At the recent IC China 2020 (China International Semiconductor Expo), Wu Hanming, academician of Chinese Academy of engineering and Dean of micro nano Electronics College of Zhejiang University, predicted that “with the evolution of process nodes, Moore’s law will become more and more difficult to continue, and it is expected to go to 2025.” The process technology of semiconductor enterprises is advancing towards this end point. The latest news is that TSMC will mass produce 3 nm process chips in 2022 and implement 2 nm process in 2024. Is nanometer the end of Moore’s law? Under the existing process technology, maybe. Unless there is a breakthrough in new technology and new materials. TSMC has optimistic forecast that under this premise, the process can reach 0.1nm by 2050. < / P > < p > but in any case, it will be a game for a few people, and for well-known reasons, China is temporarily lagging behind in this game. In the post Moore era, where are the opportunities for Chinese chips? < / P > < p > “the driving force of industrial development is computing power. According to openai’s estimation, AI’s computing power will double every 3.4 months, but the growth rate of computing power demand is 10 times / year. Even if Moore’s law does not slow down, it will be difficult to meet the increasing demand for computing power.” Zhao Lidong, founder and CEO of Shanghai Suiyuan Technology Co., Ltd., said on IC china2020. Although AI computing power has doubled every 3.4 months since 2012, it is still lower than the growth rate of computing power demand. Source: openai < / P > < p > Moore’s law holds that the number of transistors in a chip will double every 18-24 months, and its performance will also double. Since Moore put forward this idea in 1965, the performance of the chip has been improved by 10 times every 5 years and 100 times every 10 years. However, the manufacturing process has entered the “post Moore era” from 22nm, and Moore’s law began to slow down. Huang Renxun, founder of NVIDIA, even said on ces2019 that Moore’s law can only increase by a few percentage points every year, and the growth rate of chip performance has dropped to twice every 10 years. < / P > < p > the cost of chip design increases with different process nodes. According to the estimates of IBS, the development cost of 16 nm process is 100 million US dollars, and that of 10 nanometers is 174 million US dollars. At present, the mainstream 7 nanometers cost 300 million US dollars, 5 nanometers costs about 436 million US dollars, and 3 nanometers is more than 650 million US dollars. < p > < p > John Neuffer, President and CEO of the semiconductor industry association of America, revealed that American semiconductor companies spend about a fifth of their revenue each year to meet the challenge of rising chip design costs. Wu Hanming said that the development of China’s integrated circuit industry is difficult, especially in the chip manufacturing process. There are three challenges: lithography is the basic challenge, new materials & technology is the core challenge, and improving yield is the final challenge. The contradiction between computational power and Moore’s law is obvious. However, the current situation is that the advanced process technology is basically monopolized by TSMC and Samsung. In addition, due to the complicated international relations since this year, it can not meet the demand of computing power only relying on process upgrading. However, it is a good thing for Chinese enterprises. Wu Hanming predicted that “new materials and new processes will be the main theme of the future research and development of complete sets of processes. Many key technologies, including lithography technology and thin film technology, will be integrated into a complete set of commercial processes. < / P > < p > in 2019, 82% of the global foundry capacity of wafers are nodes above 20nm, and the advanced capacity of nodes below 20nm only accounts for 12%, which leaves a huge space for innovation for enterprises. ” In fact, with the advent of the era of high-density mobile phones, most of the products will not be replaced by the high-speed mobile phone. < / P > < p > in other words, many AI chips can achieve the same computing power as advanced process technologies through the innovation of complete sets of processes without the need for top-notch process technologies. < / P > < p > If TSMC and ASML of the Netherlands “lock” China’s direction of manufacturing high-end chips through process improvement, then Chinese chips must “find another way.”.
“global semiconductor industry has undergone three shifts, and each shift is due to changes in demand. Since 1990s, with the popularity of PC and mobile phone, the industrial chain has gradually shifted from Korea and Taiwan to mainland China. Until the era of intelligent Internet of things, China will become the mainstream of light design companies.” Dai Weimin, chairman of Xinyuan shares, said. < p > < p > data shows that China has become the largest and fastest growing IC market in the world. John Neuffer revealed that in 2019, the Chinese market accounted for 36% of the revenue of us semiconductor companies, and China’s exports of PCs, mobile phones, TVs and wearable devices ranked first in the global share. < p > < p > Yu Kai, founder of horizon, a domestic edge computing chip enterprise, believes that the future demand of automatic driving for vehicle intelligent computing power is an opportunity. New packaging technology, highly collaborative software and hardware, innovative architecture design for automatic driving artificial intelligence and compiler operating system will make AI computing power move forward with exponential “Moore’s law” Development, but this requires artificial intelligence enterprises to have a new thinking chip design. Where is the road to innovation? New structures, new materials, and new integrated design innovation are generally recognized in the industry. However, in terms of equipment and materials, Japan rose earlier and is now the market leader. China is most likely to make a breakthrough in the field of design. Dai Weimin thinks that chiplet is a feasible way. The so-called chiplet is the Mochi (modular chip) architecture proposed by Marvell founder Zhou Xiuwen. As the name implies, it is to mix die (bare chip) of different process nodes in a heterogeneous way, which can greatly reduce the cost of advanced process. < / P > < p > for example, apple A13 costs $4.98/1 billion transistors in a 16 nanometer process, but only $2.65 per billion transistors in a 7-nanometer process. According to Dai Weimin, the new form of chiplet is one of the important trends of chips in the future. For example, a chip can use 7Nm process for CPU and 14nm process for I / 0, which can reduce the cost by about 50% compared with the chip made entirely of 7Nm. < / P > < p > at present, Xinyuan puts forward IP (chiplet), which can form chips such as ISP (image processing), NPU (neural network processor), Vpu (video processor) and GPU (graphics processor) into Chiplets based on different process nodes and achieve “plug and play”. < / P > < p > in this way, for Chinese enterprises, chiplet will reduce the threshold of chip design on a large scale; semiconductor IP licensees can upgrade to chiplet suppliers, improve the value of IP and effectively reduce the design costs of chip customers. < / P > < p > more importantly, chiplet can make better use of China’s packaging resources. Although it has been “stuck” in the advanced process, China’s packaging process is basically close to the world’s top level. < / P > < p > < p > “leader” Changdian technology is now continuously promoting the mass production of chiplet and the development of 2.5d/3d packaging technology. It is estimated that by 2021, 2.5D packaging process can be realized, thus promoting the rapid landing of chiplet. In 2019, the scale of China’s integrated circuit industry will reach more than 700 billion yuan, with a year-on-year increase of 15.8%. Yang Xudong, deputy director of the electronic information department of the Ministry of industry and information technology, told IC China that China accounted for nearly 50% of the global market share. Although the United States has begun to implement the “decoupling” policy in recent years, China’s chip industry is rapidly rising. Data show that today China has 17% of the world’s chip production. John Neuffer predicts that by the end of the 20th century, this proportion will increase to 28%, especially in the field of wafer less factories and OSAT (packaging and testing substitute factory), and China has obvious advantages. However, the disadvantages of China are also obvious. In 2020, the latest survey of the semiconductor industry association of America shows that the friction between China and the United States will change the pattern of global IC industry. At present, China’s local IC self-sufficiency rate is only 14%, China’s domestic IC design industry’s global market share is 15%, and the global market share of the overall IC industry is only 5%. In contrast, the United States accounts for more than 50% of the total. According to Chen Gang, general manager of Shanghai Integrated Circuit Industry Investment Fund Management Co., Ltd., from the four aspects of materials, equipment, manufacturing / sealing and testing, and design, kechuangboard has a certain coverage for enterprises in material fields such as photomask and 12 inch polishing chip to seek secondary market capital. However, in terms of high-end materials, photolithography equipment and polishing machine required by advanced technology, it still has a certain coverage There are also gaps in EDA, CPU, FPGA, DRAM, NAND and other fields in the design field. < / P > < p > in the core semiconductor equipment market, China is far from the world. In 2019, the localization rate of China’s semiconductor equipment is 18.8%, but Wang Hui, chairman of shengmei Semiconductor Equipment Co., Ltd., admits that the proportion of core equipment is very small. According to the allocation of R & D funds in different countries, there is still a big gap between China and the advanced countries in the world. Especially in the field of basic research, most of R & D is still in the experimental stage, which is difficult to produce revolutionary innovation. Wu Hanming pointed out that China’s total investment in IC R & D is no less than that in Britain, the United States, France and Japan, but the proportion is seriously unbalanced. Only 5% of the funds are invested in basic R & D. in particular, some chip manufacturing enterprises have the trial and error cost accounting for 84% of the total R & D expenditure. “We must have a basic guidance to make the so-called trial and error R & D become experimental research and development.” Global Tech