In the global semiconductor market, IDM's development momentum and industry influence seem to be getting weaker and weaker, while the industry status of the wafer foundry business model continues to improve. This development trend can also be seen from the development status of leading manufacturers in the industry. At present, the two major semiconductor companies with the highest market value are Nvidia, with a market value of more than 2 trillion US dollars, which is extremely popular, and the other is TSMC, which will be in 2022 and 2022. In 2023, before NVIDIA's stock price soared, TSMC's market value was the highest among semiconductor companies, once exceeding US$700 billion. It later declined, but now it has returned to more than US$700 billion.
Nvidia and TSMC are typical companies in the wafer foundry business model. One designs and one produces, and they both focus on advanced process technology. Together, they have become the most eye-catching presence in the current semiconductor industry.
In contrast, the established IDM companies, which have stabilized among the top three companies in their respective fields, have two major representatives, one is Intel and the other is Texas Instruments (TI), one makes digital logic chips and the other makes analog chips. Both companies are leaders in their respective fields. But they seem to be having a hard time in recent years, especially Texas Instruments. Whether it is market value, revenue, profits, or business development capabilities, they are all declining compared with earlier years, and layoffs are continuing.
Wafer foundry is fully suppressing IDM. It is precisely because of this that the three companies with the strongest comprehensive strength among the world's wafer foundries: TSMC, Samsung, and Intel. Two of them - Samsung and Intel - have entered the wafer foundry industry from IDM and have invested in it. The intensity is getting stronger.
01 Development strategies vary
For the three major manufacturers, TSMC, Intel and Samsung, their original wafer foundry strategies have been different, but in the past two years they have become more and more consistent, that is, investing more and more resources into the most advanced process technology. superior.
TSMC's tone has not changed, and it continues to maximize its wafer foundry business, especially in advanced processes, which is the focus of TSMC's investment. A large amount of money is poured into it every year, and when it develops to 10nm, TSMC is relatively The advantages of industry competitors (mainly Samsung) are becoming more and more obvious. In terms of mass production of 7nm and 5nm process chips, TSMC has formed a trend of crushing competitors, and has extended this advantage to 3nm.
As for Samsung, in the era of 20nm and above processes, the gap with TSMC was not as big as it is now. But by 14nm (TSMC calls it 16nm), Samsung relied on its breakthroughs in process technology to overwhelm TSMC at this node. However, this advantage did not last long. TSMC quickly caught up and made Samsung increasingly uncomfortable in the sub-10nm process field. In order to catch up with TSMC, Samsung Electronics decided to spin off its foundry business unit in 2017 in order to seek the trust of more customers, especially large industry customers. However, judging from the results, such a spin-off was not considered a success, or in other words , for a giant company like Samsung in South Korea, it is too difficult to completely spin off the foundry business.
As for Intel, in the previous CEO’s plan, the foundry business was almost ignored, and instead the main focus and resources were invested in the core product CPU, as well as various new processor products (such as mobile phone processors and AI processor), but the actual results are not ideal. In terms of CPU, AMD has quickly counterattacked in the past five years by relying on its architecture and design innovation, as well as the process advantages of its partner TSMC, and snatched a large number of products that originally belonged to Intel. CPU market share. At the same time, the importance of GPUs in the AI field continues to be highlighted, and TSMC’s process technology advantages still play a key role. On the contrary, during that period, Intel did not pay attention to the GPU market and missed the best development opportunity period. This has also led to the company's heavy investment in GPU research and development in recent years, which has always resulted in half the effort with half the effort.
Under the leadership of the new CEO, Intel has significantly adjusted its development strategy and placed the foundry business in a top priority, almost requiring all-in investment. Judging from the development in the past two years and the future, Intel's This decision is worth looking forward to. Although it is a little late, it has not been missed. The results of the development will be known in five years.
02 The status of the industry is waxing and waning
The overall development strategy will lead to corresponding results, which is clearly reflected in the foundry business of TSMC, Intel and Samsung, especially the industry rankings, which are most obvious.
Judging from the rankings in the past two years, the changes in the top three are obvious. TSMC's market share has increased to 60%, Samsung has declined significantly, and Intel has been in and out of the top ten list.
A few days ago, TrendForce released the revenue ranking list of the world's top ten wafer foundries in the fourth quarter of 2023, as shown in the figure below.
It can be seen that TSMC's market share has increased to 61.2%, a month-on-month increase, while Samsung's market share has increased to 11.3%, a month-on-month decrease. As for Intel, the company's foundry business IFS (Intel Foundry Service) appeared on the list for the first time in history in the third quarter of 2023, ranking ninth at the time. In the fourth quarter list, Intel was squeezed out of the top ten.
In the fourth quarter of 2022, the ranking situation will remain basically unchanged. In terms of market share, TSMC has a market share of 58.5% and Samsung has a market share of 15.8%. At that time, Intel had not yet appeared on the list.
In the fourth quarter of 2021, TSMC’s market share in the list was 52.1% and Samsung’s was 18.3%. At that time, Intel was unlikely to appear on the list because the company officially launched the IFS service in 2021. , everything has just begun.
Based on the market share in the above three years, TSMC has steadily improved year by year, while Samsung has done the opposite. After two years of preparation and development, Intel appeared on the list for the first time in the third quarter of 2023, but disappeared again in the fourth quarter. These reflect from one aspect the results of the development strategies of these three manufacturers' wafer foundry business. That is, TSMC has adopted a pure foundry model from the beginning to maximize customer trust while developing advanced processes to the industry's ultimate level. , can the market share be steadily increased, while Samsung is between the IDM and pure foundry models, and has failed to develop the industry's top level in terms of advanced processes. The market share is declining. Intel's appearance and disappearance on the list reflects its instability in the early stages of business development.
03 Process technology competition
Changes in industry market share depend to a large extent on the level of process technology. Whether in the past or now, TSMC's comprehensive strength is the strongest, especially in terms of transistor density and energy efficiency. The advantages accumulated in technology will be difficult to surpass in a short time. However, in recent years, Intel has been catching up quickly. After solving the 10nm process technology that has been troubled for many years (Intel lingered on the 14nm process for nearly 5 years before conquering this node), the company's process node evolution The speed has improved significantly and is closing the distance with TSMC. Under this situation, Samsung is under increasing pressure, because it was blocked by TSMC and pursued by Intel. In the future, Samsung's wafer foundry business may not have a good time.
Recently, TechInsights released a comparative report on TSMC, Intel, and Samsung process technologies, focusing on the transistor density, computing performance, and energy efficiency of advanced processes.
In terms of transistor density, TSMC’s 3nm (N3) process and its enhanced version N3E have transistor densities of 283MTx/mm² (millions of transistors per square millimeter) and 273MTx/mm², both higher than Intel 18A’s 195MTx/mm². Intel 18A uses backside power technology, which helps reduce energy consumption, but Intel has not released energy consumption data. Overall, it is still unlikely that Intel 18A will significantly surpass TSMC's 3nm performance.
Samsung led TSMC in entering the GAA architecture Nanosheet process, trying to overtake others in a corner. However, after comparing transistor density, performance, and energy consumption, within the same year, Samsung's process technology lagged behind TSMC. TSMC's transistor density was about 1.5 times more than Samsung's. In terms of the number of advanced process customers, TSMC also far exceeds Samsung.
Another important point is yield, which directly affects production costs and customer acceptance.
Since entering the 5nm process era, yield rate has been the biggest problem faced by Samsung's foundry business. Especially at the 3nm process node, Samsung took the lead in introducing a new GAA architecture transistor, which is different from the FinFET transistors used in the past. The big difference also further amplifies the yield problem.
According to Notebookcheck, currently, Samsung’s 3nm process yield rate is hovering around 50%, and there are still some problems that need to be solved. Samsung stated in 2023 that the yield rate after mass production of its 3nm process has reached more than 60%. However, it now appears that it was too optimistic at the time.
In February this year, Korean media reported that Samsung’s new 3nm process had major problems. All trial production chips were defective and the yield rate was 0%. The report pointed out that the Exynos 2500 chip using the 3nm process failed to pass quality testing due to defects, resulting in the subsequent Galaxy Watch 7 chipset being unable to be mass-produced. The report pointed out that due to the failure of trial production of Exynos 2500 chips, Samsung has postponed mass production. At present, it is unclear whether the yield problem can be solved in time.
In order to catch up with TSMC, Samsung's 3nm process technology has adopted a more radical strategy, which is mainly reflected in the GAA transistor architecture. TSMC's 3nm still uses FinFET. The switch to GAA transistors will not be until 2nm, and the radical result is that some costs will be paid in terms of yield.
At that time, Intel's 10nm had been difficult to produce. The biggest obstacle was the yield problem that had not been solved for many years. As a result, the 14nm process had to be changed again and again in order to maintain the update of its CPU. Later, after about five years of research, the 10nm process yield problem was finally solved. After that, Intel's process technology development seemed to be much smoother. Now, Intel 4 is ready for mass production, and Intel 3 is soon.