Performance killer! TSMC's 3nm process iteration, new generation mobile phone chip battle

TSMC 3nm achieves higher transistor density and lower power consumption
TSMC's 3nm process technology adopts advanced manufacturing processes, which can achieve higher transistor density and lower power consumption, significantly improving chip performance and energy efficiency.
TSMC's 3nm process technology is the last generation based on FinFET technology. It uses up to 25 extreme ultraviolet (EUV) layers, some of which use double exposure technology to increase the density of logic and SRAM transistors.
Compared to the previous generation 5nm process, the 3nm process achieved approximately 1.6 times logic density expansion, 18% speed increase, and 34% power reduction. These improvements are mainly attributed to FinFlex ™ The introduction of technology provides standard units with different fin configurations to achieve better power efficiency and performance optimization.
FinFlex? Technology allows designers to choose different combinations of fin numbers and threshold voltage (Vt) according to their needs to meet wide range speed and leakage requirements on the same chip. This flexibility allows designers to optimize circuit performance more accurately.
N3: This is the basic version of TSMC's 3nm process and has been put into production. It provides significant performance and power improvements, providing strong support for various chip designs.
N3E: As an enhanced version of N3, N3E adopts 19 EUV layers and does not rely on EUV dual exposure technology, thereby reducing manufacturing complexity and cost. Although its logical density is slightly lower than N3, it has a wider process window and better yield.
N3P: This is the optical miniaturization version of N3E, which reduces power consumption, enhances performance, and density by adjusting optical performance. According to TSMC, N3P will provide a 5% higher speed under the same leakage conditions, or a 5-10% decrease in power at the same speed, as well as a 1.04 times chip density. It is expected that N3P will become one of TSMC's most popular N3 nodes and is planned to be launched in the second half of 2024.
N3X: A process tailored specifically for high-performance computing chips such as CPUs and GPUs. The N3X will support a voltage of approximately 1.2V and prioritize performance and maximum clock frequency. It is revealed that N3X will be put into production in 2025.
N3AE (Auto Early): A node designed specifically for advanced chips in automotive applications. It provides an N3E based Automotive Process Design Kit (PDK) and plans to launch a preliminary version in 2023, while a version that fully meets the automotive qualification requirements will be released in 2025.
Due to the progressiveness of 3nm process technology, the market demand is very strong. Several technology giants such as Apple, Qualcomm, MediaTek, Nvidia, etc. have placed orders with TSMC for chips with a 3nm process for high-end applications such as smartphones, data centers, and AI accelerators.
In order to meet market demand, TSMC continues to expand its 3nm process production capacity. According to reports, TSMC's 3nm process capacity is expected to triple this year, but supply is still in short supply.
The use of TSMC's 3nm process chips significantly improves performance
Which specific chips use TSMC's 3nm process? Apple's M4 and expected A17 are reportedly using TSMC's 3nm process. The M4 chip was released on May 7, 2024 and will be installed on Apple's next-generation iPad Pro. The number of transistors in the M4 chip has reached 28 billion, using TSMC's second-generation 3nm process, with 10 cores, including 4 performance cores and 6 efficiency cores, both equipped with a new version of machine learning accelerator.
The M4 is equipped with a brand new 10 core GPU that utilizes dynamic caching, hardware accelerated ray tracing technology, and grid coloring to enhance graphics processing performance. It also integrates a new neural engine with 38 TOPS computing power, focusing on accelerating AI related workloads. Apple claims that the CPU performance of the M4 has improved by up to 1.5 times compared to the M2 chip, and it has excellent performance per watt.
Apple plans to launch the top of the line iPhone 15 series in the second half of 2024, which will feature the A17 biomimetic chip. The A17 biomimetic chip is expected to use TSMC's most advanced 3nm process technology. Compared to the previous generation A16 Bionic chip, A17 will have improvements in transistor quantity, performance, power consumption, and other aspects.
The A17 biomimetic chip will significantly improve CPU performance. It is speculated that its single core and multi-core performance will be significantly improved compared to the previous generation A16 chips. This improvement will make the iPhone 15 series smoother and faster in handling complex tasks and applications. The A17 chip has also been upgraded in terms of GPU, and is expected to adopt a new architecture design and introduce more GPU cores to improve 3D graphics rendering and gaming performance. This allows users to have a better experience when playing games, watching high-definition videos, or performing other graphics intensive tasks.
The A17 chip is equipped with powerful neural network processors (such as neural engines) to support machine learning and artificial intelligence applications. This computing power makes the A17 chip widely applicable in various fields such as speech recognition, image processing, autonomous driving, and medical diagnosis. The A17 biomimetic chip will first be applied to Apple's upcoming iPhone 15 series. In addition to smartphones, the A17 chip may also be used in other Apple products, such as iPads, Macs, etc.
The Qualcomm Snapdragon 8 Gen4 is expected to adopt TSMC's second-generation 3nm process N3E, which is an improved version of N3B. Compared to N3B, it has better power consumption performance, higher yield, and lower production costs. Due to the use of 3nm process technology, Snapdragon 8 Gen4 has significantly improved performance compared to its predecessor. Higher transistor density and lower power consumption will result in better performance of the chip in terms of operating speed and battery life.
Snapdragon 8 Gen4 will adopt its self-developed Nuvia architecture and no longer rely on Arm's public version architecture. The Snapdragon 8 Gen4 is expected to adopt an eight core solution consisting of two performance cores and six energy efficiency cores, with a CPU design using a 2 Phoenix L+6 Phoenix M architecture. The Snapdragon 8 Gen4, as Qualcomm's flagship chip, is expected to become the preferred chip for high-end models in the second half of 2024 and is highly sought after by many mobile phone manufacturers.
MediaTek Dimensity 9400 adopts TSMC's second-generation 3nm process (N3E), which is MediaTek's first 3nm mobile phone chip. N3E is an enhanced version of N3B, expected to be more widely used than N3B, with higher yield and lower cost. This advanced process technology will bring higher performance, lower power consumption, and smaller size to the Dimensity 9400.
Due to the adoption of 3nm process technology and advanced architecture, the Dimensity 9400 will have a significant improvement in performance, which can meet the performance requirements of high-end smartphones and other mobile devices. The Dimensity 9400, as MediaTek's flagship chip in 2024, will face the high-end smartphone market and compete fiercely with competitors such as Qualcomm Snapdragon.
Nvidia B100 also adopts TSMC's 3nm process technology. The B100 chip has achieved a huge leap in performance, with its efficiency more than four times that of the previous generation H100 chip. This improvement is mainly due to its advanced process technology and optimized architecture design. The memory capacity of the B100 chip is also nearly 40% higher than that of the H200, making it more adept at handling large-scale data and complex tasks.
With the significant improvement of B100 chip performance, its power consumption has also significantly increased, with a maximum design power consumption of 1000W. The traditional air-cooled cooling scheme can no longer meet its cooling needs. Nvidia has adopted "water cooling" technology for the B100 chip, which is an important milestone in the shift towards "liquid cooling" technology. Water cooling technology has advantages such as low power consumption, high heat dissipation, and low noise, which can effectively solve the heat dissipation problem of B100 chips. The B100 chip, as Nvidia's new generation AI chip, will be widely used in fields such as artificial intelligence, deep learning, and big data analysis.
Write at the end
TSMC's 3nm process technology is one of the most competitive processes in the current semiconductor industry. It not only improves the performance and power consumption of chips, but also provides rich choices for diverse market demands. With the support of TSMC's 3nm process technology, the performance of chips such as Apple, Qualcomm, MediaTek, and Nvidia has also achieved significant leaps. In the future, with the continuous upgrading of technology and the expansion of production capacity, TSMC is expected to achieve greater success in the 3nm process field.