Top 10 highlights of the global semiconductor industry in 2025

0 1

Mass production using processes of 2nm and below

2025 is the time point for advanced process foundries to deliver processes of 2nm and below. TSMC's 2nm process is expected to enter mass production in the second half of 2025, which is also the first process node for TSMC's transition from FinFet architecture to GAA (Fully Surrounding Gate) architecture, and will introduce nanosheet transistor technology. Samsung plans to mass produce the 2nm process SF2 by 2025 and will gradually launch different versions such as SF2P, SF2X, SF2Z, and SF2A from 2025 to 2027, targeting mobile, high-performance computing, AI (both SF2X and SF2Z are aimed at this field, but SF2Z adopts backside power supply technology), and automotive fields. Intel's Intel 18A (1.8nm) will also be mass-produced in 2025, using RibbonFET full surround gate transistor architecture and PowerVia backside power supply technology. The first external customer to adopt Intel 18A is expected to complete chip fabrication in the first half of 2025.

02

HBM4 will be shipped in the second half of the year at the earliest

HBM's iteration and manufacturing have entered racing mode. It is reported that in order to keep up with Nvidia's new product release pace, SK Hynix's originally planned HBM4 for mass production in 2026 will be advanced to the second half of 2025, using TSMC's 3nm process. Samsung has also been rumored to plan to start mass production immediately after completing the development of HBM4 by the end of 2025, with target customers including Microsoft and Meta. According to the preliminary specification of HBM4 released by JEDEC Solid State Technology Association, HBM4 has increased the number of layers within a single stack, from a maximum of 12 layers in HBM3 to a maximum of 16 layers. It will support 2048 bit interfaces per stack and achieve a data transfer rate of 6.4GT/s.

03

Advanced packaging production capacity continues to increase

In 2024, the advanced packaging industry will recover and lead the packaging and testing industry to improve. In 2025, the demand for advanced packaging in the market is expected to continue to recover, and OSAT (packaging and testing foundry) and leading wafer fabs will further expand their advanced packaging production capacity and promote technological upgrades. While accelerating the expansion of CoWoS production capacity, TSMC will increase the mask size of CoWoS from 3.3 times in 2023 to 5.5 times between 2025 and 2026, with a substrate area exceeding 100 × 100mm and a maximum capacity of 12 HBM4s. Changdian Technology's Shanghai Lingang Automotive Grade Chip Finished Product Manufacturing Base is planned to be completed and put into use by 2025. The first phase of the project of Tongfu Chaowei (Suzhou) Microelectronics Co., Ltd., a new base of Tongfu Chaowei in Suzhou, is expected to achieve mass production in January 2025 and engage in advanced FCBGA packaging and testing. Huatian Technology's Jiangsu Pangu Semiconductor Board Level Packaging and Testing Project will complete the process equipment relocation and achieve project production in the first quarter of 2025, committed to promoting the large-scale production of board level fan out packaging technology.

04

AI processor shipments continue to remain strong

In 2025, a batch of new AI chip products will be released or launched, with iterative updates in architecture, process, heat dissipation methods, etc., in order to provide stronger computing power and energy efficiency. Intel will launch the Panther Lake AI PC processor and Clearwater Forest data center processor based on the Intel 18A process in 2025. Nvidia expects to launch the next generation "Blackwell Ultra" GB300 in 2025, and the previously released GB200 NVL4 supercomputer will be available for supply in the second half of 2025. AMD will launch the next generation AMD CDNA 4 architecture in 2025, which is expected to improve AI inference performance by 35 times compared to Instinct accelerators based on CDNA 3 architecture. The shipment momentum of AI processors will drive the growth of storage, packaging and other links.

05

Advanced intelligent driving chips enter the window of entry for vehicles

2025 is regarded by many automotive chip manufacturers as the final point for advanced intelligent driving and the window period for mass production and installation. The Horizon SuperDrive full scenario intelligent driving solution is expected to deliver the first mass-produced cooperative model in the third quarter of 2025, in addition to the flagship version of Journey 6, which aims to win the key production window of 2025. The Black Sesame Wudang series is expected to enter mass production by 2025, providing cross domain integration capabilities for autonomous driving, intelligent cockpit, body control, and other computing functions. Xinqing Technology's autonomous driving chip "Xingchen No.1" is planned to achieve mass production by 2025. In terms of international enterprises, Qualcomm's Snapdragon Ride Ultimate platform, released in October 2024, will be sampled in 2025. In addition, based on the previous generation Snapdragon Ride platform, Qualcomm has collaborated with more than ten Chinese partners to develop integrated solutions for intelligent driving and cabin driving, and will continue to implement them in 2025.

06

Quantum processors on a large scale

The United Nations has declared 2025 as the 'Year of Quantum Science and Technology'. At the end of 2024, the latest quantum processors such as Google Willow and the "Zuchong No. 3" from the University of Science and Technology of China made consecutive appearances, achieving breakthroughs in quantum bit count, quantum error correction, coherence time, and quantum computing superiority. In 2025, the industry is expected to usher in larger scale quantum processors and computing systems. IBM will release the Kookaburra, a multi chip processor with 1386 qubits and quantum communication links, in 2025. As a demonstration, IBM will integrate three Kookaburra chips into a system containing 4158 qubits. In addition, by 2025, IBM will showcase the first quantum centric supercomputer by integrating modular processors, middleware, and quantum communication, and further enhance the quality, execution, speed, and parallelization of quantum circuits.

07

The manufacturing technology of silicon optical chips is maturing

With the rapid increase in data transmission speed requirements for AI servers, silicon optical chips that integrate silicon chip process flow and the advantages of high speed and high energy efficiency in optoelectronics have attracted much attention. By 2025, the manufacturing process of silicon optical chips will mature. The People's Government of Hunan Province mentioned in the "Action Plan for Accelerating the Construction of the World Optics Valley" that by 2025, the development of 12 inch basic silicon optical wafer technology will be completed, forming an internationally leading silicon optical wafer foundry and production manufacturing capacity. The Action Plan for Accelerating the Innovative Development of the Optical Chip Industry in Guangdong Province (2024-2030) mentions supporting the research and optimization of optical chip related components and processes, as well as supporting the research and continuous optimization of optical chip related manufacturing processes such as silicon optical integration, heterogeneous integration, epitaxial growth and epitaxial processes, and manufacturing processes. In terms of international enterprises, Nvidia showcased a silicon photonics prototype developed in collaboration with TSMC at IEDM 2024 in December 2024. TSMC will achieve a 1.6T optical engine suitable for pluggable optical modules by 2025 and complete the COUPE (Compact Universal Photon Engine) validation for small pluggable products. According to TSMC, COUPE technology uses SoIC-X chip stacking technology to stack electronic dies on top of photonic dies, providing lower resistance and higher energy efficiency at die to die interfaces.

08

AI acceleration and integration with semiconductor production

AI is accelerating its integration with the entire process of semiconductor design, manufacturing, and more. In March 2024, Xinsi Technology will deploy the complete AI driven EDA technology stack on the Nvidia GH200 Grace Hopper super chip platform, achieving up to 15 times efficiency improvement in chip design, verification, simulation, and manufacturing processes. In July 2024, Aitomatic released its first open-source large-scale model, SemiKong, tailored for the semiconductor industry, claiming to shorten the time to market for chip design, improve first die yield, and accelerate the learning curve for engineers. By 2025, AI is expected to assist or replace EDA's fitting algorithms and work, including corner prediction, data fitting, pattern learning, etc. In terms of manufacturing, TSMC is expected to use Nvidia's computational lithography platform cuLitho in the development of 2nm and below processes. The platform provides accelerated computing and generative AI, allowing wafer fabs to free up available computing power and engineering bandwidth to design more innovative solutions when developing 2nm and more advanced new technologies.

09

RISC-V opens up high-performance productization

In 2024, RISC-V will further penetrate into the field of high-performance chips. The Institute of Computing Technology of the Chinese Academy of Sciences and the Beijing Open Source Chip Research Institute released the third generation "Fragrant Hill" open source high-performance RISC-V processor core, which has entered the world's first echelon in performance. At the same time, a group of domestic enterprises such as Xinlai Technology, Yisiwei, Saifang Technology, and Jindie Space have released IPs, toolchains, software platforms, AI PC chips, and other technologies for high-performance computing fields such as artificial intelligence, data centers, autonomous driving, and mobile terminals AI MCU、 Multimedia processors and other chips, as well as development boards and other products, have formed a number of cases in the fields of laptops, cloud computing, and industry applications. Krste Asanovi, the main inventor of RISC-V, predicts that the number of RISC-V cores will increase to 80 billion by 2025. 2025 is also regarded as a crucial year for China's RISC-V industry to bridge the gap and create high-performance benchmark products. Accelerating the creation of iconic products, deepening ecological construction, and promoting the integration of RISC-V and AI have become industry consensus.

10

Silicon carbide enters the 8-inch production capacity conversion stage

In 2024, the silicon carbide industry accelerated its transition from 6 inches to 8 inches. In 2025, the silicon carbide industry will officially enter the stage of 8-inch capacity conversion. The joint venture factory project established by STMicroelectronics in China, the Ansteel Silicon Carbide Device Factory, is expected to start mass production in 2025. The Xinlian integrated 8-inch silicon carbide production line is planned to enter mass production in 2025. The Roma Fukuoka Chikugo factory plans to start mass production in 2025. Resonac plans to begin mass production of 8-inch silicon carbide substrates by 2025. Ansenmei will start production of 8-inch silicon carbide wafers in 2025.