it will still have a market share of 90%.

Domestically, IDC data shows that in 2020, China’s server market shipments will be 3.5 million units, a year-on-year increase of 9.8%; the market size will be 21.649 billion US dollars (about 148.99 billion yuan), a year-on-year increase. 19.0%. Among them, the X86 server market shipments were 3.4393 million units, a year-on-year increase of 8.1%; the market size was 20.823 billion US dollars (about 143.32 billion yuan), a year-on-year increase of 17.7%.

The reason why the X86 architecture cannot shake its position is that its software ecosystem is already very large and rich, and everything from software to OS has been very fixed. However, this inherent pattern is changing in the cloud era – ARM’s share is increasing.

Due to the expensive and difficult licensing of X86, many companies have begun to build ARM-based self-developed server chips. Amazon launched its own server chip Graviton under the ARM architecture. At the end of 2019, it released the Graviton2 processor based on the Arm architecture, which can be used for AWS EC2 services. Compared with similar products of x86 Intel Xeon, the computing capacity is increased by 4 times, and the cost performance is increased by 40%.

With the support of Graviton2, AWS topped the Arm-based server processor list in the 2021 server leader list, ranking first in the five evaluation dimensions of market, price advantage, performance, reliability and innovation.

At present, Graviton has been widely commercialized. At the beginning of this year, aws officially announced that Graviton 2 officially landed in China. Amazon Elastic Compute Cloud (Amazon EC2) M6g, C6g, and R6g instances powered by Amazon Graviton2 processors are available in the Amazon Web Services China (Beijing) region, operated by Sinnet and Amazon Web Services China (Beijing), operated by NWCD This is the first time that its self-developed processor has landed in China.

In China, HUAWEI CLOUD’s server architectures are ARM and X86 respectively. Among them, servers based on the X86 architecture account for the vast majority of Huawei’s server shipments.

At the end of last year, Huawei had a 5% share of the global server market. Due to the US ban, Huawei has fallen into the “other” queue internationally at the beginning of this year. Affected by the unavailability of X86 shipments, Huawei’s server business is experiencing heavy losses.

In addition to the X86, Huawei has retained the ARM architecture-based server product – Kunpeng. In 2019, Huawei released the high-performance processor Kunpeng 920, manufactured with 7nm process, based on ARMv8 architecture, with 64 2.6GHz cores, supporting 8-channel DDR4, PCIe 4.0 and CCIX interconnect chips, which is lower than X86. power consumption.

On March 27, 2020, Huawei held the 2020 Developer Conference online, announcing the latest progress of the Kunpeng + Ascend computing platform ecology. Hou Jinlong, president of Huawei Cloud and Computing BG, said that he hopes to make 90% of applications run on Kunpeng within three years. Hou Jinlong said that all applications in the 5G era are going to the cloud. Mobile phones, tablets and other end-side are based on the ARM architecture. Kunpeng is also based on the ARM architecture. After the cloud and the end are isomorphic, the performance can be improved by 40%. This is the Kunpeng and x86 architectures. Compared with natural advantages. It is reported that at present, Huawei Kunpeng processors are mainly used in the scenarios of party and government organs, public institutions, large state-owned enterprises and state-owned banks.

In addition to Huawei Kunpeng, there are also X86-based Zhaoxin and Haiguang, and arm-based Feiteng. At the same time, domestic commercial open server chips also include self-developed instruction sets including Shenwei and Loongson.

The spring of memory chips

RightScale released a “2018 State of the Cloud Survey”, among 1000+ respondents, 81% of enterprises have adopted a multi-cloud strategy. Due to the needs of enterprise users in a multi-cloud environment, the choice of enterprise-level storage will not be reduced.

IDC report shows that since this year, China’s external storage market has maintained rapid growth. In the first half of this year, the scale of China’s external storage market reached 2.39 billion US dollars, a year-on-year increase of 33.5%. Among them, Huawei’s market share occupies a leading position, reaching 29.6%; followed by New H3C, Inspur, Lenovo and Hikvision, with a market share of 12%, 10.4%, 7.9% and 6.1% respectively %.

In the field of domestic self-developed memory chips, Ziguang, Jinhua, Changcun and Changxin are the main players. In recent years, Tsinghua Unigroup’s Yangtze Memory, Wuhan Xinxin, and Zhaoyi Innovation and its partner Hefei Changxin have gradually broken through technical barriers in the field of DRAM and Flash.

YMTC is the first manufacturer in China that can complete 3D NAND Flash memory chips. GigaDevice will have a market share of 15.6% in the global NOR Flash market in 2020, ranking third, second only to Winbond and Macronix.

Technically, as the process becomes more and more difficult, the drawbacks of traditional memory chips have gradually begun to appear, DRAM is more susceptible to the influence of external charges; Flash faces serious crosstalk problems during operation, thus shortening its service life, PCRAM and MRAM have become the cloud. A rookie in computing architecture.

Intel and Micron jointly launched 3D Xpoint technology in 2015, which is another area of ​​PRAM development. A previous report by Objective Analysis and Coughlin Associates, “Emerging Memories Ramp Up,” predicted that 3D XPoint, in the form of Intel Optane, could generate up to $16 billion in revenue growth by 2029, although it never explicitly stated that this emerging memory was PCRAM, But the industry consensus is that 3D Xpoint is a variant of PCRAM.

One of the two major technical directions for PCRAM to break through storage capacity: three-dimensional stacking 3DXpoint.

MRAM has been developing since the 1990s. Its speed is no less than that of SRAM, and like flash memory, it can retain data after power failure, and its energy consumption is lower than that of DRAM. It is one of the candidate technologies for general-purpose memory in the future. Earlier this month, Shanghai Gencun Technology received tens of millions of Pre-A financing, which is a company engaged in the research and development and design of MRAM memory. Its person in charge said that MRAM has a potential market of tens of billions of dollars. The premature investment of heavy assets, immature preparation process, continuous optimization of benchmarking technology, unprepared application side, and inability to implement products. The earliest batch of MRAM companies Developments were mostly uneventful, and the real timing for MRAM only started in 2019.

Today, with the rapid development of information technology, the pace of digital transformation in various industries is accelerating, the speed of intelligence and digitization centered on the cloud is accelerating, cloud services are becoming more and more mainstream, and cloud IT facilities are ushering in a major historical opportunity. For semiconductor giants, cloud computing is a must-have sector. At this stage, the 14th Five-Year Plan has just started, and new infrastructure is getting better. China’s cloud computing is developing rapidly, and it is also a time to wait and see.

The Links:   LM64K836 EL640.200-SK 7MBR50SB120-50