Author: Chen Junyi
Editor: Zhang Yuzhe
Behind the release of a brand new high-performance processor is a game on the chip battlefield.
Qualcomm released its latest flagship processor in early December - Snapdragon 8 Gen 1. This processor is based on Samsung's 4-nanometer process. The instruction set is upgraded from ARM v8 to ARM v9, which brings a new upgrade to components such as CPU, GPU, NPU and 5G baseband. The CPU performance is improved by 20% and the power consumption is reduced by 30%, the GPU performance is improved by 30% and the power consumption is reduced by 25%, the AI computing power is improved by 4 times, and the paper data performance is improved by very obvious.
mobile phone manufacturers have also announced plans to install new phones with Snapdragon 8 Gen 1. Xiaomi and Lenovo Motorola even fought a verbal battle on who first launched the Snapdragon 8 Gen 1.
Source: Qualcomm official
The previous generation of Snapdragon 888 processors using Samsung's 5-nanometer process showed a power consumption increase that does not match the performance improvement range when comparing the previous generation of Snapdragon 865 processors. In the evaluation of many digital channels such as Xiaobai Evaluation, Geek Bay, Yizhi Mushroom, etc., Snapdragon 888 has higher power consumption than Snapdragon 865, while Snapdragon 865, which shows the advantage in performance power consumption, uses TSMC's 7-nanometer process.
After the Snapdragon 888's power consumption performance "overturned", why did Qualcomm continue to choose Samsung instead of turning to TSMC's foundry? While Qualcomm and Samsung are deeply bound, why does Apple bet on all chip foundry chips? What impact will Qualcomm and Apple take on chip foundry have on the chip manufacturing field?
Qualcomm gave up TSMC, which is not the first time
As a chip design company, Qualcomm does not have its own chip manufacturing factory, and the advanced process foundry that can be selected in the industry. In fact, there are only two companies, Samsung and TSMC.
If the processor receives a lot of criticism from users for power consumption and heating problems, it will often change its foundry in the next generation of processors.
For example, the Snapdragon 810 released in April 2014 uses a large-core CPU architecture of quad-core Cortex-A57 + quad-core Cortex-A53, and is manufactured based on TSMC's 20-nanometer process. However, due to the huge power consumption and serious heat generation, mobile phones using Snapdragon 810 are generally not recognized by consumers. The Samsung Exynos 7420 processor released at the same time uses Samsung's first generation 14-nanometer process, which instead avoids the problem of excessive power consumption and helps Samsung further consolidate its market position in Android phones.
After this battle, Qualcomm gave up TSMC in the new generation of flagship processors born after Snapdragon 810, and chose Samsung for three consecutive years:
016's flagship processor Snapdragon 820/821 adopts Samsung's improved version of 14-nanometer LPP process;
017's flagship processor Snapdragon 835 adopts Samsung's 10-nanometer LPE process;
018's flagship processor Snapdragon 845 continues to use Samsung's 10-nanometer LPP process.
At the same time, Qualcomm only some mid- and low-end processors continued to use TSMC foundry, such as Snapdragon 653, Snapdragon 435, Snapdragon 439, etc.
In 2018, since Samsung's 7-nanometer process progressed faster than TSMC, the Qualcomm flagship processor Snapdragon 855 released in December of that year gave up Samsung's 7-nanometer N7 process (TSMC's 7-nanometer process has three versions, one is the first generation N7 that uses DUV technology, the second generation N7P that uses DUV technology, and the third is the introduction of EUV extreme ultraviolet lithography N7+).
In December 2019, Qualcomm's flagship processor Snapdragon 865 was continued to be handed over to TSMC's foundry and manufactured using TSMC's N7P process.
But Qualcomm has not forgotten Samsung either. Even if Qualcomm's 5G chip Snapdragon SDM7250 was completely scrapped due to Samsung's 7-nanometer process technology problems in 2019, Qualcomm still continues to cooperate with Samsung in foundry, and released the Snapdragon 888 in December 2020, once again abandoning TSMC and switching to Samsung's 5-nanometer process. Although
is not easy to compare directly, according to public information, Samsung's 5-nanometer process does lag behind TSMC in terms of transistor density and other parameters.
Qualcomm's choice this time is not so much a way to give up TSMC, but rather to say that TSMC has been "intercepted" by larger customers such as Apple because of its advanced technology. For major customers like Apple, TSMC not only has a tilt in production capacity, but also has preferential prices.
In 2021, TSMC informed customers of the news of the price increase of advanced process technology, but the prices for customers such as Qualcomm and AMD rose by 20%, while the prices for Apple rose by less than 5%. Qualcomm and AMD are therefore dissatisfied with TSMC's special treatment to Apple. It is reported that Qualcomm and AMD are both planning to transfer some chip foundry orders to Samsung to reduce their dependence on TSMC.
The only chip foundry field TSMC is obviously not conducive to upstream chip design companies. Then, when one party’s advanced processes are leading, the mid- and low-end chips are still handed over to the other party’s foundry, and even when one party’s advanced processes do not have the leading advantage, the flagship chips are handed over to the foundry. For example, supporting Samsung as the "second supply" will also help Qualcomm have more chip foundry companies in the future as a choice.
But another question is, why is Apple not worried about TSMC's store bullying customers? Even if the price increase in OEM is lower than that of Qualcomm and other companies, compared with TSMC, does Apple have the ability to bully stores?
Apple was also "Aquaman"
Apple has always adhered to the multi-supplier strategy. It will choose multiple suppliers in terms of foundry assembly, screens, batteries and other links or components, but in the field of chip foundry, Apple is more specialized. Before
iPhone 4, Apple used a third-party processor. Since the iPhone 4 was equipped with its first A-series processor, A4, in 2010, it has been manufactured by Samsung. It was not until the A8 processor released in 2014 that Apple switched to TSMC and was manufactured using TSMC's 20-nanometer process.
The A9 chip that Apple released in September 2015 was equipped with Apple's A9 chip, which was produced by Samsung and TSMC for the first time. To produce this chip, Samsung uses 14nm 14LPE technology, while TSMC uses 16nm 16FF+ technology.
Some reviews pointed out that there are obvious power consumption differences between different A9 chips produced by Samsung and TSMC, and the maximum power consumption of Samsung version will exceed 20% of the TSMC version.
In response to this situation, Apple explained in an official statement sent to the US technology website Techcrunch: "Some laboratories keep processors at high load until power is exhausted, which is not in line with the real life situation. Our test data and customer data show that the difference in real battery life of iPhone 6s and 6s Plus using all different parts is only between 2-3%. "
In Apple's statement, although the maximum difference in battery life of different versions of chips is only 3%, consumers do not agree. Finally, Apple removed the software that can detect whether the processor is TSMC or Samsung foundry, and starting from the A10 series chips, Apple's A series processors are exclusively produced by TSMC.
How deep is the binding between TSMC and Apple? In addition to the A-series processors, TSMC has formed an exclusive R&D team of more than 300 people to cooperate in depth with Apple to develop the M-series processors of Apple computers.
For Apple, Samsung has multiple identities. This Korean giant is not only a chip foundry, but also a world-leading mobile phone giant. TSMC is a chip foundry that has never had a direct competition with Apple.
completely handed over the chip to TSMC foundry, which not only saved the additional cost of the two foundries jointly producing a chip in the A9 chip era, but also was able to use the most advanced process technology earlier than competitors with their exclusive advantages.
For Qualcomm, sometimes choosing TSMC and sometimes choosing Samsung, you can choose the best solution at that time based on the advancedness, cost, and coordination factors of both chip processes. Qualcomm's "Aquaman Strategy" seems to be more flexible.
Source: Qualcomm official website
But at present, because the deeper the TSMC is bound to Apple, Qualcomm has a trend of completely betting on Samsung OEM and no longer "Aquaman".The greater the advantage of Apple phones over Android phones, the more unfavorable it is to Qualcomm's development.
Moreover, Apple plans to use TSMC's 4-nanometer process to produce customized 5G baseband chips for iPhones starting from 2023, replacing the Qualcomm X60 baseband integrated in the current A15 processor.
After resolving a patent lawsuit in 2019, Qualcomm's competition with Apple continues. In the highly competitive chip track, the more deeply TSMC is bound to Apple, the less it is to be accepted by Qualcomm. Qualcomm will choose to hand over the flagship chip to Samsung for production, and some mid- and low-end chips to Qualcomm for production, and in the end, both parties will not form each other's major customers.
For Apple, the "two-ship" strategy of flagship chip A9, which was once jointly produced by two major chip foundry companies, has long been proven invalid and is prone to controversy. Focusing on TSMC, forming deep binding, also brings benefits to Apple's current development.
The two strong men competed, who could benefit?
Qualcomm is deeply bound to Samsung, Apple is deeply bound to TSMC, and the two strong groups have opened up the battle.
TSMC is still absolutely leading in revenue scale. According to TrendForce data, in the third quarter of 2021, Samsung Electronics' OEM sales increased by 11.0% month-on-month to US$4.81 billion, continuing to rank second; TSMC's sales increased by 11.9% month-on-month to US$14.884 billion, accounting for 53.1%, firmly ranking first. In addition, the 7-nanometer and 5-nanometer processes contributed more than half of TSMC's overall revenue, and this proportion is still growing.
In the competition for advanced processes, TSMC is ahead of the 7-nanometer process, while both sides are almost moving forward in the 5-nanometer process, and both sides have plans for the more advanced 4-nanometer and 3-nanometer processes.
Samsung's 4nm process has been mass-produced and applied to Qualcomm's latest Snapdragon 8 Gen 1 processor. In addition, Samsung is also considering establishing a state-of-the-art 3nm chip factory in the United States. It is reported that Samsung will start installing major equipment from 2022 and start operating in 2023.
Due to the advantages of TSMC in 5-nanometer process to Samsung, the industry also has higher expectations for TSMC in 4-nanometer process. It is reported that Qualcomm may launch the Snapdragon 8 Gen 1+ flagship in 2022 and adopt TSMC's 4-nanometer process.
However, the plus version of the same generation of processors has not happened in Qualcomm's previous history. Snapdragon 865, Snapdragon 865+ and Snapdragon 870 are all made using the same foundry and process (TSMC N7P). Therefore, the reliability of this news is still questionable. Qualcomm flagship processors will most likely not replace the foundry within the time period of at least one generation.
TSMC's 3nm process is planned to be mass-produced in the second half of 2022. By then, the company's monthly production capacity of related processes will start at 55,000 pieces, and will reach 105,000 pieces in 2023. According to the disclosed information, TSMC's 3nm still uses FinFET (finfield effect), and GAA (full-surround gate architecture) will only be used at N2 technology nodes.
It is reported that Samsung has officially used GAA's 3-nanometer process technology to cut the flow, with the logic area efficiency increased by more than 35%, power consumption reduced by 50%, and performance improved by about 30%, which is better than TSMC's 3-nanometer process in terms of performance.
Samsung, whose performance and power consumption are slightly weaker than TSMC in the 5-nanometer era, may catch up with TSMC in the 4-nanometer and 3-nanometer era. The market structure of chasing each other will also have more process routes for chip design companies to choose from, and even attract more entrants.
Intel is preparing to open its chip foundry business again. The company's IDM 2.0 plan announced in March 2021 shows that from 2023, some consumer-grade CPUs and data center CPUs will be handed over to third-party foundries such as Samsung and TSMC, and an independent "chip foundry" department will also be established to cover various IPs such as x86, ARM, and RISC-V.
In other words, Intel no longer insists on the vertical integration from chip design to chip drilling and manufacturing, but will not only hand over some chips to others for foundry, but will also help others to find various IP chips.
When it comes to handing over its own chip to someone else's foundry, Intel and TSMC seem to be more optimistic about their cooperation. On December 13, Intel CEO Kirsinger also flew to Taiwan on a private plane to visit TSMC and other companies, and told the outside world that the partnership between TSMC and Intel is very far-reaching. TSMC has assisted Intel in many technologies, thus releasing greater wafer production capacity and creating unprecedented products.
In terms of OEM for others, Kirsinger once said that Apple is a potential customer they particularly want to develop.
But in the Mac product line, Apple is giving up Intel chips, and it can be foreseen that all Mac products in the future will no longer use Intel processors. In the field of chip foundry, Apple seems to have no reason to give up TSMC and choose Intel. For a long time, when Intel is in the market leading position, it tends to use its own advanced processes to consolidate its chip advantages; and when Intel is no longer in the market leading position, its process technology often no longer has a leading advantage in the market.
Intel's attempts in advanced processes in recent years have not been successful. The company's 14-nanometer process is continuously used in multi-generation processors, but after 2014, more and more "+" numbers have been criticized by consumers for their process upgrades. At the same time, the company's 10-nanometer process was not able to be successfully mass-produced until the 11th generation mobile version of Core processor, but there is a gap compared with the TSMC's 5-nanometer process at the same time. TSMC's advanced manufacturing process is still progressing. Based on the current 5nm process node, the N4P process was launched in October this year, with performance being 11% higher than the earliest N5 process and 6% higher than the N4 process. On the morning of December 16, after N4P, TSMC announced the launch of N4X process technology, with performance 15% higher than N5 and 4% higher than N4P. It is expected to enter risk production in the first half of 2023.
In 2010, Intel also founded external chips. However, since Intel wants to get a share of both PC processors and mobile processors, it is difficult to reach a cooperation intention when using its advanced processes to provide wafer foundry business to possible competitors. At the same time, Intel's production capacity must always give priority to ensuring its own chip production, which can provide insufficient production capacity to the outside world.
Under multiple factors, in 2018, Intel announced that it would "re-concentrate manufacturing resources on its own products" and basically gave up its foundry business. Until 2021, the company proposed the IDM 2.0 plan again.
The market is not optimistic about Intel's opening of its foundry business again. But in addition to TSMC and Samsung, there is also a large demand for more technical foundries to enter the market.
In the battle between Qualcomm and Apple, the production capacity of chip foundry is also deeply bound, and its exclusivity becomes stronger. In addition to TSMC and Samsung, more other chip foundries may also gain their own prey in the battlefield of competition between the two strongest people.
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-