Not long ago, Yole said in a research report released by it that in the past few decades (since 1965), Moore’s Law has been guiding the global semiconductor industry, and the development of advanced processes has increased performance and cost. After 2002 (130 nanometers), the industry has undergone several rounds of integration, and the scale effect has profoundly affected the development of this field. At present, advanced manufacturing is an oligopoly market, with only a few key players remaining.
(Image source: Yole)
Yole also gave a more intuitive chart of changes in its report. From this picture, it is not difficult for us to discover the fact that after entering the 21st century, the number of players dedicated to advanced manufacturing has dropped from 26 to 3. In other words, in just 20 years, the number of advanced process players has decreased by nearly 90%.
A great decline in the number of advanced process players
Moore’s Law guides advanced manufacturing processes into the 21st century. At that time, the edge of TSMC, which pioneered the foundry model, was not as dazzling as it is today. It was still an era when the IDM model was king. Advances in advanced technology can still bring huge profits to the semiconductor manufacturers at that time.
Therefore, the United States, which first realized the take-off of the semiconductor industry, and the Japanese manufacturers that benefited from the first semiconductor industry transfer dividend, occupied almost the entire chip manufacturing field.
However, from 2002 to 2006, players began to withdraw from the competition in advanced manufacturing processes, including Sanyo, Rohm, ON, Mitsubishi, Hitachi, Atmel, HLMC, and ADI who did not launch the 90nm process in the first time. It can be seen from this that more Japanese manufacturers have withdrawn from the advanced node competition during the period.
Based on the semiconductor market situation at that time, after experiencing a round of rapid growth, Japanese semiconductor manufacturers were suppressed by the United States. In 1991, the United States unilaterally claimed Japan’s breach of contract and once again forced Japan to sign the second semiconductor treaty. Semiconductor manufacturers have ushered in a lost two decades. At that time, it was DRAM that was driving the growth of advanced technology, which was also an area where Japanese semiconductors saw their strengths. Then Japanese manufacturers began to turn to the upstream semiconductor industry chain, which may be one of the reasons why they have given up on the most advanced chip manufacturing process.
Next, during the transition from 90nm to 65nm process, six semiconductor companies have withdrawn from the advanced process competition. At the same time, new players have appeared in the 65nm node competition.
Judging from the data provided by Yole, in the two years from 2006 to 2008, companies that disappeared from the 65nm process manufacturing list include Sony, Sharp, Infineon, Freescale, Cypress and AMD.
From this, we can see that in addition to Japanese manufacturers, those well-known manufacturers that have a reputation in the field of automotive chips have gradually withdrawn from the competition of advanced technology. Behind their withdrawal from the competition in advanced manufacturing processes, chip manufacturing capabilities can no longer provide them with sufficient advantages.
Take Infineon as an example. In 2008, they announced the abandonment of the 65nm manufacturing process. According to relevant media reports in 2008, Infineon’s CEO Wolfgang Ziebart said that if semiconductor companies want to survive the current wave of industry consolidation, they should shift their attention from building fabs to Build on the system, and must establish a deep-level technical cooperation relationship with the customer. He said: “The competitive advantage that semiconductor manufacturers once had has disappeared. Now every company can basically use the same process technology at the same time. In the past, this was the key to distinguishing the strength of semiconductor companies.”
In his words, we can clearly see the rise of the foundry model. From the perspective of the foundry market, after TSMC was firmly established as the world’s largest foundry in 2002, he has made rapid progress in advanced technology. On the road from 90nm to 65nm for TSMC, another important role needs to be mentioned-ASML. At that time, the 153nm lithography machine encountered a bottleneck. In 2002, Dr. Lin Benjian, the vice president of TSMC R&D and the world’s lithography technology authority, proposed the concept of immersion system. ASML seized this opportunity and decided to cooperate with TSMC. They developed it in 2003. The first prototype TWINSCAN AT:1150i was released, which successfully upgraded the 90nm process to 65nm.
And we all know that semiconductor process equipment, especially lithography equipment, has never been a small expenditure. This may have also contributed to some IDM companies moving towards the fabless mode or lite FAB. This sign has appeared during the transition from 90nm to 65nm. The representative of this is that AMD sold its chip manufacturing department in 2008, and this has become the most significant strategic change in AMD’s history.
According to related reports, AMD had suffered losses for 7 consecutive quarters, most of which were related to operating expensive chip manufacturing plants. The construction cost of those chip production plants is generally between US$3 billion and US$5 billion, and a lot of funds are needed to upgrade their technology and equipment each year. AMD’s new management believes that splitting the chip manufacturing industry is the best choice to continue to maintain its competitiveness and avoid being marginalized. At the time, AMD’s CEO Dirk Meyer explained this: “AMD has explored an innovative path that allows us to focus on innovative designs without the need to invest heavily in semiconductor production.”
Since then, IDM players have begun to get lonely in the competition of advanced technology, and the foundry model has begun to show its edge.
Speed of elimination
As IDM began to seek transformation, the race for advanced processes continued to extend downward under the guidance of Moore’s Law. During the downward development of the 65nm process, the advanced process market ushered in a brief calm. And just after entering the 45nm node, the market has undergone new changes. The number of players exiting the advanced process no longer decreases significantly, but the frequency of market elimination has begun to accelerate-each new generation of new nodes is updated, Some players are eliminated.
It can be seen from the figure at the beginning of the article that from the transition from 45nm to 22nm/20nm devices, the number of advanced process players has decreased from 14 to 7, including TI, STM and other companies.
And we all know that TI, as the world’s largest analog chip manufacturer, has extremely high gross profit margins. Why do such “not bad money” manufacturers have to abandon the pursuit of advanced process technology? Is advanced process bad? According to relevant reports in 2010, TI announced that it would abandon the R&D of 32nm and below, while on the other hand they were actively acquiring second-hand production lines. Analysts at the time said that TI does not need very advanced manufacturing processes, because the current (2010) first-class analog circuit is at the 0.18 micron level, so its acquisition of second-hand production lines is very reasonable. “For TI, the complexity of the 32nm process, immersion lithography, and repeated pattern exposure are too difficult,” said Didier Scemama, a semiconductor analyst at ABN Amro Bank in London. cost.”
So after this round of knockouts, after entering the 22nm/20nm node, most companies with advanced technology manufacturing are foundries. It can be said that after entering the 22nm/20nm node, advanced technology has become a competition among wafer foundries, and they have also become the main force in advancing advanced technology.
The next competition among foundries is a retelling that has been written countless times by us. For example, whether the transistor architecture that drives 22nm and below process progress is FinFET or FD-SOI, for example, benefiting from the rise of emerging markets, the increase in chip design manufacturers will bring opportunities for foundries, and whether the naming of new nodes is consistent. The original intention of Moore’s Law, another example is whether Moore’s Law, which has been widely discussed in the industry, has expired.
Just as the discussion surrounding the continued downward development of advanced process nodes has not stopped, some foundries have grown amidst the controversy, while some foundries have given up and continued to go down-GF and UMC The withdrawal of the advanced manufacturing process has gradually become an oligopoly market. Therefore, the competition among the remaining three players in the competition of 10nm and below technology has attracted the attention of the industry.
This has also led to some new changes. Samsung and Intel have successively opened up their foundry business in the course of competition. Samsung separated its foundry business in 2017 and formed direct competition with TSMC. . This year Intel announced that the company will become a major provider of foundry capacity, starting in the United States and Europe, and providing services to customers around the world. From this point of view, the competition of advanced technology seems to have entered a white-hot stage.
So this has formed such a situation-after IDM companies have adapted to the situation of outsourcing manufacturing, advanced process manufacturers represented by TSMC, Samsung, and Intel can almost represent the global chip manufacturing capabilities. In order to seize more opportunities, they are now building factories in Manchuria. Among them, their deployment in the United States has attracted particular attention. In addition to TSMC and Samsung who have announced that they will build new foundries in the United States, Intel’s new CEO also recently made a series of deployments for its chip manufacturing capabilities. According to its plan, Intel will invest US$20 billion in two new wafer fabs in Arizona, USA. From the public information, they may all deploy 5nm and below nodes in American fabs, and this may also improve the strength of the United States in chip manufacturing in the future.
On the other hand, benefiting from the demand for mature processes in emerging markets, we see that those foundries that have abandoned the development of the most advanced nodes and IDM manufacturers that still have chip manufacturing capabilities have become the industry’s attention again. Objects, they also constitute an important part of today’s chip manufacturing capabilities.
The darling of chip manufacturing piled up heavily
Today, when the status of semiconductors is rising day by day, especially in the case of frequent shortages of chip production capacity, manufacturers’ chip manufacturing capabilities have been valued. But both for foundries and IDM companies, improving chip manufacturing capabilities will be a significant expense.
According to statistics released by the analysis agency Semico Researc not long ago, in 2020, semiconductor capital expenditures will increase by 9.2% to $112.1 billion. This is US$14.1 billion higher than its forecast in the spring of 2020 and US$3.2 billion higher than its forecast in the fall of 2020. The report also shows the top 15 spenders from 2020 to 2021. The report stated that these companies accounted for 91% of total capital expenditures in 2020 and are expected to remain at 91% of total capital expenditures by 2021.
(Source: Semico Research Corp. and company information)
The report pointed out that both Samsung and Intel are currently choosing not to provide guidance. Therefore, Semico provided estimates for the two companies in the table above based on our expectations for equipment and construction needs. Among them, according to Semico’s estimate of TSMC’s US$26.5 billion expenditure, about 80% will be used in advanced process technologies: 3nm, 5nm and 7nm. The rest will be roughly equally divided between packaging/mask manufacturing and expertise. A small portion of the total amount will be used for the new plant to be built in Phoenix, Arizona.
For Samsung, its company’s spending in the second half of 2020 was 18.3 trillion won, an increase of 24% from 14.7 trillion won in the first half of 2020, and an increase of 33% from 13.8 trillion won in 2H19. During the year, the company’s capital expenditures increased by 46% in 2020. Semico predicts that due to the continuous increase in the number of ongoing construction projects and the continued transition to more advanced process nodes, Samsung’s expenditure level in 2021 will be comparable.
The report finally pointed out: “By 2021, capital expenditure will increase by nearly 15 billion U.S. dollars. Among them, TSMC alone can contribute more than 9 billion U.S. dollars, but there are many other companies that are expanding production capacity and migrating to new process nodes.” From here At one point, the improvement of chip manufacturing capacity requires heavy capital investment.