For Apple, the development of 3nm (nanometer) technology is undoubtedly at the forefront of its agenda.
According to previous reports by Bloomberg, Apple has been testing its new M3 chip's base version on the new Mac mini. Now, the same source has revealed another report about the M3 Mac, this time concerning a more powerful M3 Max chip. Apple has reportedly been experimenting with a chip containing as many as 40 GPU cores for a new high-end MacBook Pro.
Though details about the M3 chip have not been disclosed, sources have shared that a new laptop, codenamed J514, will be utilizing an Apple Silicon chip with a 16-core CPU and 40-core GPU. Specifically, the chip known as M3 Max will have 12 high-performance cores for demanding tasks and 4 efficient cores to save battery power during less intensive applications. Compared to the M2 Max, it includes four additional CPU cores and two more GPU cores.
In an earlier report, the new M3 base chip was mentioned to have an 8-core CPU and 10-core GPU, similar to the M2. However, the M3 Pro is expected to have 12 CPU cores and 18 GPU cores, two more than the M2 Pro.
Furthermore, it's reported that the upcoming top-tier iPhone 15 Pro model will feature a brand-new A17 processor, which shares many similarities with the M3 chip and will be built using the new 3nm process.
Why the Emphasis on 3nm?
Apple has long been known for implementing cutting-edge technology. The company's last significant leap in manufacturing processes occurred in 2020 when it utilized Taiwan Semiconductor Manufacturing Company (TSMC)'s 5nm process to produce the A14 Bionic and M1 chips. Additionally, the Apple Watch's S6, S7, and S8 chips continued to use the 7nm process, based on the A13 Bionic-the last 7nm chip designed by Apple for the iPhone.
Last year, Apple introduced the A16 Bionic chip in the iPhone 14 Pro and iPhone 14 Pro Max, claiming it was a 4nm chip made with TSMC's "N4" process, though it was actually manufactured using an enhanced version of TSMC's 5nm N5 and N5P processes. This constraint was due to the increasing difficulty of advancing the process further.
Creating chips with a 3nm process is a complex task that requires high precision and innovative lithography, etching, and deposition methods to create intricate circuits on a microscopic scale. Even the slightest deviation could result in significant yield loss. Manufacturers like TSMC, Samsung, and Intel have encountered challenges with this technology, and the price of the 3nm process is steep, exceeding what most customers are willing to pay.
Now, with TSMC's groundbreaking new process, Apple is poised to embrace the next big leap in growth.
Data from manufacturers suggest that 3nm should offer Apple's chips the most significant performance and efficiency leap since 2020. The increased transistor count enables the chip to execute more tasks at faster speeds while using less power.
According to TSMC's chairman, Mark Liu, the 3nm process's logic density will increase by about 70%, boosting speed by 10-15% with the same power consumption or reducing power consumption by 25-30% at the same speed. Additionally, 3nm chips may support more advanced specialized hardware for artificial intelligence, machine learning, and graphics. This kind of enhancement has been rare in recent years and is likely a reason why Apple is interested in 3nm technology.
TSMC's Full Support
As mentioned, Apple plans to use 3nm technology in several chips. Reports have emerged that Apple has reserved 90% of TSMC's capacity this year to produce A17 Bionic and M3 series processors for iPhones, Macs, and iPads.
As a long-term partner of TSMC, Apple reportedly began ordering 3nm chips as early as 2020 and will continue to do so. The expectation is that Apple will use 3nm chips to introduce new product categories, such as self-driving cars.
TSMC has also made significant efforts to support Apple in this partnership.
According to a report by EE Times in May, Arete Research analyst Brett Simpson estimated that the yield rate for TSMC's A17 and M3 processors was around 55%. However, TSMC's CEO, CC Wei, later stated that the company had achieved "large-scale production, with good yield," though customer demand exceeded supply capacity. In the second half of the year, TSMC will increase production of A17 and M3 chips for Apple, as well as for Intel, AMD, and Nvidia.
However, TSMC's CEO, CC Wei, subsequently stated that the company had achieved "mass production with a good yield rate." However, the demand from clients surpasses their supply capabilities. In the latter half of the year, TSMC plans to increase production of the A17 and M3 chips for Apple, while also manufacturing chips for Intel, AMD, and Nvidia.
According to a report from The Information, TSMC, the chip supplier, took an unusual step. Before the launch of the iPhone 15 Pro and the A17 Bionic chip, they stopped charging Apple for the flawed 3nm chips. The Information claims that TSMC is only charging Apple for "known good dies," and not charging at all for the defective chips. This is very unconventional since TSMC's clients typically have to pay for the entire wafer and all the chips it contains, including any flawed ones.
Given the massive orders Apple secures from TSMC, it's apparent that the chipmaker has every reason to absorb the costs of the flawed chips. Apple's willingness to be the pioneer client for the supplier's new manufacturing process aids in covering the R&D costs of the new nodes and the expenses of the facilities that produce them.
For a new manufacturing process, the cost savings from this decision might be considerable. The Information reported that around 70% of the early 3nm chips are now usable, though this figure might fluctuate depending on the chip being manufactured and typically improves over time with advancements in the process.
Of course, the scale of Apple's orders also allows TSMC to learn faster how to improve and expand the nodes in a mass production context.
As financial reports indicate, TSMC, the world's largest semiconductor foundry, served 532 clients in 2022. However, its top 10 clients contributed to 68% of its revenues. Among them, Apple was the largest, accounting for 23% of TSMC's revenue. Reports highlight that TSMC's revenue from Apple grew by 18.2% in 2022, reaching 17.5 billion US dollars.
Challenges Faced by Other Clients
For Apple and TSMC, the decision made between them is a compromise based on mutual needs. But once TSMC improves the production and yield issues of the 3nm chip with Apple's support, other clients wanting 3nm chips might inevitably face higher prices and charges for flawed chips.
TSMC had previously disclosed that its N3 node has been adopted by designers of High-Performance Computing (HPC) and smartphone SoCs, and early in its lifecycle, the number of adopters was higher than N5. However, TSMC never mentioned how many companies opted for its 3nm manufacturing process.
Nevertheless, John Koeter, Senior VP of IP Marketing and Strategy at Synopsys, once stated, "Synopsys IP for TSMC's 3nm process has been adopted by dozens of leading companies to accelerate their development speed, quickly achieve chip success, and hasten time to market." This gives a hint about the popularity of TSMC's 3nm process.
Yet, these manufacturers seemingly lack the bargaining power Apple holds with TSMC.
Statistics show TSMC's top clients include Apple, Qualcomm, AMD, Broadcom, NVIDIA, MediaTek, Intel, Marvell, NXP, and Unigroup Spreadtrum. Affected by chips based on the N6/N5/N4 processes and a broader influence in the automotive and IoT sectors, Qualcomm became TSMC's second-largest client. Qualcomm spent more than 15 billion US dollars on foundry services in 2022, with the largest share going to TSMC and Samsung. This is a significant contrast to the revenue proportion TSMC gets from exclusively producing Apple's chips.
According to a report from DigiTimes late last year, a wafer using TSMC's 3nm process might cost clients about 20,000 US dollars. In comparison, the cost for TSMC's 5nm process wafer is around 16,000 US dollars.
TSMC's CC Wei also noted in April, "Given the demand from our client (presumably Apple) for N3 exceeds our supply capabilities, we expect N3 to be fully utilized for HPC and smartphone applications by 2023."
In other words, for clients wanting to use TSMC's 3nm this year, not only will they face challenges in capacity, but pricing will also be another significant hurdle for them.
