The foundry market is "Three Kingdoms", who will stand out from TSMC, Samsung and Intel?

Recently, Samsung is negotiating with foundry customers and plans to increase foundry fees by 20% this year to cope with rising pressure on material and logistics costs. According to relevant reports, the overall price increase is between 15% and 20%, depending on the complexity of the foundry chips, and the increase in mature process chips will be even greater. Samsung has completed negotiations with some customers, but is still negotiating with others. A Samsung spokesman declined to comment. Samsung's decision is a shift from its relatively stable pricing policy last year, which withstood industry pressure and maintained a relatively stable pricing policy last year.

TSMC has previously announced plans to raise prices. The company plans to increase its foundry quotations by 6% from January 2023. The company said the price hike was mainly due to inflationary pressures, rising costs and meeting its own massive capital spending plans. Unlike last August, when the company raised its foundry prices across the board for the first time in a decade, next year, TSMC’s advanced manufacturing process and mature manufacturing process prices will increase at the same rate.

At present, TSMC and Samsung have two-thirds of the global foundry market share. However, recently, from chemicals, natural gas, to wafers to equipment and materials, the manufacturing costs of foundries have increased on average in all aspects. About 20% to 30%, so we have to further increase the price of wafer foundry.

The three pillars of the foundry market

There is already fierce competition in the foundry market. TSMC and Samsung are already "fighting", and Intel is also trying to shake up the competition by announcing its re-entry into foundry. Samsung and TSMC are currently preparing to mass-produce 3nm processes.

According to Taiwan media reports, TSMC's 3nm process will be introduced into mass production in August this year, but in order to gain a stronger initiative, TSMC decided to let the 3nm R&D team switch to 1.4nm, and is scheduled to start shooting next month and invest in confirming technical specifications. Staged development, which also prepares TSMC to step into the 1nm generation. Operating a 1.4nm R&D team that will be commercialized in the distant future, while 3nm has not yet started mass production, TSMC is consciously chasing competitors.

Samsung aims to start mass production of 3nm semiconductors in the first half of this year and to start mass production of 2nm semiconductors in 2025. By the end of 2022, Intel should increase their 4nm node, in 2023, Intel's 3nm node should increase, and by 2024 their 20A (2nm) and 18A (1.8nm) nodes should increase. All of these are EUV-based nodes, and by the end of 2024 Intel will rarely use non-EUV-based microprocessor production processes. Intel is also building an EUV-based production facility.

Can TSMC remain overwhelmingly competitive?

From the current technology point of view, according to analysis, TSMC in Taiwan is likely to maintain a strong position in the 2nm or smaller process. TSMC is a company with a motto of "not competing with customers." Its clients include semiconductor design companies such as Apple, Qualcomm, Nvidia and AMD. The fact that the foundry's largest customer, Apple, has long entrusted all of its mobile processors (APs) to TSMC is a testament to TSMC's technological prowess.

Samsung has also produced APs for Apple's iPhones in the past, but since the iPhone 6 in 2015, TSMC has become Apple's main supplier. TSMC's long-term partnership with design houses such as Apple has been built up through TSMC's know-how. In the process of semiconductor production by foundries, each node needs to be fine-tuned countless times, so long-term cooperation is conducive to improving product quality or yield.

Taiwan's semiconductor ecosystem is also cited as a factor that could give TSMC a long-term competitive advantage.

Unlike memory semiconductors, which are characterized by standardization and mass production, SoCs are usually manufactured according to customer needs. It has to go through a design house specializing in semiconductor design, a foundry that takes orders to produce semiconductors, and a packaging and testing process for the chips produced.

In Taiwan, China, there are not only large design companies such as MediaTek, but also ASE, the world's first packaging and testing factory.

Samsung, close the gap with TSMC

Samsung's foundry started as a foundry business team within the semiconductor division in 2005, but in 2017, the foundry business team was promoted to a separate division and seriously expanded.

Although Samsung is a latecomer, it is closing the gap with TSMC by focusing on advanced processes below 10nm.

In 2016, Samsung started mass production of the 10nm process for the first time in the industry, and in 2017, it introduced extreme ultraviolet (EUV) exposure equipment into the 7nm process for the first time in the industry. Due to their focus on cutting-edge processes, Samsung and TSMC are currently evenly split on processes above 10nm. As of 2021, Samsung's foundry market share is only 18%, but it is understood that Samsung will reach 40% of the foundry market above 10nm.

Samsung did not stop there, taking the lead in introducing a new transistor production technology "Gate All-Around (GAA)" at 3nm, and is implementing a strategy to defeat TSMC. GAA is a technology where the channel and gate of a transistor are in contact on four sides. The voltage at which the semiconductor operates can be reduced and performance improved compared to existing FinFET approaches that limit the channel and gate contacts to three sides.

If Samsung succeeds in GAA technology at 3nm, it is expected to be able to gain a competitive advantage over TSMC, which plans to introduce GAA at 2nm for the first time.

Intel, the traditional semiconductor powerhouse

Intel is a traditional semiconductor powerhouse, but it withdrew from the foundry business in 2018 and chose to focus on the central processing unit (CPU) of PCs. At that time, Intel's market share in foundry was too low to help profit, so it decided to focus on CPUs, which were almost monopolized at the time. Failure to overcome the technical barriers of the 7nm foundry process at that time was also one of the reasons for its withdrawal.

However, Pat Gelsinger announced a re-entry into the foundry business in 2021 with a radical change in management strategy. According to the interpretation, he decided not to miss the fast-growing foundry market. Intel is lagging behind in the generation process, however, Intel is praised for its excellent efficiency maximization, even if the line width process is the same, the product quality varies according to the way the circuit is designed, and Intel is invincible in terms of design. Even Intel's 14nm offerings are rated to have similar performance to rival 10nm offerings. "Intel's 10nm ultra-fine process will be comparable to TSMC's 7nm process, and 7nm will be similar to TSMC's 3nm process. If successful, it will be able to surpass TSMC and Samsung in performance," said Scott Jones, president of market research firm IC Knowledge. "

Another positive factor is that, with the full support of the US government, Intel is very likely to win orders from US chip designers. Mizuho Securities researcher BJ Rakesh said, "Intel's plan is in line with the US Department of Defense's security strategy of wanting to produce semiconductors in the United States."

Therefore, Intel will be able to gain 15-20% of the foundry market share. "However, there are also views that Intel's foundry road is full of difficulties and obstacles. On the one hand, technology is a major obstacle; on the other hand, Intel, which has large factories in Europe and the United States with high labor and production costs, also has difficulties in reducing production costs. .