Samsung and Qualcomm have once again set off a "self-developed architecture" war, is it the key to success or a death throes?

According to a report by market research institute Counterpoint Research, in the context of the overall decline of the global smartphone market in 2022, high-end smartphone sales of more than $600 increased by 1%. But among them, Apple is still the biggest winner, accounting for 75% of the high-end market share and 41.25% of the total revenue of the global smartphone market.

Why is the iPhone so successful? There are many reasons, but at least the industry generally agrees that the success of the iPhone is inseparable from Apple's self-developed chips, and even self-developed CPUs and GPU architectures. Looking around today's mobile market, Apple is also the only chip manufacturer that has implemented its own CPU architecture.

Photo/Apple

In fact, a few years ago, Qualcomm and Samsung also developed their own CPU architectures - Krait (ringsnake) / Kryo and Mongoose (mongoose), but then both abandoned self-research and turned to the Arm public CPU architecture. However, in the past two years, Qualcomm has clarified the direction of its self-developed CPU architecture, and even did not hesitate to go to court with ARM for this, and Samsung has the same idea.

According to the Korea Business Daily, Samsung Electronics formed a CPU core development group within the company and hired former AMD senior developer Rahul Tuli to lead the group. If the development process goes well, Samsung Electronics will be able to use CPUs developed with its own core architecture by 2027.

From abandonment to restart, what has changed in the Qualcomm, Samsung and mobile chip industry?

From onboarding to abandonment to rebooting

In 2015, in order to deal with Qualcomm's "ring snake" architecture, Samsung named the newly launched CPU large-core architecture "Mongoose". As the natural enemy of the "ring snake" in the biological sense, the first appearance of "Mongoose" - M1 did bring a great surprise, directly helping the Samsung Exynos processor to beat the Snapdragon 810 (the original fire dragon) and Arm public version of the same period, and won the reputation of "Android Light".

M1 architecture, Fig/Samsung

However, in the following four years, Qualcomm returned to the Arm public CPU architecture, and Samsung's hundreds of billions of investment in "Mongoose" did not exchange for a leading position in technology and products. In 2018, AnandTech, a well-known review media, wrote in an article involving the Samsung Exynos 9810: "The Samsung Exynos 9810 (2018) with M3 core consumes twice as much energy as the Apple A11 (2017), but the performance lags behind by 55%."

At the end of 2019, Samsung can't stand it. Just after the release of the Exynos 990 using its self-developed large core, it was reported that Samsung would stop developing its own "mongoose" big core and return to the Arm public version of the big core, and it was officially announced in the following months. Earlier, Qualcomm was frustrated to find that the Arm public version had higher energy efficiency and better PPA (performance, power consumption, area) advantages, and abandoned its self-developed CPU architecture from the Snapdragon 835 released in 2016.

But Qualcomm, which was the first to give up, was also the first to start the idea of restarting its self-developed CPU architecture. In 2021, Qualcomm announced the acquisition of Nuvia, a chip startup founded by former Apple's chief chip architect, for $1.4 billion, almost publicly predicting that it would replace Arm's Cortex CPU with a self-developed CPU. Facts have proved that this is indeed the case, at the Snapdragon Technology Summit at the end of last year, Qualcomm officially announced a new self-developed CPU "Oryon", and is expected to see related products on the market as soon as the second half of this year.

From Qualcomm to Samsung, why have the two major chip designers given up for many years and returned to the road of self-developed CPU architecture?

Samsung and Qualcomm: Same direction, different purpose

If you have to use a conclusion to summarize the reason why Qualcomm and Samsung restart their self-developed CPU architecture, it is undoubtedly because the product is not good. More specifically, in Apple's comparison, Arm Cortex CPU is not very good, even the launch of super large cores has not changed the status quo, Qualcomm and Samsung want to break the deadlock they face, can only continue to go deep into the core architecture.

Very large core Cortex-X3, pictured / ARM

But when it comes to the two companies, there are still big differences. On Samsung's side, the core still hopes to revive the mobile phone business. According to a report by Counterpoint Research, Apple has surpassed Samsung in the fourth quarter of last year to become the world's largest smartphone manufacturer, and Samsung has almost no recovery in the Chinese market.

Although the Galaxy S23 series generation chose to carry the Snapdragon 8 Gen 2 chip on the whole line, Samsung did not intend to give up its own chip. In addition to restarting its own CPU architecture, Samsung's cooperation with AMD on mobile GPUs has not stopped, and Sungboem Park, vice president of Samsung responsible for GPU development, has made it clear that it will continue to cooperate with AMD.

AMD CEO Su Zifeng at the Samsung press conference, photo/Samsung

By the end of last year, it was also reported that Samsung will work with the AMD Radeon team and Google Tensor team to jointly develop a new mobile SoC, which is expected to be launched exclusively in 2025. Samsung Electronics President and MX business leader Lu Taiwen also revealed a few days ago that Samsung Electronics is developing exclusive custom SoCs for its Galaxy mobile phone product line, and said that this chip will be:

Unlike any other product on the market, it is "unique".

One of the benefits of self-developed chip core lies in the integration of the underlying hardware layer, while controlling the product cycle, customizing the chip design according to needs, and better adapting to the software ecology. This is not only the key to Apple's core experience, but also the main driving force for more and more mobile phone manufacturers to invest in self-developed chips. A deeper self-developed CPU/GPU architecture can make greater breakthroughs in core differentiation.

Sheng Linghai, research director of Garner, also said: "Self-developed cores can enable chip manufacturers to reduce their dependence on public designs when developing processors, and have greater autonomy and differentiation."

However, limited by the Arm public version architecture, Qualcomm has seen Apple's expanding advantages in the PC and high-end mobile phone markets in recent years, but it does not have much to do.

On the PC, thanks to the low power consumption and high performance of the M series chip, the MacBook has greatly improved the performance of heat dissipation, battery life, noise and other aspects while improving the performance of performance, almost riding in the PC winter, constantly increasing Apple's market share.

MacBook Air doesn't need a fan anymore, pictured by iFixit

In the high-end mobile phone market, Apple continues to consolidate its dominance, accounting for three-quarters of the global share last year. There is no doubt that the Android camp's move to hit the high-end and the poor performance of the Snapdragon 888 and 8 Gen 1 not only did not challenge Apple, but even deepened consumers' recognition of Apple in the high-end market to a certain extent.

It's just that the A16 generation has encountered its own mistakes (related to mobile optical tracing technical support, see our previous report for details), coupled with the significant improvement brought by Qualcomm's switch to TSMC's 3nm process + new architecture, it can be narrowed the gap.

However, the more essential problem has not been solved, Qualcomm wants to help the Android camp win more high-end market share, must come up with more energy efficiency and performance of better products, since the Arm public version design is difficult to meet, can only like the self-developed Adreno GPU, their own development of CPU core architecture.

Arm public version, long time must be divided

Luo Guanzhong pointed out in the first episode of the opening chapter of "Romance of the Three Kingdoms": The general trend of the world, divided for a long time, must be united for a long time, and a long time must be divided.

In the "world" of smartphone chips, Qualcomm, Samsung, MediaTek, Huawei HiSilicon, UNISOC Zhanrui, and even the early Apple A series chips have chosen Arm's public version architecture, even if Qualcomm and Samsung's "short defection" has never affected the status of Arm public version.

Figure/ARM

However, as the market size of smartphones peaks, mobile phone manufacturers need to establish a larger and longer-term competitive advantage, and have to go deep into the underlying chips, systems, etc., and even personally create self-developed SoCs. As a result, chip manufacturers are actually forced to reconsider whether they can create more long-term competitive product advantages under the limitations of the public version architecture.

Once the answer is clear, chip manufacturers will naturally choose to break the "shackles" of the public version and recreate more competitive CPU architectures and products.

Of course, we must also understand that Arm's public architecture is both a "constraint" and a "protection", and globally, there are only a few that have achieved large-scale success based on self-developed CPU architectures. So for Qualcomm and Samsung, it is still just a few key steps, and the remaining road is still far away.

Image from Pexels.