Mo Dakang: A preliminary study on China's memory dream

China's launch of memory chip manufacturing has caused a global response. I am afraid that the veil will be lifted and the "true face" will be revealed in 2019 and beyond. It is a milestone for China's semiconductor industry. In essence, it is to lay a solid foundation for achieving the goal of "independent and controllable" industry, so "the spirit can only be encouraged, not discouraged". Although the difficulties are still great, we must take it seriously, attach importance to the protection of intellectual property rights, and work hard to strengthen the process of research and development.

Three Chinese companies have already launched a charge into memory chip manufacturing, namely, Wuhan Yangtze Memory's 32-layer 3D NAND flash memory; Fujian Jinhua's 32-nanometer DRAM niche products; and Hefei Changxin (Rui Li)'s 19-nanometer DRAM. All three companies claim that they will achieve trial production and open production lines by the end of 2018.

If we also take into account that Tsinghua Unigroup has just announced the construction of two more memory bases in Nanjing and Chengdu, the total number of companies is five.

A difficult decision to make

When China's semiconductor industry was about to start memory chip manufacturing, most people were cautious. It was not because they underestimated themselves, but because the competition in the memory industry was too fierce.

What are the difficulties in the memory industry? There may be several main aspects:

No newcomers

Since the 1990s, there has not been a single "new entrant" in the global memory manufacturer industry. During this period, Qimonda went bankrupt and Micron merged with Elpida, resulting in only three companies in the DRAM field in the world, including Samsung, Hynix and Micron (the several companies in Taiwan account for 5% in total, which can be ignored), and only four consortiums in NAND flash memory, including Samsung, Toshiba and Western Digital, Hynix, Micron and Intel. Among them, Samsung has a monopoly position. In 2017, its DRAM accounted for 45.8% of the world's total and NAND accounted for 37%.

Cycle ups and downs

The basic "rule" of the memory industry is to make a profit in one year and lose money in two years, but Samsung is an exception. It dominates the world and is good at making reverse investments. According to Gartner data, global memory grew by 64.3% in 2017 to approximately US$120 billion, and grew by 13.7% in 2018, and fell by 12.9% in 2019, and fell by another 10.2% in 2020.

The investment is large, and the competition is about production capacity and cost

Due to the special nature of memory products, their design is relatively simple, so the line width, production capacity, yield rate and depreciation of the product have become the biggest cost items. Any newcomer can hardly compete with Samsung and others due to factors such as production capacity ramp-up and depreciation, so even if they are willing to invest heavily, it is probably difficult to win, and there are also issues such as patents.

China's semiconductor industry is facing a difficult choice. The realistic solution is to choose between the processor (CPU) and memory. As we all know, the processor has been invested for nearly 20 years. The result of Loongson is that there are achievements, but it is difficult to promote and apply. Therefore, the only choice is memory, which is what everyone expects. It's just that it feels too difficult, and most people hesitated at the beginning. Now "the die is cast", we can only work together and strive to move forward.

Difficulties in 2019 and beyond

China may face three major challenges in launching memory manufacturing: 1) breaking through technical barriers; 2) competing on costs and prices; and 3) patent disputes.

From the analysis of the situation, it may not be a problem for Chinese memory manufacturers to break through technical barriers and successfully conduct trial production. Obviously, the investment pressure in 2018 will increase compared with 2017.

The second hurdle is expected to be the most difficult, which is the beginning of the ramp-up of production capacity and the stage of competing on product costs and prices. The two are linked together. When the cost difference is large, the rate of capacity ramp-up will definitely slow down, and it is difficult to immediately expand the capacity to 50,000-100,000 pieces. Because compared with our competitors, our production capacity is only 5,000 to 10,000 pieces when the line is put into production, while our competitors have already exceeded 100,000 pieces. Its yield rate is nearly 90%, while ours may be 70-80%. Samsung has already mass-produced 64-layer 3D NAND, while we may still be at 32 layers. Its depreciation is 30%, or less, while ours may be greater than 50%, and their line width size is small. Each 12-inch silicon wafer may have 900 dies, while ours only has 800, or less. So there is no doubt that the cost difference is very obvious. It depends on how long our company can tolerate losses in terms of funds.

Therefore, the most difficult time for China's memory industry should be in 2019 or later.

The third difficulty is patent disputes. Recently, there have been rumors that China cannot avoid crossing the patent "red line" when making DRAM. It is unpredictable how the opponent will fight with you. This is the "price" that China's semiconductor industry must pay for its growth. Therefore, we must prepare patent lawyers and materials from now on to prepare for the battle.

The Chinese semiconductor industry must pay enough attention to intellectual property protection, which is also the only way to move towards globalization.

Because our competitors are watching us closely, they usually use two weapons. One is the "patent stick". Their goal is to completely defeat us, and then even if they can't defeat us, they will drag us down. The other more fierce one is "price war", which makes our products become inventory and cannot be sold. So this battle is very difficult. We must make plans in advance for their irrational moves. Obviously, in addition to being able to sustain funds, we must make full use of the advantages of the Chinese market.

Recently, Samsung, Micron, Hynix, Intel and Toshiba have all begun to expand their production capacity, which is not a good sign. It is said that one of their purposes is to cope with the rise of China's memory industry.

Thinking beyond memory

The risk of conquering memory this time is very high, and success or failure is still unpredictable. However, from the perspective of China's semiconductor industry, since the arrow has been shot and the momentum is imminent, we must work hard to achieve it. It will have a far-reaching impact on China's semiconductor industry.

An attempt at IDM mode

Why China must get involved in the IDM model is closely related to the goal of achieving the industry's "independent control". Since China's semiconductor industry is in a unique position, and at this stage its chip manufacturing mostly adopts the foundry model, lacks its own products, and the only fabless is very biased, focusing on areas such as mobile phone processors. Therefore, to achieve the industry's "independent control" goal, it must quickly enter the IDM model and solve some of the most critical products that affect its own needs. Therefore, memory chips are first included in the candidate list.

China previously had its own IDMs, such as Hangzhou Silan Microelectronics, but due to their relative weakness and lack of technological advancement, they cannot represent the level of China's chip manufacturing industry.

The IDM model has its own characteristics and certain difficulties, otherwise the Chinese semiconductor industry could have entered it long ago. Its difficulties are more closely related to marketization. Because IDM products must be able to meet market demand, and foundry only provides process conditions for customers to choose processing. And because IDM has its own products, it has inventory risks, and it has to continuously compete with competitors. Many well-known manufacturers in the world are almost all IDM models, such as Intel for processors (CPU), Samsung for DRAM and NAND, NXP for automotive electronics, TI for analog products, etc.

Therefore, this foray into memory manufacturing using the IDM model is a new beginning for China's semiconductor industry and is of milestone significance.

Localization

The concept of localization is very important because the West still uses embargoes against China. From the perspective of national security, China's semiconductor industry must have some key IC products that can replace imports. Even if only 5%-10% of the products are produced in China, the West must buy them. In this way, the two sides can rely on each other and adjust the balance.

The so-called "localization rate" often used in many articles, such as 40% in 2020 and 70% in 2025, may need to be rethought. It lacks a correct definition, that is, the components of its numerator and denominator may be vague. This leads to a large difference in the data of the so-called localization rate abroad and at home. If it is assumed that the "localization rate" has reached 70% by 2025, what practical significance does it have?

Memory products have broad prospects for improving the application of domestic IC replacement.

Some thoughts on Tsinghua Unigroup's construction of three memory bases

Tsinghua Unigroup is willing to take the responsibility for the development of China's memory industry, and its chairman Zhao Weiguo is worthy of respect for his entrepreneurial responsibility. From the perspective of the development of China's memory industry, Tsinghua Unigroup is needed, because its mechanism, which is dominated by state-owned funds, cannot adapt to the fast-changing market characteristics of memory. Otherwise, Wuhan Xinxin would not have invited Tsinghua Unigroup to invest and make it a major shareholder.

Looking from another perspective, China's largest chip manufacturer, SMIC, has accumulated sales of only US$3 billion after 16 years, so it is urgent to explore a new model. Having terminal companies drive the manufacturing industry may be one of the methods, but many terminal companies have tried to make breakthroughs before, but the results were unsuccessful, reflecting that chip manufacturing is difficult and has its own unique rules.

Tsinghua Unigroup now claims to adopt methods that have never been used by predecessors (semiconductor industry), including using capital operation to accumulate development funds and mergers to promote industrial progress. Recently, it also claims to strengthen research and development. Although the industry is somewhat skeptical, observing Tsinghua Unigroup's practices in the past three years, at least there is "something in sight".

The industry is worried that the investment of up to 100 billion US dollars and the establishment of three bases at the same time, including Wuhan, Nanjing and Chengdu, seem too confident and unnecessary. Moreover, the investment amount is not necessarily proportional to the actual effect of the investment. Because under the current conditions in China, "the memory industry is not willing to spend a lot of money to succeed", and learning from Samsung's experience may not be applicable to China, let alone where the "money" is? Tsinghua Unigroup can raise some funds by relying on its reputation, and in the future it will mainly rely on performance, which is the normal rule of industrial development.

The industry once questioned whether Tsinghua Unigroup, as a company, could take on the responsibility of developing the memory industry and last long. My initial understanding is that Tsinghua Unigroup is willing to explore a "new path", which is beneficial to the development of the industry and is very necessary, so we should first welcome and support it, and should not criticize or stop it. According to Zhao Weiguo, chairman of Tsinghua Unigroup, it will take about five years to gain a firm foothold, and according to my observation, even if it takes another 2-3 years, it will be a sign of success if it can persist, and it will make a huge contribution to the development of China's memory industry.

Memory Tips

The current status of global memory, the following provides some key data, such as monthly production capacity. As of the beginning of 2017, in terms of DRAM, Samsung's monthly production capacity of 12-inch wafers is 400,000, Hynix's is 300,000, and Micron's is 330,000. For NAND flash memory, Samsung's is 400,000, Hynix's is 210,000, Micron and Intel's are 270,000, and Toshiba and Western Digital (formerly FlashDigital) are 490,000. The total global memory monthly production capacity is about 2.4 million 12-inch silicon wafers.

Samsung's Pyeongtaek plant is named Fab18 and will start mass production in Q2 2017. It will produce fourth-generation 64-layer 3D NAND flash memory. The monthly production capacity in the first phase is 40,000-50,000 pieces, accounting for 1/4 of the production line's designed capacity of 200,000 pieces. The investment amount is US$2.72-3.17 billion.

Currently, Samsung's Xi'an plant mass-produces 64-layer 3D NAND flash memory. Each 12-inch silicon wafer has approximately 780 256GB dies. When the average yield reaches 85%, the cost is estimated to be US$3 per die, which is equivalent to the price of 16Gb capacity of the mainstream 2D ​​NAND process.

For 20nm DDR4 8Gb, each 12-inch silicon wafer has about 950-1100 dies. When the yield is 85%, the cost of each 12-inch wafer is US$1,450. After taking into account the packaging and testing costs, the cost of each die is US$1.79-2.24.

Therefore, in the future, whether it is 3D NAND or DRAM, the competition will be about the cost of each die, and it is obvious that the cost pressure is very high.

Conclusion

China's launch of memory chip manufacturing has caused a global response. I am afraid that the veil will be lifted and the "true face" will be revealed in 2019 and beyond. It is a milestone for China's semiconductor industry. In essence, it is to lay a solid foundation for achieving the goal of "independent and controllable" industry, so "the spirit can only be encouraged, not discouraged". Although the difficulties are still great, we must take it seriously, attach importance to the protection of intellectual property rights, and work hard to strengthen the process of research and development.

Some people believe that China's national strength is more than six times that of South Korea, so China will definitely win the game if it consumes national strength. This is correct. However, it may not be realistic for many memory companies to rely on national funds to make up for losses, because no one can predict what the final result will be.

In addition to the factors of technology, talent and capital, there are two key internal and external issues that affect the development of China's chip manufacturing industry. One is the obstruction of the West, which uses methods such as controlling the outflow of cutting-edge talents and preventing international mergers, and occasionally using the Wassenaar Treaty to interfere. The other is the improvement of the overall industrial environment, that is, to solve the "structural" contradictions in the development of many industries. Since one of them is not in our own hands, and the other involves the overall process of national reform, the Chinese semiconductor industry will not have a smooth road ahead in the memory industry, and we really need the determination and courage to fight for a long time.

-Mo Dakang

January 22, 2018

Related Reading:

How did China's telecommunications industry achieve a turnaround from being a laggard to leading the world?

7nm→5nm→3nm, can Morris Chang really leave with peace of mind?

Smart speaker giants are fighting each other, who can become China's Amazon?