On January 29, the Ministry of Industry and Information Technology and other seven departments jointly issued the "Implementation Opinions on Promoting the Innovation and Development of Future Industries" (hereinafter referred to as the "Implementation Opinions"), involving advanced semiconductor materials, brain-like chips, GPU chips and other new materials and core devices. In addition, innovative products such as quantum computers and future high-end equipment such as humanoid robots involved in the "Implementation Opinions" also need semiconductor products such as quantum chips and sensors.
Advanced semiconductor materials: the cornerstone of the information technology industry
The "Implementation Opinions" pointed out in the future materials section that it is necessary to develop key strategic materials such as advanced semiconductors.
Advanced semiconductor materials are a new strategic highland for the development of the global semiconductor industry and the cornerstone of the information technology industry. Materials involve every link in the semiconductor industry chain such as equipment, manufacturing, and packaging, and any small fluctuation will affect the yield and performance of the finished chip.
Under the category of "advanced semiconductor materials and new display materials" in the "Guidance Catalogue for the First Batch of Application Demonstration of Key New Materials (2024 Edition)" of the Ministry of Industry and Information Technology, it includes high-purity arsenic for compound semiconductor materials, gallium nitride single crystal substrates and epitaxial wafers, silicon carbide single crystal substrates and homogeneous epitaxial wafers, precision ceramic components for semiconductor equipment, photoresists for integrated circuits and their key raw materials and supporting reagents and other semiconductor materials, and there are clear requirements for performance indicators, such as gallium nitride epitaxial wafers of 8 inches and above, square resistance 400 etc.
Xinao Technologies' epitaxial and SOI wafers.
The progress and innovation of advanced semiconductor materials is an important driving force for the development of the semiconductor industry. Silicon wafers are the most market-demanded semiconductor materials. From the 60s of the 20th century to the present, silicon wafers have developed from 1 inch to the current mainstream 12 inches, and the purity of large silicon wafers for semiconductors has reached 11N (99.).999999999%). Wide bandgap semiconductors are known as the third breakthrough development of semiconductor materials in the world, and their wafer size and preparation technology are constantly improving, and Chinese companies have achieved mass production of 8-inch gallium nitride on silicon. Photoresist for integrated circuits is known as the "crown jewel" of the electronic chemicals industry, and chemical amplification photoresist using photoacid is currently the most commonly used EUV photoresist, and the industry has also carried out research and development of metal oxide resists for photoresists required for 2nm and below processes. New semiconductor materials, such as gallium aluminum nitride, gallium aluminum carbide, gallium phosphide, etc., have attracted much attention, and have entered commercialization in specific fields such as ultraviolet LEDs and lasers. Guo Hui, chairman of Xi'an Shengguang Silicon Research Semiconductor Technology Co., Ltd. and researcher of the School of Microelectronics of Xidian University, told the reporter of China Electronics News that to promote the development of advanced semiconductor materials, efforts need to be made from many aspects to accelerate the process of technological innovation and industrialization. First, it is necessary to strengthen basic research and technological research. Increase investment in basic research of advanced semiconductor materials, and invest in R&D from material preparation, processing, testing and other aspects to meet the needs of different fields; the second is to promote industry-university-research cooperation, and promote the process of technology transformation and industrialization; The third is to formulate support policies, transform research institutes as the main body to capable private enterprises to take the lead in research, encourage enterprises to increase R&D investment, and improve technological innovation capabilities; Fourth, strengthen personnel training. It is suggested that enterprises should increase the training of talents in related majors and improve the treatment of relevant scientific research and development personnel. Fifth, strengthen international cooperation, build a semiconductor materials research community, and promote the development and application of advanced semiconductor materials.
Brain-like chips: the future development trend of artificial intelligence
The "Implementation Opinions" pointed out in the part of innovative landmark products that for brain-computer interfaces, breakthroughs in key technologies and core devices such as brain-like chips.
Brain-like chips are chips that use circuits to simulate the neural network architecture of the human brain to achieve ultra-low power consumption and parallel data processing similar to the human brain. Von Neumann, the "father of computers", once mentioned that the biggest advantage of biological brains compared with machines is that they can use relatively outdated components to quickly make correct calculations in an uncertain environment and in a low-energy manner. Therefore, "brain-like" is an important technological trend in artificial intelligence.
For a variety of application scenarios such as medical ** and autonomous driving, brain-like chip technology is constantly maturing.
International companies are experimenting with the use of brain-computer imaging to capture EEG signals and assist them. On January 30, 2024, Elon Musk said that his brain-computer interface company NeuraLink has achieved the first experiment of quadriplegic patients** with implantable brain-computer interface devices, using the N1 chip developed by Neuralink, and at present, the transplant recipient is recovering well.
Zhejiang University United Zhijiang Laboratory released the brain-inspired computer (**Zhejiang University Official***
China's research in the field of brain-like chips is also making breakthroughs. In 2019, the team of Professor Shi Luping of Tsinghua University developed the world's first heterogeneous fusion brain-like computing chip "Tianji". The chip combines a variety of neural networks to realize the driving of unmanned bicycles and roadblock avoidance. In 2020, the United Zhijiang Laboratory of Zhejiang University successfully developed China's first brain-like computer, which contains a total of 792 Dahl second-generation brain-like chips developed by Zhejiang University, with the number of neurons comparable to the number of neurons in the mouse brain, and the power consumption is controlled at 350W to 500W. In October 2023, the "Wentian" brain-like supercomputer service independently developed by the Nanjing Institute of Intelligent Technology of the Chinese Academy of Sciences was officially launched. According to reports, the computer has the characteristics of large scale, high real-time performance, and strong flexibility. Taking the "Wentian" intelligent transportation system developed by the research institute as an example, in the face of the scale of Nanjing's transportation network, the response time of the "Wentian" intelligent transportation system is only 5 of that of other systems3% to help reduce carbon emissions.
GPU chip: the computing power base of the new intelligent computing center
The "Implementation Opinions" pointed out in the part of innovative landmark products that for ultra-large-scale intelligent computing centers, breakthroughs in GPU chips and other technologies will be accelerated.
The Intelligent Computing Center (Intelligent Computing Center) is a new type of public computing infrastructure based on the theory of artificial intelligence, using artificial intelligence computing architecture, and providing computing power services, data services and algorithm services required for artificial intelligence applications.
Originally designed for graphics rendering, GPUs (graphics processing units) have shown great potential for parallel computing in the process of technological development, and have gradually become the main processors for high-concurrency deep learning tasks since the 10s of the 21st century. GPUs are widely used in many fields, such as large model training, autonomous driving, and virtual reality, because they can significantly accelerate data processing and model training, and improve overall computing efficiency.
AMD introduces the MI300X (**AMD).
At present, GPU chips are mainly made up of companies such as NVIDIA and AMD, and a number of star companies of GPU and AI acceleration chips have also emerged in China. In terms of parallel computing capabilities, NVIDIA's H100, A100 and other products have accumulated many cloud service vendor customers with high computing power and relatively mature software development ecosystem. On the other hand, the high storage performance of the GPU also means that the data access capability is strong. The Mi300X, launched by AMD in December 2023, has 192GB of RAM and 53 TBS bandwidth. Fast memory access can help accelerate data reading and writing and model computing, and further accelerate large model training and inference on the basis of computing power.
Quantum chips: solve the problem of massive data processing
The "Implementation Opinions" mentioned that breakthroughs in high-end equipment products such as quantum computers will be made.
With its unique parallelism, superposition and entanglement, quantum computing is regarded as an important technical direction for the next generation of massive data processing. Among them, quantum chips, as the core components of quantum computing, are the key to solving the problem of massive data processing.
Schematic diagram of the comparison between quantum computers and traditional computers (compiled by China Electronic News).
It is understood that quantum chips use the principles of quantum mechanics to store, process and calculate information, and their core is qubits. Compared with traditional bits that can only store 0 or 1 states, qubits can be in the superposition state of 0 and 1 states at the same time, which enables quantum chips to achieve parallel computing and efficient information processing. At present, silicon-based qubit chips and superconducting qubit chips are the two major branches of technology that have attracted the most attention.
Silicon-based qubit chips embed a single electron into a silicon lattice to realize the fabrication of silicon-based qubits, which has the advantages of relatively low manufacturing cost and natural connection with the traditional semiconductor industry. In June 2023, Intel announced Tunnelfalls, a quantum chip containing 12 silicon spin qubits that integrate more than 24,000 quantum dots on each 300mm wafer to form 4 to 12 qubits that can be isolated from each other or manipulated simultaneously.
Superconducting qubit chips enhance the stability and controllability of qubits by introducing superconducting technology, so as to better adapt to the needs of quantum computing. Previously, IBM released the world's first modular quantum computing system, IBM Quantum System 2, and the industry's first 1000-qubit quantum chip, Condor.
China has also made phased progress in the field of quantum chips. On January 7, the Anhui Provincial Key Laboratory of Quantum Computing Chips and the Anhui Provincial Quantum Computing Engineering Research Center jointly released China's third-generation independent superconducting quantum chip - "Wukong Core" (Kuafu KFC72-300). The superconducting quantum chip has been running on China's third-generation autonomous superconducting quantum computer "Origin Wukong", which is capable of achieving characteristics such as quantum superposition and entanglement. Based on this quantum chip, the "Origin Wukong" quantum computer can deliver and execute up to 200 quantum circuit computing tasks at one time.
High-precision sensors for humanoid robots: Giving machines human-like senses
The "Implementation Opinions" proposes in the part of strengthening the future high-end equipment, facing the major national strategic needs and the needs of the people's better life, accelerating the implementation of major technical equipment research projects, and breaking through high-end equipment products such as humanoid robots.
Compared with industrial robots, humanoid robots are characterized by higher complexity and higher precision in their working environment and movement methods. In order for humanoid robots to be "more human-like", sensors applied to humanoid robots need to be more accurate on the one hand, and "complement" the human-like senses of the machine on the other. 
Xiaomi's full-size humanoid bionic robot CyberOne
There are many different types of sensors that can be installed in humanoid robots, but they can be summarized into five categories: torque sensors, IMUs (inertial measurement units), vision sensors, tactile sensors, and encoders. This type of sensor is a key product to help robots realize important components including dexterous hands, bionic perception and cognition, and electronics. In September 2023, the Ministry of Industry and Information Technology (MIIT) issued the Notice on Organizing and Carrying out the Unveiling of Future Industrial Innovators in 2023 and the Announcement of the Humanoid Robot Unveiling Task List, which focuses on three types of sensors, including: force sensors, MEMS posture sensors and tactile sensors.
The force sensor is aimed at the needs of humanoid robots to accurately obtain the tactile mechanical signals of the driving joints and limb ends, and is mainly used in key joints and measurement parts such as joints, fingers, and soles. At present, some of the leading companies on the market provide six-dimensional torque sensors to meet this demand. At present, the technology is leading in Japan and European and American countries, and the world's six-dimensional force sensor manufacturers mainly include Japanese companies Sintokogio and Wacoh-Tech, European and American ATI, AMTI, Kistler, etc.
The MEMS posture sensor is aimed at the attitude control needs of humanoid robots, which requires low power consumption, anti-vibration and jitter capabilities, smaller size, and real-time attitude solving. Currently, there are a number of robots on the market that use inertial sensors (IMUs) to achieve this function, including UCLA Artemis, Boston Dynamics' Atlas, and Xiaomi's CyberOne.
The tactile sensor is mainly to meet the ability of the dexterous hand of the humanoid robot to use tools, operate equipment, sort items, and assemble with high-precision, and the tactile sensor is configured in the dexterous palm to help the robot perceive the posture, hardness, texture and other characteristics of the operation target, and improve the intelligent operation ability of the dexterous hand. At present, flexible sensors are the most widely used type of tactile sensors in the industry. Tencent Robotics X Lab has launched its self-developed robot dexterity hand TRX-Hand and robotic arm TRX-ARM, covering the fingertips and palms with self-developed high-sensitivity flexible sensor arrays.
Author丨Zhang Xinyi, Wang Xinhao, Shen Cong, Ji Xiaoting, Editor丨Qiu Jiang Yongmei, Editor丨Maria Producer丨Zhao Chen.