At the end of 2020, a “core-lack” accident caused the global auto industry to fail. Under the epidemic, many chip suppliers reduced production capacity or closed factories, which directly led to the squeeze of automotive chip production capacity.
Volkswagen, Ford, Toyota, Honda and other car factories have shut down in the first quarter of this year. Texas Blizzard, Renesas Electronics Fire, China Taiwan Earthquake, water and power outages, and the spread of the epidemic have made the problem of “lack of cores” worse.
Subsequently, the domino effect appeared, and the lack of production capacity spread from chips to packaging and silicon wafer raw materials, resulting in a rapid rise in chip prices. According to industry insiders, the cost of semiconductor particles rose from 14 yuan to 80 yuan, an increase of 4.7 times, and the As many as 169 industries worldwide have been hit.
According to statistics, in the automotive market alone, the global production of 2.99 million vehicles will be reduced in 2021, a 3.9% decline compared to the 77.6 million vehicles in 2020, and the economic loss will exceed US$60 billion.
However, with the impact of the “lack of core” incident, the output value of the foundry industry has hit a record high again. TrendForce data shows that in the first quarter of 2021 alone, the total output value of the world’s top ten foundries will reach US$22.75 billion, a 20% increase from 2020.
In this list, Chinese mainland manufacturers accounted for 7% of the market share, Taiwan, China took 65% of the market share, South Korea accounted for 17%, the United States accounted for 5%, and Israel accounted for 1%.
China Taiwan occupies a pivotal position in the global semiconductor industry chain, and has become a pillar industry accounting for 15% of its GDP.
Since the 1980s, Taiwan, China has seized the golden period of rapid development of semiconductor technology and made every effort to build the semiconductor industry. From IC design to wafer foundry to packaging and testing, a series of industrial clusters around chip semiconductors have emerged. Among them, the foundry industry driven by high barriers has the highest concentration, and its wafer foundry and packaging and testing fields are globally The market occupies the first place, and Jingyuan foundry has won 65% of the market share.
In this context, several wafer foundries are leading the pack. In addition to the familiar TSMC, UMC, and the world’s advanced ones, a foundry named Lixi is quite special.
As the only semiconductor company in Taiwan that has survived after delisting and debt, RSMC can become the third largest wafer foundry company in Taiwan and the sixth in the world. It also proves that opportunities in the semiconductor industry are everywhere. For today’s increasingly prosperous semiconductor market, what enlightenment does NSMC, known as “small TSMC”, have to the chip semiconductor industry in mainland China?
“Trinity” and taking advantage of the situation
In the 1970s, Taiwan, China stood at a time node when the industrial structure of the United States and Japan shifted to a knowledge-intensive one. It started by doing basic low-end processing for American and Japanese manufacturers who built factories in Taiwan, and began to accumulate knowledge and technology.
In the initial stage, Taiwan, China has developed correspondingly in the four links of semiconductor design, manufacturing, testing and packaging, but the implementation of the “Trinity of Industry, Government and Education” model has found its own unique development model for it.
In 1974, Taiwan, China followed the US model of industry-university-research institutes to establish an electronics industry research center, with the purpose of planning technology and accelerating talent exchanges.
In 1985, through the establishment of the Silicon Valley office, Chinese Taiwanese engineers were convened while learning advanced technology. Zhang Zhongmou, who was the vice president of Texas Instruments at that time, was invited back to Taiwan for this reason. Circuit Corporation (TSMC).
This incident became a footnote for Taiwan’s semiconductor industry to absorb a large number of overseas talents, and continued to provide talent reserves for related fields at an average growth rate of 42% from 1983 to 1997. In 1996, with reference to Japan’s VLSI (Very Large Scale Integrated Circuit) program, Taiwan’s education department promoted the VLSI education reform plan and carried out industry-government cooperation.
The VLSI education reform plan helped schools cultivate talents in the IC field by providing funds, and launched many national science and technology plans to allow semiconductor applications to penetrate into various fields. Hsinchu Science Park (Bamboo Branch) was established during this period.
(Hsinchu Science Park, source: Bamboo Branch Administration)
At that time, the VLSI program included many well-known universities and research institutions such as Taiwan’s Tsinghua University, Taiwan’s Jiaotong University, and Taiwan’s Institute of Electronics Technology. Free implementation of chip pipelines will further promote the rapid development of the semiconductor industry.
A netizen who lives in Taiwan mentioned that 20 years ago, in Taiwan’s National Chiao Tsing University, next to the Hsinchu Science Park, almost all the graduates of the polytechnic institutes were employed in the park. Even business school and law school students went to the semiconductor industry at that time. On the wall of the famous Yang Jiayi next to Hsinchu Railway Station, you can still see small advertisements for the purchase and sale of Yingcai and Kelin parts.
At the same time, the investment in policy and capital market is further concentrated. At that time, Taiwan’s funding for innovative technology accounted for more than 20% of the total plan, and it was stipulated that newly established enterprises could be exempted from income tax for five consecutive years within nine years, and the annual business tax after five years did not exceed 20%.
From 1985 to 1990, a total of NT$2.4 billion entered the field of semiconductor investment. Zhang Yaowen, a professor at the School of Electrical and Information Engineering at National Taiwan University, said: “The ‘Trinity’ cooperation between industry, government and academia is the reason for Taiwan’s semiconductor success.”
During this period, Zhang Zhongmou, who used to be the third person in Texas Instruments, not only brought advanced technology and management experience to Taiwan’s semiconductor industry, but also brought in his old friend – General Electric Semiconductor President Dyke to join TSMC, and persuaded Intel President Andy Grove became its client.
After a series of efforts such as finding people and technologies, Zhang Zhongmou led the development of the semiconductor foundry industry in Taiwan with the achievement of winning orders from Intel.
The reason why TSMC was able to pull Intel’s big orders is precisely because in the late 1980s, the US semiconductor industry represented by Intel gradually shifted from the traditional IDM model to the Fabless (fabless chip supplier) model to promote the opportunity of global vertical division of labor.
The characteristics of the IDM model are that it is large and complete. It is completed independently from chip design to manufacturing, packaging and testing. However, its defects of large investment, long battle lines and slow results gradually emerged at that time. Not only did each link cost a lot, but also the corresponding Research and development costs are also rising.
Foundry (foundry) and Fabless (fabless chip supplier) bring industry collaboration through vertical integration. This division of labor not only improves efficiency, but also accelerates the progress of the entire semiconductor industry. By driving more upstream and downstream industries, Taiwan, China can give full play to its geographical advantages to strengthen information technology communication and form healthy industrial competition and cooperation.
This became more and more obvious after Taiwan’s leading semiconductor manufacturers such as TSMC, UMC, and Acer settled in. From the simple foundry model of a certain enterprise to the distribution of all links in the industry chain, to the formation of a joint production group, the “Japanese import” was finally formed. , Taiwan processing and assembly, export to the United States” semiconductor development strategy.
At the same time, by 1999, the global share of more than ten hardware industries such as notebook computers, monitors, motherboards, optical drives, and graphics cards in Taiwan, China, exceeded 30%. SMC predecessor), it was during this period that it was established and grown.
“Nine Lives Strange Cat” Huang Chongren
Before founding Lijing, Lijie Computer Co., Ltd. founded by Huang Chongren made a lot of money by developing printers and scanners. A gold mine – OEM for Apple.
At that time, almost all scanners on the market were black and white products, but Huang Chongren quickly launched a color scanner. Due to this investment in new technology, Li Jie obtained the authorization of Apple Computer in 1996 and made MacClone series compatible computers for it. After that, Li Jie gradually transformed into a computer manufacturer and successfully went public. Huang Chongren became rich overnight. During this period In addition to getting acquainted with Jobs, Huang Chongren also made adequate preparations for the establishment of Lijing.
In 1994, Powerchip Semiconductor, founded by Huang Chongren, entered the industry with DRAM (dynamic random access memory, currently the most common system memory). At the golden age of the DRAM industry, relying on the technical authorization obtained from Mitsubishi Electric, Powerchip began to build a DRAM production line, and successively built an 8-inch wafer fab and two 12-inch wafer fabs, and purchased Macronix. 12-inch fab, and reached an AG-AND flash memory technology licensing agreement with Renesas Electronics.
The technology licensing method was the mainstream model for developing DRAM in Taiwan at that time. Manufacturers who have obtained technology authorization can quickly transform technology into products and occupy the market, and cooperate with upstream and downstream industries to further develop more advanced products.
With these technical authorizations, Powerchip made a profit of NT$200 billion a year, even surpassing TSMC and UMC for a time. Huang Chongren’s wealth also rose steadily. At its peak in 2005, his worth reached NT$3.684 billion. , ranked the 28th richest in Taiwan, but at this time, the good days of DRAM foundries are coming to an end.
In 2008, with the impact of the financial crisis, DRAM prices plummeted periodically. In addition, Korean memory manufacturers represented by Samsung invested heavily in the storage market, and DRAM manufacturers in Taiwan suffered huge losses.
In December 2012, Powerchip’s share price plummeted to NT$0.29. According to relevant regulations, Powerchip was facing delisting. Huang Chongren was carrying a debt of NT$120 billion. Investors held up banners and stood in front of Huang Chongren’s house. Make money by speculating in stocks, and kill retail investors”, 280,000 shareholders cursed Powerchip for this.
This is a microcosm of many DRAM manufacturers at that time. In the face of the market slump and the invasion of rival capital, small and medium-sized manufacturers that could not bear the risk closed down one after another, and some large DRAM manufacturers began to make difficult transitions, such as Nanya Technology and Winbond at that time. For consumer Electronic products such as LCD TVs and digital set-top boxes), Promos transformed into an IC design company, and Powerchip also began to transform from DRAM to foundry.
But it was during this period that Power Crystal seized three important opportunities and successfully “resurrected” eight years later.
The first is Micron Technology’s $2.5 billion acquisition of Japanese chip maker Elpida. Since Powerchip and Elpida established Regin Electronics at that time, according to the agreement, Micron will acquire 24% of Regin Electronics shares held by Powerchip Technology for US$334 million at the same time as the acquisition of Elpida. This won the technical patent authorization of Micron’s most advanced 25nm technology memory at that time, giving it the capital to survive.
An internal engineer of Powerchip recalled: “Powerchip purchased Elpida’s ‘half set’ technology long before Micron acquired Elpida, and continued to develop it internally based on this technology.” “Poor Man’s 5nm” 3D WoW technology ready.
The second is Apple’s driver IC orders. When Powerchip went out of the market and owed a huge debt, everyone was not optimistic about Powerchip, but when Apple iPhone 4 and iPhone 5 were launched, everyone discovered that Powerchip’s products were used in its driver ICs. These orders not only made Powerchip profitable Continuing to repay the debt to the money also saved its bad reputation.
In 2014, Powerchip became one of the iPhone 6 LCD driver IC production lines. The iPhone 6 series with a total sales volume of 250 million units not only made Apple’s stock price rise rapidly, but also became the key to Powerchip’s transformation from debt of 100 billion to tens of billions of dollars. Huang Chongren mentioned: “When the bank doubted whether Powerchip had a way to switch from a DRAM company to a foundry company, Apple let them know Powerchip’s strength, because Apple is the most stringent.”
The third opportunity is the big shortage of MOSFETs in 2018. MOSFET is the most basic electronic device. With the characteristics of high frequency, voltage drive, and good breakdown resistance, it is used in many fields such as power supply, inverter, CPU and graphics card, communication, and automotive electronics. The shrewd Huang Chongren keenly captured the market changes, and let Power Semiconductor expand its production by 50,000 pieces, eating all the MOSFET production capacity, and Powerchip made a big profit again.
From 2012 to 2020, Huang Chongren paid off NT$120 billion in debt in 8 years, and became the only semiconductor company in Taiwan, China to be re-listed after delisting. Weird cat”.
In 2019, Powerchip Technology transferred its fab and related assets to Powerchip Semiconductor Manufacturing Co., Ltd. (Powerchip Electronics), and Huang Chongren completed Powerchip’s “rebirth”. When it relisted in 2020, its opening stock price soared 170% and went public. The stock price once reached NT$84 that day, and 280,000 shareholders were relieved.
Transformation: “The Poor Man’s 5 Nano”
Different from wafer foundries such as TSMC and UMC, RSMC’s transition from DRAM to wafer foundry can be described as a “halfway”, its technical strength is not outstanding, not only the factory does not have many rivals, but also the wafer process technology is not as advanced as its rivals. .
According to statistics, TSMC has 3 12-inch fabs, 4 8-inch fabs and 1 6-inch fab in Taiwan, 1 12-inch fab in Nanjing and 2 8-inch fabs in the United States. The factory has a monthly production capacity of 1 million wafers. UMC has 12 fabs with a monthly production capacity of 750,000 wafers, while NSMC has only 2 8-inch and 3 12-inch wafer fabs with a monthly production capacity of 200,000 wafers. Very different from the first two.
From the point of view of the process, LiSMC is also not inferior to TSMC or even UMC. The former’s advanced process technology will target 14nm, 10nm, 7nm, 5nm or even more advanced, while UMC focuses on 25nm for its wafer foundry, and has formed an alliance with Samsung to become its OLED driver foundry, while the force SMC is concentrated between 90 and 40 nanometers. Under this situation, how can Power Semiconductor achieve rapid growth and sit on the throne of Taiwan’s third semiconductor industry?
Huang Chongren once mentioned his view of “anti-Moore’s Law” in public. In his opinion, as the process in the chip field becomes more and more advanced, chip design will become more and more difficult. From the perspective of input and output, can huge amounts of money be exchanged? The huge rewards are in question, and the huge risks are obvious, and such risks are likely to be fatal to companies with low tolerance.
(R&D cost and benefit relationship forecast, data from IC Insight)
Moore’s Law points out that in the past, under the condition of constant price, the number of components that can be accommodated on an integrated circuit has doubled approximately every 18 months, and the performance will also double, which means that the chip process is every 1-2 Progress will be made every year, and related industries have always followed this rule for more than 20 years since Moore’s Law was proposed in 1965.
But by 1999, Moore’s Law encountered its first crisis – leakage.
At that time, due to the increase of transistors in the wafer, the internal gate of the chip was continuously squeezed. When the gate length was less than 20 nanometers, the current of the chip was out of control, and the current of the source penetrated the gate and directly hit the drain, resulting in leakage. , At this time, the heat of the chip rises sharply, and it is directly scrapped.
The leakage problem is difficult to solve, and Moore’s Law began to be questioned by some people, but after Dr. Hu Zhengming proposed the FinFET solution, the doubts gradually disappeared. In the traditional Transistor structure, the gate that controls the passage of the current can only control the on and off of the circuit on one side of the gate, which belongs to a planar structure. This design can greatly improve circuit control and reduce leakage, also known as fin structure.
The fin structure solves the leakage problem in theory, but it is extremely difficult to implement. After all, the volume of transistors is calculated in nanometers. It is conceivable that it is difficult to change the shape with such precision. Therefore, the initial investment required for more advanced processes is huge. Huang Chongren once calculated that the construction of a 12-inch fab to produce 28-nanometer products will cost about 3.6 billion US dollars, but in the 7-nanometer era, the fab investment will easily cost 20 billion. From US dollars.
A bigger risk is still to come. Since the development of new processes is not based on the stable yield of the previous generation process, scientists have always been challenging the limits of the process for semi-research and development of chips, which is also a major factor in the development of chips such as Intel and Qualcomm. When the factory announces the technology roadmap, it often develops multiple process nodes at the same time, because it is likely that after spending a lot of money, the probability of making it is still very low.
In this case, the way the industry used to evaluate chips in terms of power consumption, performance, and area has begun to change, and indicators such as price, energy consumption ratio, maturity, and yield have become the focus of some manufacturers.
In fact, in addition to the higher requirements for advanced manufacturing processes in mobile phones and computer chips that we use every day, most of the chips required for smart application scenarios may not even use the 28nm process, but there is still a broad market space behind them. This opens up another avenue for semiconductor companies that don’t pursue cutting-edge technology.
(Most of the chips required for smart scenarios do not even use the 28nm process, the data comes from Deloitte)
PSMC is one of the semiconductor companies “pursuing market share, not the most advanced technology”. Although it lags behind TSMC and UMC in the chip manufacturing process, Huang Chongren has found a breakthrough for Power Semiconductor through technical accumulation.
One is 3D WoW (3D Wafer on Wafer), a logic and DRAM wafer stacking technology. R&D in cooperation with Aipu, through the heterogeneous integration of the 55nm CPU produced by TSMC and its own 38nm DRAM by Aipu, has achieved a performance and speed that far exceeds the advanced process, which is much higher than that of NVIDIA’s 16nm processor. Compared with AMD’s 7-nanometer chip, it has 6 times more computing speed, but it is cheaper than advanced process chips, so it is also called “poor man’s 5-nanometer”.
3D WoW technology is developed based on the valuable experience of PSMC in the DRAM era. Huang Chongren once said that everyone can do DRAM and logic separately, but only Aipu and Powerchip can do it when they are stacked together. This technology will not only make chips faster, but also “a technology for the next generation”. In fact, similar technologies are also the focus of Apple, Google and other manufacturers.
The M1 chip launched by Apple last year, the core concept of the chip is the design method of stacking the DRAM memory on the mobile processor on the AP (application processor). This method not only reduces the chip area, reduces the transmission delay and heat generation, It can also improve the running speed for the “unified memory architecture”, which has become the most critical change for Apple to switch from X86 to ARM architecture.
Second, PSMC also has a weapon to reduce costs and occupy the market, which is to use the aluminum process to make chips. Compared with other Epistar foundries that use copper process to make chips, the cost of aluminum process wafers is further reduced. This is also the technology accumulated by Powerchip in the DRAM field in the past, and it has become the key for Power Semiconductor to occupy the market and improve gross profit margin. .
According to the financial report of RSMC, from 2019 to 2020, its gross profit margin has increased from 7% to 25%, which is inseparable from its investment in 3D WoW and aluminum process technology. It can be said that although PSMC cannot catch up with manufacturers such as TSMC and UMC in the field of advanced manufacturing process, it can seize market opportunities and launch products in a targeted manner, so as to obtain profits with low prices and high gross profit, which is the key to the success of PSMC. .
Relying on industrial clusters to break through the “stuck neck” difficulty
With the rapid expansion of China’s new energy vehicles and electronic consumer products, the “hurricane” mainland market has become an important part of the global semiconductor field that cannot be ignored. According to the International Semiconductor Industry Association (SEMI) data Equipment sales increased by 39% over the previous year to US$18.72 billion, ranking first in the world.
However, the serious semiconductor shortage and the impact of trade frictions on Chinese companies are gradually expanding. This extreme contradiction between supply and demand caused by industrial development and the consumer market makes people wonder what areas of mainland semiconductors still need to be strengthened, and Taiwan and South Korea semiconductors, which seized the opportunities in the 1980s, brought valuable experience.
Under different economic environment, South Korea and China Taiwan took different development paths in the semiconductor industry in the 1980s. The former relies on the national government to support large-scale enterprises such as Samsung and Hynix and concentrate resources to accelerate the development of the electronics industry; while the latter finds a breakthrough point in numerous small and medium-sized enterprises clusters. The upstream and downstream of the semiconductor industry continue to take root and become the pillar industry of this island.
At the same time, when we look from TSMC and UMC to NSMC, a semiconductor industry pattern with a clear division of labor and coexistence of industry competition and cooperation appears in front of us. The success of PSMC is the success of the development of small and medium-sized enterprises, the success of the variables generated by the accumulation of technology, and the success of internalization after the absorption of advanced technology.
From the initial cut into DRAM to the high point with the tuyere, and then to the difficult transition of debt delisting, technology and talents are the foundation for this company to turn around against the wind, using past experience to create new products to compete for the market, and to be able to use its own Good at technology to avoid head-to-head with giants, looking for their own opportunities.
Nowadays, more and more mainland semiconductor companies are booming. Not all of them can become the industry leader like TSMC, but they are likely to become the mainstay of PSMC. In the anti-Moore concept proposed by Huang Chongren, “risk sharing, profit sharing” will become the core development concept of semiconductor companies now and in the future.
This rule means that the foundry must not only seize market opportunities, but also establish a risk-sharing mechanism with the other party when choosing partners, so as to have a better ability to withstand pressure while designing and producing close to the needs of customers .
In fact, the cooperation between PSMC and MediaTek in 2020 is based on this principle. PSMC spent NT$278 billion to build a new 12-inch fab to provide capacity for MediaTek, but in fact, the cost of building a 12-inch fab At least NT$600 billion, of which half of the investment is provided by MediaTek, which purchases equipment and leases it to LSMC, and LSMC needs to give priority to providing production capacity to MediaTek.
At a time when the semiconductor process is becoming more and more advanced, and the investment amount and risk are getting bigger and bigger, such a concept has its own special significance for the industry, and this is the biggest lesson that Powerchip has learned after its 120 billion failed debt.
The semiconductor industry chain is a highly globalized industry. The Semiconductor Association of America (SIA) once pointed out in a report that the chip production process of a typical American semiconductor company involves five countries: Japan, the United States, Malaysia, Singapore and China. Now, after Biden took office, the U.S. government has repeatedly warned about its semiconductor industry’s over-reliance on other regions, and believes that the United States should make every effort to develop local semiconductor products. In this case, the global semiconductor pattern will also change.
(Data from Deloitte)
For the mainland semiconductor industry, this is both a crisis and an opportunity. In the past, due to the long-term use of other people’s technical solutions and paths without internalizing their own technology, China’s semiconductor industry has never been able to establish a large enough influence. But today, China has become the most concentrated consumer market in the semiconductor industry. In 2020, China’s semiconductor manufacturing capacity will account for 15% of the world’s total. The expansion of the market requires us to continuously digest, absorb and innovate technology, which not only requires enterprises , the government, but also the consensus reached by academia, research institutions and the whole society. Time is running out, and we must catch up.
Zhang Zhongmou also mentioned in a speech in April this year that the wafer manufacturing in mainland China is more than 5 years behind SMC, and the IC design lags behind the United States or Taiwan by about 1 to 2 years. Far”.
Today, the semiconductor industry, which is catching up with the whole country, can form an industrial cluster. We cannot be separated from the development of the world, but at least we can establish our own advantages.