news

the rise of ai has caused the prices of new process chips to skyrocket, but has caused a price war for mature process chips.

2024-10-05

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

​if we paint a picture of the future, does that mean we are worried about the present?

the demand for advanced wafer foundry processes is booming, and behind the upward development of the overall industry, the market situation of mature processes appears to be relatively clear. the latest report released by the research organization trendforce pointed out that although the capacity utilization rate of mature processes can increase by 10 percentage points next year, regarding this "pie", the industry said that the continued expansion of mature processes will lead to continued pressure on prices.

01

the reason why it still doesn’t work

affected by the low visibility of demand for consumer products in 2025, the supply chain has a conservative attitude towards inventory establishment. orders for wafer foundries will be the same sporadic urgent order mode as in 2024. however, the inventory of application components such as automobiles, industrial control, and general servers has been reduced. it will gradually be corrected to a healthy level in 2024 and will join the ranks of sporadic stocking in 2025. it is expected that the capacity utilization rate of mature processes will increase by 10 percentage points and exceed 70%.

however, as fabs have slowed down their production expansion plans for two consecutive years due to weak demand, they are expected to gradually open up previously deferred new production capacity in 2025, especially 28 nanometer, 40 nanometer and 55 nanometer. , under the dual pressure of low demand visibility and the opening of new production capacity, the price of mature processes continues to be under pressure.

on may 9, mainland china’s largest wafer foundry released its first quarter 2024 financial report. at the earnings conference the next day, the ceo admitted that the company's 12-inch wafer production line has been operating at full capacity since february, but competition in the industry remains fierce. "many of our strategic customers, whether they are set-top boxes or smartphones, may lose their orders if competitors cut prices in the market, and tens of millions of orders may be lost. we still have to adapt to the market and face market competition with our customers. "

companies that focus on mature process wafer foundry have their own strategies, but they can also be described as clear soup and little water.

world advancedthe 12-inch factory and compound semiconductor invested, and subscribed for 50,000 private equity shares of hanlei. the cooperation between the two parties estimates that mass production is expected in the second half of 2026.

umcit focuses on increasing its ability to receive orders for special processes. it is currently able to provide special processes such as 22/28nm, envm and rfsoi. the company is optimistic that the generative ai market has huge potential and will not be absent from the ai ​​market.

power semiconductorchairman huang chongren has previously said that it will transform in two major directions to get rid of price competition among manufacturers. the first is fab ip that charges factory building authorization fees, and the second is 3d ic technology with stacking technology advantages.

02

ai takes advanced manufacturing processes to the next level

tsmc leads the way, and the global wafer foundry output value is expected to return to an annual growth rate of more than 20% next year, and the annual growth rate will be the highest in the past three years. research institute trendforce has released its latest forecast. advanced processes such as 5/4/3 nanometers used in high-speed computing (hpc) products and flagship smartphones will remain fully loaded and will continue until 2025. tsmc’s revenue performance will surpass the industry. average, expected inaithe rise in (ai) applications will drive the growth of industrial output value.

according to trendforce's latest survey, although the visibility of the consumer end market next year will still be low, the inventory of supply chains such as automobiles and industrial controls has gradually declined since the second half of this year. sporadic stocking will be restarted next year, and edge ai will push up the single-unit inventory. with the wafer consumption of mobile phones and the continuous deployment of cloud ai, it is estimated that the output value of wafer foundry will increase by 20% annually next year, which is better than this year’s 16%.

however, according to industry observations, if tsmc is excluded, global wafer foundry output will only increase by 11.2% next year, which means that tsmc alone will contribute nearly half of the growth. advanced processes maintain high growth momentum, and advanced packaging is increasingly important.

based on the analysis of the performance of various wafer foundries, trendforce expects that advanced processes and advanced packaging will drive tsmc’s annual revenue growth rate to exceed the industry average next year.

in addition, supported by the continued promotion of ai and the low inventory of various application components, although the annual revenue growth of the foundry industry will return to the 20% level next year, manufacturers still have to face many challenges, including overall it will affect terminal consumer demand, whether high costs will affect the deployment of ai, and whether expansion of production will increase capital expenditures.

trendforce estimates that 7, 6 nanometer, 5, 4 nanometer and 3 nanometer processes will contribute 45% to global foundry revenue next year. according to industry observations, the relevant processes are technology nodes that leading manufacturer tsmc is good at and leading.

advanced packaging production capacity is tight. trendforce analyzes that driven by the large-area demand for ai chips, 2.5d advanced packaging will be in serious short supply from 2023 to 2024. major manufacturers such as tsmc, samsung, and intel that provide complete solutions for front-end manufacturing and back-end packaging are all actively building production capacity, it is estimated that the 2.5d packaging revenue provided by wafer foundries will increase by more than 120% annually next year. although it accounts for less than 5% of the overall wafer foundry revenue, its importance is increasing day by day.

tsmc chairman and president wei zhejia mentioned at a previous press conference that in the packaging part, tsmc will only focus on the most advanced back-end technologies. these technologies will help customers develop forward-looking products.

03

the “excess contention” of mature nodes

first, the term "mature semiconductor" covers a range of different types of chips, each with its own supply and demand dynamics. these include both specific types of semiconductors, including logic, power, rf, and hybrid analog and digital semiconductors, as well as mature semiconductors for specific types of end uses, such as automotive, robotics, drones, industrial automation, aerospace, and other industries . therefore, there is no single market for "traditional semiconductors" and we cannot generally say "overcapacity".

secondly, most of the world's mature semiconductor node production capacity is in the hands of idms. in china, mature node production capacity is dominated by companies engaged in foundry services, which includes a high degree of specialization. foundries manufacture semiconductors based on designs provided by customers, relying on market demand determined by the foundry's customers rather than the foundry itself. the company designs semiconductors closely aligned with the needs of specific industries, striving to carefully balance supply and demand, and our business model is designed to avoid "overcapacity." foundries tend to be highly specialized, and typically each fab operated by a company will be set up for a very specific customer product.

in addition, because the profit margin of mature node semiconductors is very low, it is difficult for foundries to switch production lines quickly, so foundries and customers prefer to sign long-term contracts to lock in the supply of specific types of semiconductors. this is particularly suitable for industries with long product life cycles, high safety requirements, and strict certification of product quality and reliability, such as medical equipment and automotive applications.

finally, there is the concept of “economic overcapacity” that is often overlooked or misunderstood in the industry. this refers to the fact that the global industry actually considers a certain amount of excess supply to be desirable and critical to smoothing out expected and common fluctuations in supply and demand. some of these risks include unpredictable tool downtime; natural disasters, such as the fukushima earthquake that affected front-end manufacturing and materials suppliers; the texas winter freeze that affected samsung; and other incidents that have occurred in the past few years, such as major facilities a fire breaks out. some people in the industry believe that the optimal level of overcapacity is around 15-20%! even after severe chip shortages occurred during the epidemic, some mature nodes will continue to experience rolling shortages, and the health of the entire system requires a certain degree of economic excess capacity.

04

advanced processes are just a stage for giant battles, while mature processes are the "wealth" pursued by the world.

the u.s. commerce department surveyed about 100 u.s. companies in the first quarter of 2024 to determine their reliance on chinese companies for mature semiconductors. the dependence issue is closely tied to the overcapacity issue, as u.s. officials worry that overcapacity could lead to greater dependence, leaving companies vulnerable to supply chain disruptions at mature nodes.

the situation is also complicated by the fact that a large number of chips are assembled in china. u.s. companies rarely import individual semiconductors, but rather import products containing one or more mature semiconductors that may originate in china, either from domestic foundries or from foreign foundries operating in china. as a result of u.s. export controls and china's increasing pressure on certain government and state-owned enterprise supply chains to reduce the volume of foreign semiconductors, the proportion of it products produced in china by foreign companies that are mature semiconductors from china may will rise.

companies that make cutting-edge chips often attribute the increase in demand to the growth of ai applications that rely on cpus, gpus or specialized neural processing chips. applications that make less headlines include smartphone application processors, high-performance computing (hpc) and cloud server chips.

when next-generation technologies become available and key customers for leading applications are ready to move to the next leading edge node, there will then be capacity gaps in the fabs, especially if volumes are high.

but more chips are built on more mature nodes. for example, electric vehicles have increased demand for power management ics (pmics). pmics usually use mature nodes such as 180nm or 130nm, but using bcd processes (bipolar, cmos, d-mos), pmics are becoming more and more intelligent, integrating more and more digital logic in addition to analog circuits. therefore, designs are moving to 90nm, 55nm and 40nm bcd process nodes.

at the same time, sensor demand may still be below the 180 and 150nm nodes. for automotive applications that require high voltage resistance, they are integrated with other analog circuits on the bcd process and mainly use 180nm or 130nm. advanced smart sensors integrated with microcontrollers are moving to 65nm or 40nm, but this is the latest technology for these applications. . top cmos image sensors use 22nm low-power process and are moving to 12nm finfet process.

process nodes typically target specific applications and use cases. chips for iot systems represent some divergence in target process nodes, with most staying at nodes like 40 and 22nm for cost reasons. but as ai moves to the edge, more devices will have some inference capabilities, and the chips that perform that function will need higher performance than other digital logic, so they are moving to 6nm.

analog and mixed-signal chips also tend to lag behind. umc noted: "if the application has a mix of analog and digital circuits, then we believe 55nm is the best choice. pure analog tends to stay at the 8-inch advanced nodes - typically 180 and 150nm."

mature processes are not static. some fabs breathe new life into old processes by making improvements to attract new designs. these include introducing specific transistor devices to improve performance or minimize leakage, shrinking the process to improve cost and tool utilization, adding specific rf capabilities or high voltages to enable mixed-signal systems, or adding automotive-grade certification.

the emergence of chiplet technology also affects these choices. in theory, one no longer needs to migrate certain functions to more advanced nodes and can simply put everything on a single chip. instead, only the parts that truly require advanced node functionality can be moved there, minimizing the die size of expensive nodes. the rest can be integrated within the package as a separate chiplet. while chiplets can save chip costs, advanced packaging costs must be reduced to achieve net cost savings.