Warum haben die 2-nm-Chips im Jahr 2025 alle "verspätet" erscheinen lassen?

All flagship mobile phones in 2025 are not equipped with 2 - nm chips.

The A19 and A19 Pro - series chips of the iPhone 17 use TSMC's N3P process. The upcoming MediaTek Dimensity 9500 and the fifth - generation Qualcomm Snapdragon 8 Extreme Edition will also use this process.

Unexpectedly, MediaTek has brought a surprise.

Recently, MediaTek officially announced that the 2 - nm chip (Dimensity 9600) has reached the design and tape - out stage, making it one of the first companies to use this technology. Mass production is expected to start at the end of next year. Remarkably, they announced this progress a year before mass production. This lead is truly remarkable.

If we look at the release timings of new products from major mobile phone manufacturers, by the end of 2026, in addition to the 2 - nm Dimensity 9600, the Apple A20 series, the sixth - generation Qualcomm Snapdragon 8 Extreme Edition, and the Samsung Exynos 2600 will also use the 2 - nm process.

It is certain that the "fierce battle" over 2 - nm technology will fully commence in 2026. TSMC and Samsung have been promoting 2 - nm technology for years. Why isn't the A19 chip of the Apple iPhone 17 equipped with 2 - nm technology? And why didn't the "2 - nm battle" take place this year?

01 "No one could have dreamed that the demand for 2 - nm chips would be greater than that for 3 - nm chips"

During the earnings report on October 17, 2024, TSMC CEO Wei Zhejia talked about the demand for 2 - nm chips and used two phrases: "Very high" and "No one could have dreamed that the demand would be greater than that for 3 - nm chips."

Here is a question: TSMC only started accepting orders for 2 - nm chips on April 1 this year and began mass production in the second half of the year. Why could Wei Zhejia predict the demand for 2 - nm chips as early as October 2024?

"TSMC has a very high - quality market research team that can summarize the demand from all industries worldwide, including the demand from Nvidia, Tesla, AMD, etc.," said former TSMC factory builder Wu Zihao. "It takes about four years to build a manufacturing facility. This includes capacity planning. Fabless companies like Apple and Nvidia naturally have to submit order forecasts."

Wu Zihao also revealed that from the perspective of fabless companies, not only does tape - out need to be planned in advance, but research and development also need to be adapted to the platforms and technologies of wafer fabs. These are also sources of demand information.

In addition, the capacity forecast clause in wafer manufacturing agreements requires customers to provide wafer fabs with a reasonable order forecast to assist factory operators in capacity planning. This also partially reflects the demand situation of fabless companies.

According to data from TrendForce, Apple, AMD, Nvidia, MediaTek, etc. have already reserved capacities for TSMC's 2 - nm technology. Most of them are among TSMC's top - ten customers. Apple even contributed 25.18% of the revenue in 2024, making it TSMC's largest customer.

Among the above - mentioned customers, MediaTek has already announced the mass - production time. According to the release plans of mobile phone manufacturers, it is quite certain that Apple will be the first to gain access to TSMC's 2 - nm capacities. AMD announced in April, when TSMC released the capacities, that it will introduce the 2 - nm process in the next generations of EPYC data - center processors codenamed "Venice".

For Nvidia, Rubin is already being manufactured using 3 - nm technology. Rubin Ultra consists of four GPU dies integrated in a single package. The package size cannot be further increased, so the 2 - nm process will also be introduced.

An insider revealed that Bitmain is also a customer of TSMC's 2 - nm process and may be the world's first fabless company to use TSMC's 2 - nm technology. "ASICs for mining devices are relatively easy to manufacture. Introducing new technologies can serve as practice. Bitmain might try to start shipping in the second half of the year."

Compared with TSMC, there is less information about the customers of Samsung's 2 - nm technology. Apart from its own Exynos 2600 getting the label of "the world's first 2 - nm chip", there are rumors that Qualcomm might return to Samsung for the 2 - nm technology.

The customers' demand is unprecedentedly high. Essentially, this is due to the performance improvement achieved by transitioning from 3 - nm to 2 - nm technology.

TSMC announced the parameters of the N2 node early on: Compared with the first N3E, the transistor density increases by 15%, the performance at the same power consumption increases by 10 - 15%, and the power consumption at the same performance decreases by 25 - 30%.

MediaTek's press release about the "lead" in 2 - nm technology essentially confirms the accuracy of these data. MediaTek stated that the improved 2 - nm process technology from TSMC increases the logical density by 1.2 times, boosts the performance at the same power consumption by up to 18%, and reduces the power consumption at the same speed by about 36% compared with the existing N3E process.

In summary, the performance improvement brought by 2 - nm technology has prompted major fabless companies to adopt this technology. However, the mass production of major manufacturers will mainly start in 2026.

02 TSMC has fallen short

The reason why flagship mobile phones in 2025 cannot be equipped with 2 - nm chips is that TSMC has fallen short.

According to TSMC's plan, the 2 - nm capacity should be opened in mid - 2025. Currently, everything is going as planned, but mobile phone manufacturers have too little time to start mass - producing 2 - nm chips in 2025.

"Tape - out and wafer return take several months. After the return, the functions and performance still need to be tested. Usually, performance tuning also takes several months," said a chip designer.

This means that even if a major customer like Apple completed the tape - out and testing of the A20 chip by the end of 2024, it can only start mass production in June this year. This does not fit into the inventory plan for the iPhone 17, as Foxconn's assembly lines also need to be in operation.

Yield is another factor why mobile phone manufacturers are not betting on 2 - nm technology this year. However, this factor has less impact than the mass - production schedule. The sensitivity of different fabless companies also varies.

With the 3 - nm technology, the initial yield was only about 60%. Later, N3E and N3P gradually increased to over 80%. A similar process will also occur with the 2 - nm technology.

"The yield when introducing 2 - nm products may already be over 70% and will gradually increase. It will reach 80% next year," said the above - mentioned insider.

The initial yield is low and the price is relatively high. Customers who are sensitive to price will plan mass production after the yield has increased and adopt the "wafer - purchase" model. Otherwise, the more products they produce, the more losses they will incur. However, price is not an absolute obstacle.

Take Apple as an example. It has signed a "finished - product - purchase" agreement with TSMC and only pays for functioning chips. As long as the yield is not extremely low, price will not be a decisive factor. Regarding this question, Guo Mingji, an analyst from Tianfeng Securities, has a different opinion. He believes that although Apple buys finished chips, the costs of defective chips are already included in the purchase costs.

"The best proof is that the costs of new processors for new iPhone models have increased significantly every year. This also applies to the A17 this year," said Guo Mingji.

03 The battle for wafer manufacturing

Overseas wafer fabs are currently all working on the mass production of 2 - nm chips. However, there are slight differences in technology nomenclature, such as N2, 20A, SF2, 2 - nm, etc. However, all of them have "mysteriously" adopted the new GAA transistor architecture and have unanimously planned the back - side power supply technology in subsequent iterations.

In addition, there is an important point: The back - side power supply technology can separate the power supply lines and signal lines and move them to the back of the integrated circuit. This reduces resistance, increases transistor density, and improves performance.

In terms of the mass - production schedule, Samsung, TSMC, etc. generally adhere to their plans. In contrast, Intel, which was originally the most enthusiastic, planned to release 2 - nm capacities by the end of 2024. However, due to technological challenges, management changes, and other factors, it ultimately canceled the 2 - nm process (20A) and stopped the development of 18A (1.8 - nm). Instead, it is now focusing on the 14A (1.4 - nm) technology.

In terms of specific capacities, TrendForce revealed that TSMC is expected to fully utilize four 2 - nm wafer fabs next year, and the total monthly capacity will reach 60,000 wafers.

The above - mentioned insider said: "Fab 20 in the Hsinchu Science Park has a monthly capacity of at least 60,000 wafers. Fab 22 in Kaohsiung is expected to have a monthly capacity of 30,000 wafers. Next year, the monthly 2 - nm capacity will be at least 90,000 - 120,000 wafers."

In terms of Samsung's capacity, TrendForce cited data from SEDaily in April and said that the monthly capacity of 2 - nm wafers is 7,000 wafers.

2025 is a key year for opening 2 - nm capacities. However, this battle for wafer manufacturing can be traced back several years.

In October 2021, Samsung started the research and development of 2 - nm technology at its annual Foundry Conference and announced the corresponding schedule and technological route. TSMC was even earlier. In June 2019, it informed the public about the start of research and development. It revealed at the Global Technology Forum that it will establish a new research and development line for 2 - nm technology and deploy more than 8,000 engineers.

Overall, the research and development of 2 - nm technology at leading wafer fabs takes between 4 and 6 years. During this phase, the annual research and development costs of wafer fabs usually exceed 1 billion US dollars. TSMC even invested 3.6 billion US dollars in research and development in 2022.

The enormous research and development costs are reflected not only in technological solutions but also in the competition for research and development facilities. A typical example is the competition for the ASML High NA EUV lithography machine.

In 2022, Samsung tried to obtain advanced lithography machines by having Lee Jae - yong visit ASML. In the end, however, Intel acquired the world's first High NA EUV lithography machine at a price of nearly 400 million US dollars at the end of 2023. In 2024, Intel received another machine of the same kind.

Compared with Intel and Samsung, TSMC has been more reserved in the competition for the best facilities and was initially unimpressed by the visits of ASML managers who were trying to arrange the delivery of advanced lithography machines. In the face of competition from its rivals, Wei Zhejia also visited ASML in 2024. According to rumors, he received a "package deal" from ASML: discounts