Major chip manufacturers are rushing to grab 3nm production capacity

Driven by Apple's new iPhone shipments, coupled with the demand for Huida's new generation AI chips, and the blessing of self-developed chips by Microsoft, Amazon, Google and others, the supply chain pointed out that TSMC's leading process, huge production capacity support and increased yield are three major benefits. As a result, performance has begun to recover, and 3-nanometer foundry production capacity is expected to reach 60,000 to 70,000 pieces by the end of this year, and the annual revenue share is expected to exceed 5%, and there is a chance of reaching 10% next year.

According to legal analysis, with Huida, Qualcomm, MediaTek and other manufacturers competing to introduce it, the 3nm era will be entered in the second half of 2024. It is estimated that TSMC's monthly production capacity will reach 100,000 pieces by the end of 2024, establishing a long-term mainstream process. direction.

TSMC has received orders for artificial intelligence chips from major cloud service providers, including Huida, Google, and AWS, as well as Microsoft's latest 5-nanometer self-developed chip MAIA. The legal person pointed out that while TSMC has secured orders from many mobile phone manufacturers such as Apple, it has further obtained orders for AI chips and boosted the capacity utilization rate of advanced processes.

According to legal estimates, advanced process quotations are high, with 3-nanometer wafer foundry quotations approaching US$20,000. Both volume and price have increased, simultaneously boosting TSMC's revenue growth, and more manufacturers are wafer-producing, giving TSMC the opportunity to improve yield rates. , the climbing slope is expected to be steeper than originally expected.

Judging from TSMC's consolidated revenue in October, annual growth has resumed again after seven months, and the monthly growth rate has also increased significantly and hit a new high. The company's optimistic scenario predicts that TSMC will reverse the unfavorable situation of revenue declining for many consecutive months with its outstanding performance in the last three months.

CSP players are competing for a once-in-a-lifetime inflection point in computing. Over the past five years, the number of parameters for large language models has increased tenfold every year; therefore, CSP players need cost-effective and scalable artificial intelligence infrastructure. The legal person pointed out that self-developed chips are in the nascent stage. Microsoft's MAIA second generation, Google TPUv6, and new AWS products are all under development. The volume and demand will only increase in the future, which will be a great supplement to TSMC's advanced processes. .

Among them, Amazon AWS not only has an increasing forecast of Inf2/Trn 1 shipments, but also the customer satisfaction rate of AWS cloud services is still low; Google has even scheduled mass production of TPU v4 (7nm) in 2023, and TPU v5 (5nm) in 2024. ) mass production, and TPU v6 (3nm) mass production in 2025. Several major customers rely on TSMC's production capacity. Whether it is advanced processes or advanced packaging, next year will be a healthy year for TSMC.

As more technology companies opt for self-designed artificial intelligence chips, Samsung Electronics Co. plans to use its advanced nano-silicon process technology to increase artificial intelligence chip manufacturing sales to about 50% of its total manufacturing sales within five years. .

Samsung Electronics' semiconductor manufacturing business Samsung Fab recently set a target to increase wafer sales from 19% of high-performance computing (HPC) chip orders in 2028 and 14% of automotive chip orders in 2023 from 2023. 11% increased to 14%.

Over the same period, it will reduce mobile chip fab sales to a low of 30% from an estimated 54% this year.

With the diversification of chip wafer orders, the South Korean chip giant is expected to improve profitability through high value-added chip production.

Both HPC and automotive chips are considered artificial intelligence chips. With the confirmation of chip foundry orders, the Korean chip industry is increasing profitability through high value-added chip production. Both HPC and automotive chips are considered AI chips.

In order to diversify its foundry orders, the company plans to double the number of non-Samsung customers during the same period.

With mobile chip foundry orders accounting for more than half of its total foundry sales, Samsung Electronics relies heavily on its in-house chip developers and affiliates such as Samsung System LSI, which develops Samsung application processor Exynos and image sensor ISOCELL for smartphones .

Samsung Electronics is confident in its foundry business due to the recent increase in foundry orders for artificial intelligence chips, such as graphics processing units (GPUs) and central processing units (CPU) centers for artificial intelligence servers and data.

There are rumors that the Korean chip giant has acquired a global HPC giant as a new foundry customer.

According to foreign media reports, CPU giant Advanced Micro Devices (AMD) is considering entrusting Samsung to produce its next-generation chips at the 4-nanometer node.

Some analysts predict that the South Korean chipmaker may win orders as Samsung and AMD expanded their strategic partnership in April to develop next-generation AP chips.

Samsung, which still lags far behind global leader TSMC in foundry orders, is also said to have improved its 4-nanometer chip production yield to TSMC's level.

The South Korean company is also expected to benefit from companies such as Google, Microsoft and Amazon competing to develop their own artificial intelligence chips. Because they are fabless, chip production must be outsourced to foundries such as Samsung and TSMC.

TSMC currently uses its 5-nanometer node to produce Microsoft's AI chips. Still, considering that fabless chip companies should have the upper hand in price negotiations with multiple OEMs, Samsung may still win OEM orders from the US technology giant later.

Jeong Ki-bong, vice president of foundry business at Samsung Electronics, said during the company’s third-quarter earnings call on October 31 that “(Samsung’s) reputation and capabilities in foundry are improving” and expected “annual growth to Strong.”

Thanks to a large number of orders from AI chip customers, Samsung's foundry business recorded its highest quarterly orders in history last quarter.

The key to the steady growth of Samsung's foundry orders lies in its nanochip processing technology, which is an essential technology for producing high-performance, low-power, and high-efficiency AI chips.

Samsung plans to advance its 3-nanometer and below nanometer process technology to attract more customers for artificial intelligence chips.

It has high hopes for its next-generation gate-all-around (GAA) architecture, which is expected to significantly improve chip performance and energy efficiency.

Jeong said: "GAA has received great attention from the HPC industry, and we will plan its production according to demand."

It also plans to use the 2nm process node to produce automotive and HPC chips starting in 2026, and launch 1.4nm, considered a dream technology, in 2027.

*Disclaimer: This article is original by the author. The content of the article is the personal opinion of the author. The reprinting by Semiconductor Industry Watch is only to convey a different point of view. It does not mean that Semiconductor Industry Watch agrees or supports the view. If you have any objections, please contact Semiconductor Industry Watch.

Today is the 3591st issue shared by "Semiconductor Industry Observation" with you. Welcome to pay attention.