Semiconductor Industry Observation

Semiconductor Industry Observation

follow

This article is from WeChat official account: Semiconductor Industry Watch (ID: icbank), written by Shao Yiqi, and the picture is from Visual China

Article Summary

The Apple M3 chip faces a short-lived fate, while the M4 appears to improve performance, but faces sales challenges and process limitations.

• •    M4 shows significant performance improvement compared to M2

• •    M3 chip failed to salvage the decline in Mac sales

• • ️   Apple's self-developed chips face process limitations and product development challenges

Half a year after the release of the Apple M3 chip, the successor M4 suddenly caught everyone's attention. Although it was a processor used on the iPad Pro, the performance improvement was not insignificant.

According to Apple, the M4 has a 50% increase in CPU compared to M2, a 4-fold increase in overall GPU rendering, and can provide equivalent performance even with half the power consumption. In short, it is another Apple style performance leap.

But Apple intentionally or unintentionally overlooked the M3- which may be Apple's shortest lived processor. The last Apple processor with such a short lifespan may have been traced back to the A5X, which was first released on the iPad 3 in March 2012. In October of that year, it was announced that the iPad 4 would be equipped with A6X, with an actual lifespan of only 7 months.

The M3 is now facing the same predicament as the A5X in the past. Its powerful performance has always been one of the biggest driving forces for consumers to replace phones. However, in just six months, Apple launched a stronger and better product, and the latest processor became the second latest processor in the blink of an eye. Such iteration speed is inevitably too fast.

A more fatal question has been raised: how many chances do Apple have of winning with its self-developed chips?

1、 The M3 that died without any progress

At the MacBook launch event in the second half of last year, Apple announced the M3 series chips using 3nm technology, which is also the first M-series chip to use 3nm technology. At the launch event, Apple also boasted about its performance parameters:

Apple stated that the M3 chip is equipped with 25 billion transistors -5 billion more than the M2- and is equipped with a 10 core graphics processor using the new generation architecture, bringing graphics processing performance up to 65% faster than the M1. It is equipped with an 8-core central processing unit, including 4 performance cores and 4 energy efficiency cores, which can achieve up to 35% performance improvement compared to the M1. It also supports up to 24GB of unified memory.

The M3 Pro chip is equipped with 37 billion transistors and an 18 core graphics processor. Compared to the M1 Pro, the graphics processor provides a speed increase of up to 40%. The support for unified memory has been increased to up to 36GB, and the 2-core CPU design consists of 6 performance cores and 6 energy efficiency cores, achieving up to 30% single threaded performance improvement compared to the M1 Pro.

The number of transistors in the M3 Max chip has increased to 92 billion, and the 40 core graphics processor is up to 50% faster than the M1 Max, with support for up to 128GB of unified memory. The 16 core central processor is equipped with 12 performance cores and 4 energy efficiency cores, with a speed increase of up to 80% compared to the M1 Max.

From Apple's description, we can see that they are trying their best to avoid direct performance comparisons between M3 and M2 chips, which is a strange thing. After all, the earliest M1 chip was released in 2020, and compared to the M1 series with upgraded technology and core, the 2023 chip can only be said to be somewhat inferior.

So why did Apple release the M3 chip and corresponding MacBook even though it knew the performance improvement was not significant?

The answer is quite simple, it is that the sales performance of the previous generation M2 chip's Mac product was too dismal.

According to The Elec, due to the severe downturn in the PC market and a sharp decline in Mac sales, Apple completely suspended production of the M2 series processors in January 2023.

Unlike A-series chips that use Fan Out Wafer Level Packaging (FO-WLP), which TSMC refers to as InFO (Integrated Fan Out), the final processing of M2 chips adopts a universal flip chip packaging process. Once TSMC completes the preprocessing process, the wafer processing products will be sent to Amkor's Korean factory for packaging work. In January and February 2023, TSMC did not deliver these products, and the production line was idle for two months. According to insiders, although wafers have been arriving since March, the quantity is only half of previous years.

Apple CEO Cook also acknowledged during the Q1 2023 earnings conference call that Apple is facing a "challenging" situation in the PC market. "This industry is shrinking," Cook said. "Apple has a lower share, but we have a competitive advantage in Apple chips, so strategically, Apple is in a favorable position in the market, but it will be a bit difficult in the short term."

In order to salvage the lack of competitiveness in Mac products equipped with M2, Apple's choice is to launch M3 chips using 3nm technology and corresponding MacBooks by the end of 2023.

But when Apple made up its mind to launch a 3nm process Mac in 2023, it encountered a new problem - TSMC's N3B process.

TSMC experienced a long nightmare on the N3 node. The initial N3, also known as N3B, had 25 EUV layers, almost twice the size of N5, which made production difficult and expensive. Considering the lackluster improvements in performance, power, and density, most customers were unwilling to pay high fees.

The problems with N3 ultimately caused TSMC to miss the 2-year upgrade cycle for the main process nodes. N3B was ultimately put into production in the fourth quarter of 2022, while N3E had to wait until the middle and later stages of 2023. This not only forced Apple to postpone its chip plan - it had originally planned to use N3 technology on the iPhone in 2022, but many customers also gave up on N3B, such as Zen 5, Intel GPU, and Broadcom custom ASIC, either continuing to use N5 or turning to N3E as a subsequent improvement process.

In fact, apart from Apple's M3 series and A17 Pro, almost no manufacturer is willing to use TSMC's N3B process, which also lays the groundwork for the Sweetheart deal between Apple and TSMC.

In August 2023, foreign technology media The Information revealed that Apple had placed a huge order for 3nm chips with TSMC, but required TSMC to bear the responsibility for unqualified chips. TSMC's initial yield rate for 3nm was about 70%, and after reaching such an agreement, Apple could save billions of dollars.

This has also formed a strange landscape. On the one hand, TSMC's 3nm technology stands out in the world, while on the other hand, Apple is unwilling to pay more for substandard wafers. Rather than being a strong demand from Apple, it is more like TSMC's half hearted efforts, allowing Apple to form a de facto monopoly on N3B.

This situation inevitably reminds us of TSMC's N10 and N20, two equally short-lived processes.

At that time, after experiencing a peak in the fourth quarter of 2016, some customers of TSMC's 16nm/20nm process shifted to 10nm in 2017, including Huawei HiSilicon's Kirin 970 and Apple's A11 and A10X, with A11 being TSMC's main customer of 10nm.

But apart from these two individuals, there are few TSMC customers who use 10nm technology. 10nm once accounted for 11% of TSMC's revenue in its financial reports, but after 2017, this proportion rapidly decreased, and 7nm technology became a choice for more manufacturers.

As for TSMC's 20nm technology, it is even more notorious. Although many processors, such as Qualcomm Snapdragon 810, Apple A8, Nvidia Tegra X1, MediaTek Helio X20, etc., have used this technology, they have all encountered severe heat generation and power loss issues, allowing everyone to experience the horror of "Taiwan leakage" in the past.

The poor performance of 20nm resulted in customers either quickly migrating to the next 16nm node or reluctantly returning to the 28nm process. In 2015, not only did it leave phone and processor manufacturers on pins and needles, but it also brought TSMC through its darkest moments.

Nowadays, the N3B from TSMC is likely to follow the footsteps of N10 and N20. According to WikiChip, the N3 node seems to be a one-time node, and TSMC engineers encountered some obstacles on their way forward, so they decided to make changes midway. At a higher level, N3B has different PPAs and the company claims to have "completely different" design rules aimed at increasing production, resulting in subsequent N3E not providing any direct migration path from N3B. Ultimately, N3B became a deadlock for designers, which is why TSMC wants most customers to use N3E.

The question is, why did TSMC specifically launch an N3B? A simple explanation is that TSMC is only fulfilling its promise to early technology adopters like Apple, and perhaps the N3B process will fade out of everyone's sight after the end of this year.

Of course, it's not just the N3B, but also the Apple M3 series and A17 Pro from N3B. After deciding to adopt this short-lived process, Apple may have foreseen the outcome of these processors long ago.

2、 The lackluster M4

Looking back on the debut of Apple's first self-developed Mac processor in 2020, it caused a whirlwind in the technology and semiconductor circles. The once dominant Intel was kicked out by Apple, and all the media was inventing Apple's self-developed path starting from A4. Everyone was writing a cool article about Apple's comeback and breaking the monopoly.

This is one of the most glorious moments in Apple's history. Some have listed Apple's Mac transitioning from the Motorola 68000 platform to the Power PC platform, then to the Intel x86 platform, and now choosing the self-developed ARM platform, in an attempt to prove the beginning of a new era.

In the first quarter of 2021, due to factors such as the pandemic, the overall shipment volume of the PC industry increased by 55% year-on-year, while Apple's Mac sales saw an astonishing 111.5% growth. In the three years from 2020 to 2022, the Windows PC market grew by 6%, while the Mac market grew by 60%. In 2022, the Mac market share exceeded 10%, reaching 10.8%. In Apple's headquarters - the US market, the Mac market share is even higher, reaching 17% in the fourth quarter of 2022.

But all the good news stayed in the past three years. Starting from the beginning of 2023, Apple Mac began to decline endlessly: in Q1 2023, global shipments of Mac reached 4.819 million units, with a market share of 8.7%, a year-on-year decline of 34.2%; In Q2, the global shipment volume of Mac was 5.293 million units, with a market share of 8.9%, a year-on-year decrease of 0.3%; In Q3, the global shipment volume of Mac was 6.266 million units, with a market share of 9.7%, a year-on-year decrease of 24.2%.

At Apple's Q3 2023 financial report meeting, Apple CFO Luca Maestri attributed the decline in Mac sales to overly fierce market competition and suppressed sales releases in 2022, resulting in disproportionate data.

Apple CEO told CNBC that the market performance of Mac is expected to improve during the Christmas shopping season. He believes that due to Apple's launch of the M3 series chip and new Mac products equipped with the chip, the Mac business will have a significant improvement in the fourth quarter of 2023.

Unfortunately, the cunning Cook also miscalculated this time. According to Apple's Q1 2024 financial report, revenue from Mac for that quarter was $7.735 billion, compared to $10.852 billion in the same period last year, far below analysts' previous expectations of $9.63 billion.

M3 did not boost Mac sales, which surprised analysts and Apple executives. It also put the M3 MacBook at the forefront of the market. It should be noted that Apple has spent a lot of money on M3. According to foreign analyst Jay Goldberg, Apple spent $1 billion on chip development for the M3 series, and the astronomical cost of chip development was much lower than expected.

Interestingly, renowned analyst Guo Minggui stated in October 2023 that if the M3 chip still cannot boost MacBook shipments, Apple may launch a completely redesigned MacBook Pro in 2025, and may also consider (yet to be determined) launching a low-priced version of the MacBook to further increase shipments, with a target annual shipment of over 8 million to 10 million units.

In other words, Apple has abandoned the ARM version of Mac with chip upgrades as its main selling point, and instead considered focusing on design and pricing. Isn't it a bit familiar? This is exactly what the Intel version of MacBook did before, can't we sell large-sized ones anymore? Let's make a 12 inch MacBook, lack upgrade highlights? Then add flashy hardware features like TouchBar and butterfly keyboard, and even if there is no selling point, it is necessary to forcefully create one.

At this moment, the Apple Mac was so proud when it was launched in M1, but now it is so disappointed. Apple thought it had created a storm of ARM laptop popularity, but three years later, it found itself just a pig in the wind, and the Mac returned to its rightful market position.

The success of M1 is partly due to the surge in demand brought about by working from home, and partly due to the contribution of TSMC. Although the TSMC N3B process mentioned earlier has brought a big leap forward, at this time point in 2020, TSMC N5 has taken several steps ahead of Intel and Samsung's respective processes, which has also contributed to M1's extremely excellent power control.

Nowadays, it seems that the self-developed chips of Apple's M series have a feeling of "one go, another go, three runs out". Although Apple is still emphasizing the improvement of M2 and M3 compared to M1, the best-selling and most recognized products in the market are still the M1.

With the large-scale production of TSMC's N3E technology, the M3 series is destined to become Apple's abandoned product. However, Apple seems not to give up and has prepared the N3E M4 at a relatively fast pace to replace the expensive M3, first reducing costs before making any other plans.

According to Mark Gurman from Bloomberg, Apple's goal is to release new Mac products from the end of this year to the beginning of next year. The new iMac, low-end 14 inch MacBook Pro, high-end 14 inch and 16 inch MacBook Pro, and Mac mini will all use the M4 chip.

Gurman believes that Apple plans to launch more M4 Macs throughout 2025, including the 13 inch and 15 inch MacBook Air updated in the spring of 2025, Mac Studio updated around mid-2025, and Mac Pro updated later in 2025.

It stated that the M4 chip series includes the entry-level Donnan, the more powerful Brava, and the high-end processor codenamed Hidra. Apple plans to emphasize the artificial intelligence processing capabilities of these three processors and how they will integrate with the next version of macOS released at WWDC this summer.

The Donan chip will be used for entry-level MacBook Pro, new MacBook Air, and low-end Mac mini, while the Brava chip will be used for high-end MacBook Pro and high-end Mac mini. As for Mac Studio, Apple is testing versions using the yet to be released M3 series chips and M4 Brava processor variants, while the Hidra chip will be installed on Apple's most high-end desktop Mac Pro.

Will M4 perform better than M3? Not necessarily.

Firstly, N3E is not a qualitative leap compared to N3B, and it can even be assumed that the two chips are actually Apple's two 3nm test papers. When we put M3 and M4 together, we will find that even the improvement on paper is somewhat lackluster.

Firstly, regarding the CPU part, the full version of the M4 includes 4 performance cores and 6 efficiency cores, with 2 more efficiency cores compared to the M3. On the other hand, the beggar's version of the iPad has made a move on the core, with a configuration of 3 performance cores and 6 efficiency cores. On the GPU part, the M4 is almost identical to the M3, with 10 cores, and also supports the dynamic caching, ray tracing hardware acceleration, and grid coloring functions that the M3 already has.

The NPU part is the biggest improvement for M4, with its 16 core neural network having a computing power of up to 38 TOPS, which is a full 20 TOPS higher than M3's 18 TOPS, and also surpasses Apple's A17 Pro's NPU's 35 TOPS computing power. According to Apple, this is of great help for future AI applications.

Looking at the actual performance score, in the latest GeekBench 6 performance test, the iPad Pro equipped with the full version M4 scored 3767 points on a single core and 14677 points on a multi-core. Compared with the previous generation M3 (3053), the single core performance improved by 25%, while the multi-core performance was roughly equivalent to the latest 11 core M3 Pro (15261). The GPU's Metal test score was 53792 points, which was 13.4% higher than the M3 (47420) and did not exceed the larger M3 Pro. It should be noted that the processor frequency of the iPad Pro in the benchmark has been increased to 4.4GHz, while the actual frequency of the M3 Max with the highest frequency is only around 4.1GHz.

In the Geekbench ML 0.6.0 test representing NPU, the score of M4 was 9234, which is about 22.9% higher than the 7511 score of iPad Pro equipped with M2, and only 10.4% higher than the M4 (8365) of 14 inch MacBook Pro.

At this point, people may feel that something is wrong. Isn't there an additional 20 TOPS of computing power? Why did the running score only increase by 10.4%? In fact, Apple played a text game here. The 18TOPS of the M3 was obtained with FP16 accuracy, while the 38TOPS of the M4 was obtained with INT8 accuracy. The latter actually had to be halved, which is 19TOPS. Therefore, the actual improvement in TOPS is only about 5%.

Compared to the original Intel x86 processor, M1 is undoubtedly a cross era processor, but the subsequent M2, M3, and M4 seem to have only been continuously improved on the basis of the M1 architecture, without significant improvement. The frequency of M4 on the iPad has even increased to 4.4GHz, and it should be noted that the frequency of M3 running on MacBook Pro is only 4.05GHz.

Why did Apple choose to launch the M4 chip with the iPad Pro? Mainly to drive sales of iPads. In February of this year, iPad revenue became a rare highlight in the first quarter. Apple's net revenue from iPads for the quarter was $9.396 billion, an increase of 29.7% from $7.248 billion in the same period in 2023, exceeding Wall Street analysts' expectations of $7.76 billion. The East is not bright, and the West is bright. Since M3 cannot boost Mac sales, let's use M4 to drive iPad sales.

But in terms of the performance improvement of M4, whether this intention can be achieved may be a big question mark.

3、 The "gap" that Apple cannot bridge

For M3 and M4 chips, their current situation is the same as the challenges faced by everyone in the chip industry. For example, process limitations, physical laws limitations, and even mask limitations. And the current situation of these chips reminds us of Apple's self-developed baseband curse.

From the earliest revelation of its possible debut in 2023 to the earliest revealed debut in 2028 at the end of last year, in the blink of an eye, Apple has spent nearly 10 years pondering whether it can replace the current Qualcomm baseband after investing billions or even tens of billions of dollars in self-developed basebands.

There is also the so-called self-developed wireless chip. When news came out in 2021, the stock prices of companies such as Broadcom and Skyworks Solutions fell in response. However, for more than three years, the progress of the project was slow. At the end of last year, there were reports that Apple's wireless chip team seemed to have stopped production, and self-developed WiFi chips were unlikely to be installed on the 2025 iPhone.

At the end of last year, Bloomberg revealed that Apple also planned to develop its own image sensor because it is the core of hybrid reality and autonomous driving. At the same time, Apple's self-developed microLED screen and non-invasive blood glucose detection system are also being promoted. However, this year, two of these three plans were hindered: the microLED team was disbanded and restructured, and the US Food and Drug Administration (FDA) also explicitly opposed non-invasive blood glucose related devices.

With the fading popularity of the Vison Pro and the failure of Apple cars, the future of Apple's self-developed image sensor team may also be shrouded in a layer of gloom. Regardless of the progress of research and development, even if it is made, where can it be used?

The success of the A-series chip has filled Apple with confidence in the chip, while the success of the M-series chip has further expanded Apple's confidence. Apple may feel that it is just a few square centimeters of silicon chip, and as long as it is willing to invest a lot of money, it is as simple as exploring things. Coupled with years of confrontation with Qualcomm before 2019, it has also strengthened its idea of holding the chip in its own hands.

But contrary to expectations, Apple still only has the A and M series of chips that can be considered successful, and even the latter is struggling with the continuous decline in Mac sales. Now, let the iPad be the last straw in life.

What was the significance of reducing costs for self-developed 3nm chips, which cost billions of dollars just for chip fabrication? If there were no corresponding products with high sales to dilute this cost? It's not enough to choose self-developed products just for the sake of a high-end brand. If the sales of the M4 iPad Pro are still not satisfactory, which category should Apple look for as a rescue?

Perhaps at this point, it can be said that even an apple is powerless.

This article is from WeChat official account: Semiconductor Industry Watch (ID: icbank), written by Shao Yiqi

This content is the author's independent viewpoint and does not represent the Tiger Smell position. Reproduction is not allowed without permission. For authorization matters, please contact us hezuo@huxiu.com

If you have any objections or complaints about this manuscript, please contact us tougao@huxiu.com

People who are changing and want to change the world are all on the Tiger Smell app

appreciate

苹果芯片，也没办法

半导体行业观察关注