|
Original title: "A panoramic view of China's chips, read the 10,000-word article first! "
Original source: Chuangyebang Author: Cheng Kuiyuan
In April 2018, the U.S. Department of Commerce imposed sanctions on ZTE, a major Chinese telecommunications equipment manufacturer, for violating the trade embargo against Iran and North Korea, requiring relevant U.S. companies not to provide components to ZTE for seven years, including the most critical microchips and other products.
In May 2019, the U.S. Commerce Department said it had added Huawei and 70 of its subsidiaries to the Entity List, a move that prohibits telecom giant Huawei and others from buying components from American companies without U.S. government approval.
From the ZTE incident to Huawei being included in the Entity List, although it is a sanctions action on the surface, it essentially reflects the lack of domestic chip autonomy.
According to customs data, China's chip imports in 2018 exceeded US$312.058 billion, a year-on-year increase of 19.8%, setting a record high. Chips have surpassed crude oil to become the largest category of imports in my country.
The export of chips was only 84.636 billion US dollars, and the import amount was 3.7 times the export amount. And judging from the data of the past five years, China's chip trade deficit is getting bigger and bigger. China's chip self-sufficiency rate is seriously insufficient. In 2018, China's chip self-sufficiency rate was only about 15%. From the perspective of core chip self-sufficiency, the self-sufficiency rate of core chips such as processors, GPUs, and memory is seriously insufficient, but domestic companies have a high self-sufficiency rate in mobile phone chips, artificial intelligence, packaging, etc. In terms of mobile phone chips, Huawei Kirin chips are the representative, and their performance has reached the world's leading level.
The technology in the packaging and testing links is relatively low, and my country, as a large labor-intensive country, has inherent advantages. There are three leading companies in the domestic packaging and testing field, namely, Changdian Technology, Huatian Technology and Tongfu Microelectronics. All three companies have entered the top ten in the global packaging and testing industry.
In terms of artificial intelligence chips, domestic traditional Internet giants and artificial intelligence startups have actively made plans. Artificial intelligence chips represented by Alibaba's Hanguang 800, Cambrian NPU, and Horizon's autonomous driving chips have achieved remarkable results. Overall, the domestic chip market is large in scale but lacks self-sufficiency; mid- and low-end products are developing rapidly, with breakthroughs achieved in niche areas, but the core is subject to control by others.
Therefore, with the support of national policies and funds, my country should step up efforts to conquer core technologies, vigorously develop domestic chips, accelerate domestic substitution of chips, and build a Chinese chip.
This article will give a relatively comprehensive introduction to chips, including the chip industry chain, digital chips, AI chips, analog chips, as well as the main structure and players of the world's chips, and attempt to explore opportunities for domestic chips.
Chip Introduction and Industry Chain
1. Chip Introduction
Chips are everywhere in our lives, including smartphones, computers, home appliances, cars and even the military industry. Although small in size, chips have laid the foundation for the informatization and intelligence of all walks of life.
The history of chips can be traced back to the birth of transistors. In 1947, Bell Laboratories in the United States manufactured the world's first transistor. The emergence of transistors made it possible to integrate various devices and circuits on a dielectric substrate, and the concept of integrated circuits was born.
In 1958, Jack Kilby, who worked at Texas Instruments (TI), successfully developed the world's first integrated circuit by connecting several transistors, resistors, and capacitors on a germanium (Ge) substrate. He found that the efficiency of integrated circuits had a huge advantage over discrete components. A few months after Jack Kilby invented the germanium-based integrated circuit, Robert Noyce invented the silicon (Si)-based integrated circuit. Most of the semiconductor applications today are based on silicon-based integrated circuits.
The transistors, resistors, capacitors, inductors and other components and wiring required in a circuit are interconnected and made on a small piece or several small pieces of semiconductor wafers or dielectric substrates, and then packaged in a tube shell to form a microstructure with the required circuit function. This is an integrated circuit (IC), also known as a chip. All components in the chip have formed a whole in structure, making electronic components a big step towards miniaturization, low power consumption, intelligence and high reliability.
2. The Importance of Chips in Electronic Devices
Let’s take the iPhone 11 Pro Max as an example to look at the importance of chips in electronic devices. Remove the back cover Remove the motherboard Disassemble the motherboard
iPhone 11 Pro Max motherboard structure Orange: Apple APL1092 power management chip Green: Decawave package chip (U1 ultra-wideband chip)
Yellow: Cirrus Logic 338S00509 audio decoder Blue: Avago 8100 mid/high frequency PAMiD RF transceiver Purple: Skyworks 78221-17 low frequency PAMiD RF transceiver Red: Apple APL1W85 A13 Bionic SoC, integrated SK Hynix 4GB memory Pink: STMicroelectronics STB601A0N power management chip
iPhone 11 Pro Max motherboard back structure Red: Toshiba TSB 4226VE9461CMNA1 1927 Flash
iPhone 11 Pro Max RF circuit board
Blue: Verizon 81013 -Qorvo packet tracking module Orange: Intel X927YD2Q modem baseband chip Green: Skyworks 78223-17 power amplifier Purple: Skyworks 13797-19 5648169 1927 MX
Yellow: Intel 5765P10 A15 08B13 H1925 transceiver Pink: Intel 6840P10 409 H1924 baseband power management chip Red: Apple, USI 339S00648 WiFi/Bluetooth chip
Charging module, source: iFixit
Orange: Apple 338S00411 audio amplifier Yellow: Texas Instruments (TI) SN261140 battery charging chip Red: STMicroelectronics (STMicroelectronics) wireless charging chip
In summary, we can see the importance of chips to electronic devices. There are a large number of chips in a smartphone, providing various functions. Specifically, the CPU, GPU, neural network engine, and external baseband chip integrated in the A13 Bionic SoC are microprocessors; 4G RAM, namely DRAM, and flash memory, namely NAND flash, are storage chips; RF chips (transceivers, power amplifiers), audio and video multimedia chips, power management chips, etc. are analog chips.
Similarly, chips play an important role in other electronics and devices. Chips are present in non-intelligent devices such as remote controls, air conditioners, LED bulbs, etc. With the rapid development of 5G and AIoT, intelligence, interconnection, and cloud computing increasingly require chips to play a role. In particular, artificial intelligence chips have become a competitive highland for traditional chip manufacturers, Internet/technology giants, and AI startups.
3. Classification of Chips
There are many ways to classify chips. Depending on the signals they process, chips can be divided into digital chips and analog chips.
Digital signal: A signal whose information parameters are discrete in time and amplitude. It is a discrete signal obtained by sampling and quantizing an analog signal, expressed in binary (0/1). Its characteristics are: discrete changes in time and amplitude, easy to store, no attenuation, and more suitable for high-speed processing.
Analog signal: Information parameters are expressed as continuous signals within a given time range, such as temperature, pressure, sound, and image. Its characteristics are: the amplitude changes continuously over time, and it can truly and realistically reflect the physical world we live in, but it is easy to decay and difficult to store.
Digital chips process digital signals, while analog chips process analog signals. Taking smartphones as an example, external analog signals such as images obtained by taking photos, fingerprints or faces required for recognition, etc. are processed by analog chips, and then converted into digital signals through analog-to-digital modules, and then processed by digital chips; the processed digital signals are also converted into analog signals through digital-to-analog chips for external transmission as needed.
4. Chip Industry Chain 1. Upstream: International giants monopolize high-end general-purpose chips, and domestic companies strive to catch up
Chip design is the top of the chip industry chain, including a series of steps such as architecture selection, logic design, circuit design, and packaging design. According to DIGITIMES Research, only one mainland company, Huawei HiSilicon, made the top ten global IC design companies in terms of revenue in 2018. Although mobile processor chip design companies represented by Huawei HiSilicon have entered the forefront of the world, the overall level of domestic chip design is still far behind that of international chip giants, and high-end general-purpose chips such as CPU, GPU, and FPGA are still monopolized by international giants. In recent years, my country's design industry has developed rapidly, with the industry's growth rate far exceeding the international average. Huawei HiSilicon has achieved 7nm advanced process and is also at the forefront of the world in 5G chip technology; Ziguang ZTE and Datang are actively deploying 5G; Goodix Technology's fingerprint recognition chips are used in most domestic smartphones; Cambrian and Horizon Robotics' AI layout is emerging internationally.
(1) Business model of chip design
According to the business model of chip design, it can be divided into IP design and chip design.
IP design, that is, designing chip IP core (IPcore). IP core is a commonly used chip design module. Nowadays, chip design is no longer completely starting from scratch, but is based on some mature IP cores, and chip functions are added on this basis.
Take ARM as an example. It does not produce chips itself, but uses processor licensing, processor optimization package licensing and architecture licensing as its business model. Chip design companies obtain ARM's authorization, obtain the IP core of ARM chips, and conduct further chip development on this basis.
Chip design, that is, targeted development based on the needs of market segments through independent architecture or licensed architecture.
(2) Chip architecture
In the design stage, the most important thing is to choose the instruction set architecture.
The instruction set architecture is not just a collection of instructions, it also defines the hardware information combined with the software, and uses hardware circuits to implement the operations specified by the instruction set. Processor architecture design is the highest and most important level in the current chip industry.
The instruction set architecture can be understood as an abstract layer, which constitutes the bridge and interface between the underlying hardware of the processor and the software running on it. It is also the most important abstract layer in current computer processors.
At present, the mainstream architectures in the world are the ARM architecture led by ARM and the x86 architecture led by Intel. Overall, the ARM and x86 architectures almost divide the entire architecture market. The RISC-V architecture, which is flexible, streamlined, and open source, is also receiving more and more attention. It has great potential in the chip fields such as the Internet of Things and AI, and has become a new opportunity for Chinese chips.
① CISC and Intel x86
The commonly used PCs or servers now mainly use CPUs from Intel and AMD. The instruction set used by this type of CPU belongs to CISC (Complex Instruction Set Computer). A CPU can support many instruction sets, and early CPUs were all based on CISC.
On June 8, 1978, Intel produced the world's first 16-bit microprocessor and named it "i8086". The x86 architecture was born, which defined the basic usage rules of chips. In the following decades, the x86 architecture continued to improve, and the x86 instruction set was used as a standard to this day, making Intel the industry leader.
In 2003, AMD launched the industry's first 64-bit processor, Athlon 64, and also brought x86-64, a 64-bit extended superset of the x86 instruction set. The battle between Intel and AMD began from then on.
② RISC and ARM
The advent of the mobile Internet era has put higher and higher demands on low power consumption. In the entire instruction set of the x86 architecture, only about 20% of the instructions are frequently used. Therefore, in 1979, Professor David Patterson of the University of California, Berkeley, proposed the idea of RISC, advocating that hardware should focus on accelerating commonly used instructions, while more complex instructions should be combined with commonly used instructions.
RISC (Reduced Instruction Set Computer) is a reduced instruction set computer. RISC achieves power saving and high efficiency by streamlining the types and formats of CISC instructions and simplifying the addressing method. It is suitable for portable electronic products or IoT products such as mobile phones, tablets, digital cameras, etc.
In the 1980s, ARM started making its own chips based on the RISC architecture, and eventually rose step by step, defeating Intel and becoming the king of mobile chips today. Today, most mobile phone terminals and IoT device chips, including Huawei Kirin and Qualcomm Snapdragon, are designed based on the ARM architecture.
In 2007, the iPhone was launched, ushering in the era of mobile Internet. The processor chip of the first generation of iPhone was designed using the ARM architecture. In 2008, Google launched the Android system based on the ARM instruction set. Since then, the rapid development of smartphones has established ARM's dominant position in the smartphone market.
③ RISC-V and AIoT
Nowadays, with the booming development of technologies such as 5G, Internet of Things, and artificial intelligence, more and more companies are beginning to produce and manufacture terminals and modules that serve various vertical industries. In terms of architecture selection, x86 is a closed technology and ARM architecture requires high licensing fees. In this case, RISC-V was born and took the stage.
The RISC-V instruction set is very concise and flexible. Its first version contains less than 50 instructions, which can be used to implement a processor with basic functions such as fixed-point operations and privileged mode. The RISC-V architecture adopts an open source approach, and its instruction set can be used freely for any purpose, allowing anyone to design, manufacture and sell RISC-V chips and software without having to pay patent fees to any company.
Currently, the RISC-V Foundation has a total of 235 member companies, including 18 platinum members (data as of July 10, 2019). These member companies include semiconductor design and manufacturing companies, system integrators, equipment manufacturers, military enterprises, scientific research institutions, universities and other organizations, which is enough to show that the influence of RISC-V is constantly expanding.
Andes Technology is a representative company focusing on RISC-V, ranking fifth in the IP field after ARM, Synopsys, MIPS, and Cadence. Andes Technology was founded in 2005. Its chairman is Ming-jie Tsai, the chairman of MediaTek. Since its inception, the company has focused on embedded CPU IP, and it has a history of 13 years. At present, the company mainly develops low-power and high-performance CPUs. In addition to CPU IP, it also provides a complete set of solutions such as platform peripheral IP, software and hardware development tools, and ecosystems.
In order to avoid being controlled by overseas chip giants, China began to focus on RISC-V-based chip design to achieve chip independence from the source. In July 2018, the Shanghai Economic and Information Commission issued the first domestic policy to support RISC-V. In October, the China RISC-V Industry Alliance was established. In terms of products, Zhongtian Micro and Huami Technology have successively released processors based on the RISC-V instruction set.
Huangshan No. 1 (Huami), developed based on RISC-V, is the world's first artificial intelligence chip in the wearable field. On July 25, 2019, Xuantie 910 was officially released. This is the first product after the establishment of Pingtouge Semiconductor. Xuantie 910 is based on the RISC-V processor IP core. Developers can download FPGA code for free and carry out chip prototype design architecture innovation. On August 22, 2019, GigaDevice, an industry-leading semiconductor supplier, officially released the GD32V series of 32-bit general-purpose MCU products based on the RISC-V core, providing complete tool chain support from chips to program code libraries, development kits, design solutions, and continues to build a RISC-V development ecosystem. 2. Midstream: International giants lead in technology, and domestic manufacturers help chip design development
The midstream of the chip industry chain includes wafer manufacturing and packaging testing.
Depending on whether the upstream and midstream are integrated, the chip/semiconductor industry has two models:
Vertical integration model, also known as IDM, belongs to the business of enterprises that include design and manufacturing/packaging and testing. Representative companies of the IDM model are Intel, Texas Instruments (TI) and Samsung.
Vertical division of labor model: companies that adopt the division of labor model only specialize in one business. For example, NVIDIA and Huawei HiSilicon only have chip design and no manufacturing business, which is called fabless; while foundries represented by TSMC, SMIC and GlobalFoundries only do foundry manufacturing and do not involve chip design, which is called Foundry.
TSMC is the absolute leader in the global foundry industry, holding a 50% share. The development progress of TSMC's advanced process almost determines the development speed of the industry. Representative foundries in mainland China include SMIC and Huahong Semiconductor, among which SMIC ranks fifth among global wafer foundries.
(1) Wafer manufacturing
The pure-play foundry industry is highly concentrated, with the top four pure-play foundries accounting for 85% of the global market share, with TSMC alone accounting for nearly 60% of the market share. Mainland foundry companies represented by SMIC, Hua Hong Semiconductor, and Huali Microelectronics still have a large gap compared to international giants. In the chip manufacturing process, the chip process determines the advanced level of the foundry. The chip process is a parameter used to characterize the size of the integrated circuit. With the development of Moore's Law, the process has developed from 0.5um, 0.35um, 0.25um, 0.18um, 0.15um, 0.13um, 90nm, 65nm, 45nm, 32nm, 28nm, 22nm, 14nm, to the current 10nm, 7nm, 5nm. At present, 28nm is the dividing point between traditional processes and advanced processes.
Take TSMC as an example. The wafer manufacturing process is updated every few years. In recent years, the replacement cycle has shortened. TSMC has mass-produced 10nm in 2017, 7nm will be mass-produced this year, and 5nm is expected to be mass-produced in 2020. The A13 Bionic chip of iPhone11 uses TSMC's 7nm process. In addition to the upgrading of wafer manufacturing technology, its downstream packaging and testing technology has also been continuously developed.
Currently, TSMC has already trial-produced 5nm, and Samsung has said it will develop a 3nm process to compete with TSMC. The gap between mainland factories and TSMC is about 2 generations. The most advanced SMIC has just been able to mass-produce 14nm process in the first quarter of this year, and is currently working hard to conquer 12nm; as for Hua Hong Semiconductor, which ranks second, it is still far from advanced process.
The existence of SMIC is of great significance to the semiconductor industry in mainland China: making money is secondary, and the main purpose is to support the independence of high-level semiconductor manufacturing industry, and then promote the improvement of the entire design, manufacturing, packaging and testing industry chain. At the same time, it can also provide support for upstream local semiconductor equipment and material manufacturers.
It is the emergence of wafer foundries that has lowered the technical and financial barriers for new players to enter the semiconductor industry and has created many IC design companies.
(2) Packaging test
Packaging and testing is the last link of integrated circuit products, and the technology is relatively easy. Packaging and testing are two processes. Packaging is to wrap the circuit and leave contact pins outside; testing is to detect whether the performance of the chip meets the design requirements. The threshold for packaging technology is relatively low, and the domestic development foundation is relatively good, so the packaging and testing industry is catching up faster than design and manufacturing. The first Chinese semiconductor company to lead the world in all aspects is most likely to appear in the packaging and testing industry. There are three leading companies in the domestic packaging and testing field, namely Changdian Technology, Tianshui Huatian and Tongfu Microelectronics, all of which have entered the top ten in the global packaging and testing industry.
From a long-term perspective, domestic packaging and testing technology has kept pace with global advancement. With the rise of domestic upstream chip design companies and the realization of downstream supporting wafer factory construction logic, coupled with the support of national policies and industrial capital, it is highly likely that domestic packaging and testing companies will surpass Taiwanese manufacturers in all aspects.
3. Downstream
The downstream of the industry chain is mainly system integration companies, which provide software and hardware integration solutions, such as artificial intelligence solution providers. Through customized algorithms and system solutions for specific industries and specific needs, downstream companies are the direct party that empowers the real economy.
The main applications are smart driving, smart security, smart voice, smart robots, smart phones, AIoT, etc.
Digital Chip
A digital chip is a circuit that transmits and processes discrete signals and implements digital signal logic operations and manipulations. Digital chips are widely used in computers, digital control, communications, automation, and instrumentation. Digital chips include processors (CPU, GPU, baseband chips, etc.), memory (DRAM, NAND Flash, NOR Flash), and logic circuits (FPGA, etc.).
The CPUs and GPUs in our commonly used electronic devices such as PCs, Macs, and smartphones are all digital chips. With the development of AI, chips such as FPGAs and ASICs are gaining more and more attention. Similarly, the storage chips represented by the familiar memory and flash memory are all digital chips.
1. Microprocessor CPU and GPU: As general-purpose chips, it is extremely difficult for China to catch up
1. CPU (1) Two giants: Intel and ARM
CPU (Central Processing Unit) is a central processing unit, which is widely used in electronic devices and the cloud. As a general-purpose chip, CPU can complete a variety of different tasks and plays the role of the brain. Its main function is to interpret computer instructions and process data in computer software.
Intel is a giant that mainly develops CPU processors and is the world's largest manufacturer of personal computer parts and CPUs. In 1971, Intel launched the world's first microprocessor, the 4004, which was used in calculators; in 1978, it launched the 8086, which can process 16-bit data and has a frequency of 5MHz. This is the first x86 chip. IBM used the simplified version of the 8086, the 8088, in its first PC. The computer and Internet revolution brought about by Intel's CPU has changed the whole world. It can be said that the history of Intel is a brief history of the development of CPUs.
However, with the arrival of smartphones such as the iPhone and the advent of the mobile Internet, Intel failed to maintain its advantage and was gradually defeated by ARM chips in the mobile device CPU market. ARM uses the RISC reduced instruction set architecture, focusing on low-cost, low-power and high-efficiency chips, which has great advantages in mobile devices. Currently, more than 95% of smartphones and tablets in the world use the ARM architecture. It can be seen that ARM and Intel have completely opposite strategic routes. Intel has always adhered to the full industry chain business model, while ARM is an open win-win cooperation model.
(2) Major domestic players
For domestic enterprises, CPU is the chip with the greatest difficulty in catching up with the world's leading enterprises. The main domestic CPU enterprises are Loongson, Zhaoxin, Huawei Kunpeng and Feiteng.
Even though the performance of some CPUs has caught up with or even surpassed Intel, due to the lack of a complete industrial ecosystem support for domestic CPUs, domestic companies are not yet able to compete directly with market leaders. Domestic companies are mostly used in the fields of finance, security, military industry, aerospace, etc. at the national level, and there is no significant market in the civilian field.
① Loongson, originated from the Chinese Academy of Sciences, is the company with the largest sales volume of PC-level CPUs in China.
The latest generation of Loongson 3A4000/3B4000 processors uses 28nm technology, and the frequency has been increased from 1.5GHz of Loongson 3A3000 to 2.0GHz. The architecture has been upgraded to GS464V, and the matching chipset has also been upgraded to Loongson 7A2000, 28nm technology. The shipment volume of Loongson 3A/B3000 processors has reached more than 300,000 pieces. It is said that after switching to 14nm technology, it can reach the level of AMD's Zen series processors.
② Zhaoxin, founded in 2013, is a joint venture between VIA and the Shanghai government fund. It has obtained x86 authorization and is the backbone of the development of high-performance X86 processors in China.
Zhaoxin and Antec have cooperated to release a domestically produced, independent and controllable network security platform. They have also cooperated with Loongson and Feiteng to launch the Loongson 3A3000/3A4000, Zhaoxin C4600, and Feiteng FT1500A/2000 series.
The Zhaoxin KX-6000 and KH-30000 series released in June this year upgraded the process to 16nm process, becoming the first general-purpose CPU in China with a main frequency of 3.0GHz. It has two specifications: 4 cores and 8 cores, and also supports PCIe 4.0 and dual-channel DDR4 memory. The matching chipset has been upgraded to the KH-3000 series.
③ Huawei Kunpeng, Huawei Kunpeng 920 becomes the industry's first 64-core data center processor compatible with ARM architecture.
In terms of performance, the quad-core version is equivalent to the Core i5 6300H, and the octa-core version is equivalent to the Core i5 8300H. Manufactured using a 7nm process, the processor is based on the ARMv8 architecture, has 64 2.6GHz cores, supports 8-channel DDR4, PCIe 4.0, and CCIX interconnect chips.
In September, Huawei took the lead in deploying the Kunpeng domestic CPU ecosystem in the Shenzhen power industry, gradually replacing Intel CPUs.
④ Feiteng is a domestic CPU company that can provide high-performance products for both desktop and embedded chips, and is one of the mainstream representatives of domestic chips.
As of August 2019, we have collaborated with more than 500 software and hardware partners to develop more than 300 complete machine products in six categories and ported and optimized more than 1,000 types of software.
On August 26, 2019, China Great Wall, a leading domestic independent security company, completed the acquisition of 35% of Tianjin Feiteng’s equity and became the largest shareholder of Tianjin Feiteng.
2. GPU
GPU (Graphics Processing Unit), or graphics processor, was originally a microprocessor used to run graphics operations in personal computers, workstations, game consoles, and some mobile devices. However, due to its parallel computing capabilities, GPU is now also widely used in the field of AI chips.
The CPU architecture requires a lot of space to place storage units and control units, while the computing unit only occupies a small part, so it is extremely limited in large-scale parallel computing capabilities and is better at logical control. However, as people's demand for larger scale and faster processing speed increases, the CPU cannot meet it, so the GPU was born.
The biggest difference between GPU and CPU is that compared with CPU serial computing, GPU is parallel computing, which can use a large number of computing units to solve computing problems at the same time, effectively improving the computing speed and processing power of computer systems. For artificial intelligence, GPU just matches the deep learning algorithm that includes a large number of parallel computing, so it has become the first choice for computing acceleration hardware in the AI era. (1) GPU King - NVIDIA
Nvidia is the world's largest independent GPU supplier. Founded in 1993 by Jen-Hsun Huang and three others, Nvidia currently holds more than 70% of the global GPU market share, with GPU as its core product accounting for 84% of revenue share. The independent GPU market is dominated by two giants, Nvidia and AMD.
(2) Major domestic players in GPU
Limited by technology, talent and patents, there is a huge gap between domestic enterprise GPU manufacturers and international giants. Jingjiawei has become the only industry leader in mass production of GPUs in China.
① Jingjiawei, the leader of the domestic GPU industry and the only GPU chip design company listed on the A-share market, was established in April 2006.
The research and development is backed by the National University of Defense Technology, and we actively cooperate with domestic and foreign algorithm companies on new technologies. The first graphics processing chip JM5400 with independent intellectual property rights has been put into use.
② Xiyou Microelectronics, embedded GPU - Firefly 1, embedded GPU chip. This project fills the domestic gap and the overall technology has reached the domestic leading level.
③ Sugon, represented by Xmachine GPU servers and Sothis AI artificial intelligence platform, provides customized development products and services for financial artificial intelligence applications. From chips, boards, complete machines, platforms, and development architectures, it fully supports the artificial intelligence applications of financial institutions, improves the performance of financial servers, and controls financial risks.
2. Memory chips: large investment, high threshold, and need strong support from the state
Just as steel and oil are the "food" of the industrial age, memory chips are the most important "food" for the development of the semiconductor industry. All information in the computer, including the input raw data, computer programs, intermediate operation results and final operation results, are stored in the memory.
Based on whether the stored data is lost after power failure, semiconductor memory chips can be divided into two categories:
One type is non-volatile memory, which can store data after power failure, mainly represented by NAND Flash, commonly used in SSD (solid state drive); the other type is volatile memory, which cannot store data after power failure, mainly represented by DRAM, commonly used in computer and mobile phone memory. In addition to NAND Flash and DRAM, it also includes NorFlash, which has a relatively small capacity, generally less than 64Mb, and is used to store some algorithms and codes of driving circuits, and is used in mobile phones, automotive electronics, industrial control and other fields.
From the perspective of output value composition, DRAM, NAND Flash, and NOR Flash are the core parts of the memory industry. This is because, on the one hand, the performance of mobile phone operating systems and increasingly rich application software are greatly dependent on the capacity of mobile phone embedded flash memory; on the other hand, the emergence of new technologies such as the Internet of Everything has driven the rapid expansion of data volume.
1. DRAM
DRAM (Dynamic Random Access Memory) is the most common memory. It can only keep data for a short time. The most common application is the memory in PC. In order to keep the data, capacitors are used for storage, so it must be refreshed every once in a while. If the storage unit is not refreshed, the stored information will be lost.
From an industry perspective, early computer applications accounted for up to 90% of the entire DRAM industry. Starting in 2016, with the rise of large-capacity smartphones, smartphones gradually replaced PCs to become the mainstream of the DRAM industry. At the same time, the emergence of demand for cloud server DRAM has contributed greatly to this. Companies such as Facebook, Google, Amazon, Tencent, and Alibaba are constantly expanding their network storage systems, and the increase in demand for cloud storage and cloud computing has driven the demand for server DRAM to take off. Currently, the DRAM industry has been monopolized by the three major memory companies in the United States and South Korea. Samsung, Hynix, and Micron occupy more than 95% of the global market.
For state-owned enterprises, DRAM requires huge industry investment and high barriers to entry, and requires strong support from the state. Currently, the three major domestic memory chip projects, Hefei Changxin, Fujian Jinhua and Yangtze Memory, have become the hope for domestic substitution of memory chips.
2. NAND Flash
NAND Flash is the most important type of Flash memory. NAND Flash memory has the advantages of large capacity and fast rewrite speed. It is suitable for storing large amounts of data and is most commonly used in solid-state drives.
NAND Flash relies on storage particles to achieve storage, and the smallest unit of data is called a cell. From a technological point of view, NAND Flash can be divided into 2D process and 3D process. The traditional 2D process is similar to "a piece of paper", but the capacity of "a piece of paper" has a bottleneck. In order to meet the needs of large-capacity terminals, the four major flash memory manufacturers, Samsung, Intel, Micron, and Toshiba, have all begun to develop multi-layer flash memory (3D NAND Flash). 3DXpoint introduced by Intel and Micron is the most groundbreaking storage technology since the launch of NAND Flash. It breaks through the capacity limit of 2D NAND storage chips through single-layer memory stacking and greatly increases the memory capacity. Therefore, the technology 3D NAND has four advantages: first, it is 1,000 times faster than 2D NAND Flash; second, the cost is only half of DRAM; third, the service life is 1,000 times that of 2D NAND; fourth, the density is 10 times that of traditional storage.
In addition to traditional storage giants such as Samsung Electronics, SK Hynix, and Micron Technology, Toshiba and Western Digital are also important forces that cannot be ignored in the NAND Flash field. Like DRAM, domestic companies still need strong support from national supporting policies and funds.
3. Memory chip landscape: overseas giants continue to monopolize, and domestic companies can enter the market from niche markets Overall, DRAM and NAND Flash account for more than 96% of the memory chip market. NOR Flash has a small storage capacity and its application focuses on code storage. It has shown a trend of being replaced by NAND Flash in consumer storage applications. It is currently only used in functional mobile phones, set-top boxes, network equipment, and industrial production line control.
At the company level, as the memory industry trend dominated by DRAM and NAND Flash will continue in the future, overseas memory giants Samsung Electronics, SK Hynix, Micron Technology, Western Digital, and Toshiba will continue to control the mid-to-high-end memory market and will continue to compete in the memory industry in the future.
my country's total imports of memory chips amount to 88 billion U.S. dollars, with a foreign dependence rate of over 90%, and a self-sufficiency rate of almost zero for DRAM and NAND.
my country has begun to vigorously develop domestic memory chips. At present, my country has gradually formed three major memory projects: the NAND and DRAM projects (Yangtze River Storage) jointly carried out by Tsinghua Unigroup and Wuhan, Nanjing and Chengdu, the DRAM project (Hefei Changxin) jointly carried out by GigaDevice and Hefei, and the DRAM project (Fujian Jinhua) jointly carried out by UMC and Fujian Province.
Although the mainstream DRAM consumer market is huge, the resistance and pressure to move forward are also enormous. Domestic companies can enter the market through segmented markets, accumulate research and development and production, and then overtake others. This has become the choice of many domestic memory chip companies. For example, Dongxin Semiconductor has set its sights on the small and medium-capacity memory chip market.
The new generation of trillion-dollar blue ocean IoT devices requires a large amount of data storage and transmission, and small and medium-capacity storage chips will be more suitable for the development of the IoT. At present, the global NAND flash memory industry is in the transition period from 2D to 3D. Several major giants have focused on the competition of 3D, and storage giants will gradually abandon the small and medium-capacity storage chip market. The rapid development of the Internet of Things and smart terminals will continue to expand the demand for small and medium-capacity storage chips. The evolution of the industry landscape has created a historic development opportunity for semiconductor companies such as Dongxin that focus on small and medium-capacity storage chips.
4. Storage control chips are also a starting point for domestic startups
The structure of storage hardware such as SSD hard disk and USB flash disk usually includes PCB (including power supply circuit), NAND flash memory, main control chip, interface, etc. The main control chip is equivalent to the CPU of the hard disk and plays a vital role.
The main control chip design has mature ARM core, DDR physical layer and other IP licenses available, which greatly reduces the difficulty of research and development. Therefore, the main control chip has become a breakthrough point for domestic start-ups and is expected to break the monopoly of foreign companies.
Currently, the world's SSD controller chip companies mainly come from the United States and Taiwan. Marvell is the representative of American controllers, while Taiwan is represented by Phison and SMI.
Since 2015, domestic manufacturers have also gradually increased their investment in the storage market. Many manufacturers have chosen SSD controllers as a breakthrough point. Coupled with the support of domestic semiconductor funds for storage chips, domestic controller companies have begun to emerge.
More well-known companies include Jiang Bolong, Guoke Micro, Yixin, Hualan Microelectronics, and Zhongbo, which focuses on military industry and enterprise-level markets. In addition, there are subsidiaries established by Taiwanese manufacturers in the mainland. For example, Qunlian established Zhaoxin Technology in Hefei, and Hangzhou Lianyun Technology also has Taiwanese capital participation.
In general, there are no less than 10 companies in China that have deployed SSD main control chips, which has surpassed those in the United States and Taiwan in terms of number. However, domestic companies are still emerging forces in the field of main control chips.
【END】
Source link: weixin.qq.com | 
