The Semiconductor World Has Changed

This year is 2024, nearly 70 years from the invention of the first integrated circuit, the semiconductor industry has undergone several changes, the market ups and downs, from design foundry, to equipment materials, it seems that there is no eternal leader.

Looking back to the 1960s and 1970s, the PC market was still firmly in the hands of IBM, although personal PCS like Apple were recognized by the consumer end, IBM has a strong technical advantage, and has an undisputed notification status in the commercial field, but the subsequent rise of Intel has replaced it with x86 architecture, becoming the new hegemon.

Although the semiconductor market is often calculated in five or even ten years, its development speed and the change of the technical wind direction are far beyond many people's expectations, and several major players in the market may have experienced multiple cycles, but even they do not dare to say that they have mastered the future of semiconductors.

In just the past 10 years, new giants have emerged. Will they take over? Or will it be a note in the history of semiconductors?

The rise of Arm architecture

Intel was challenged 20 years ago.

In July 2003, foreign media compared the two most famous architectures at that time - x86 and PowerPC, and they respectively chose the most representative cpus under the two architectures at that time, that is, the x86 camp of AMD Athlon XP and Intel Pentium 4. The PowerPC camp includes the IBM 750xx (G3), MOTOROLA 74xx (G4), and IBM 970 (G5).

The advantage of x86, it said, is the huge market and Microsoft's dominance, with a lot of low-cost hardware and a lot of software to run on it, and no other CPU architecture has this advantage. RISC may be technically superior, but it can only stand in a niche market due to market forces that favor the lower cost and richer software x86, and the market does not go by technology and rarely chooses the best solution.

However, the media also pointed out that the situation may be quietly changing, more and more enterprises are adopting Linux, the PC market is approaching saturation, and users do not need as powerful processors, PowerPC this architecture may not have the opportunity to change.

The author also pointed out the fatal problem of x86 - heat. He said x86 cpus were already hot and needed more heat dissipation, and a report from Microprocessor Report Publishers noted that Intel expected to start experiencing heat problems in 2004.

He believes that x86 cpus generate a lot of heat due to the need for higher performance, but because of the inefficiency of the instruction set, it will pull up energy consumption. In order to compete with each other, AMD and Intel need to keep increasing the clock frequency, running the chips to the limit, and the temperature of the chips will be higher and higher. RISC cpus like the 970 have a clear advantage in this regard, as they provide extremely competitive performance at lower power consumption without the need to push performance to the limit. Once chips are scaled down to next-generation processes, the power consumption of existing performance is reduced.

The outlet boldly asserted that things were changing in 2003 and that Linux and other operating systems were becoming more popular and that these systems were not locked into x86 or any other platform. x86 is having problems, and PowerPC looks to be emerging as a real, effective alternative to x86 cpus, with performance that can match or exceed x86 cpus, but without the increasingly important power or heat issues.

From today's point of view, the media is obviously only know one and do not know the other, x86 is facing heat dissipation problems, but PowerPC is not better, especially the PowerPC G5 he mentioned, although it is not better than Pentium 4, but the power consumption and heat of G5 is not better, with the rise of notebook computers, PowerPC, known for its low power consumption, is in trouble of its own.

As Apple turned to Intel's x86 architecture in subsequent Mac products, PowerPC began to fade out of the consumer end, which was a major setback for the RISC instruction set on the PC side, and Intel and AMD also expanded the x86 landscape to high-performance computing and data centers through the success of the consumer end. The x86 architecture introduced by Intel has been a huge success in the PC, data center and high-performance computing three areas, and Intel has successfully achieved a dominant position in the computing industry.

But Intel's big win did not last too long, with the release of the iPhone in 2007, a rival named Arm entered its sight, ARM architecture with the mobile phone this rapid development and ubiquitous device, became the new x86 architecture rival.

AMD in the rise of Arm just in the trough period, but Intel was monopolist of the three major areas of the cake, as the largest semiconductor manufacturers, naturally will not sit back and watch the rise of Arm, from 2007, Intel with ATOM processor began to expand the mobile Internet, the major PC manufacturers have responded to launch a new netbook, However, the power consumption of ATOM is too high, but it has not made much waves in the consumer market, and it is not much attractive for mobile phones and tablets.

By 2015 or so, Intel's strategy in the mobile field basically declared failure, and its defeat, but also once in the eighties and nineties unbreakable Wintel alliance appeared cracks, Microsoft and Arm reached a cooperation, Windows system is no longer bound to x86 architecture, Windows on ARM began to emerge.

However, it is important to note that Windows on ARM has undergone a very long metamorphosis.

The first public version of Windows to run on Arm processors was Windows RT, an ARM-compatible branch of Windows 8 that ran on a small number of devices in late 2012. Windows RT has significant limitations, most notably the complete inability to run traditional x86 Windows desktop applications, all of which must come from the ecologically poor Microsoft Store, and no x86 compatibility mode at all.

Part of the reason for this limitation may have been the limited and low-performance ARM hardware available at the time. ARM processors are still predominantly 32-bit, with slow processors and Gpus, 32 or 64GB of flash memory, and only 2GB of memory. Even with x86 application translation, the translated application will be poor, because ARM hardware already has difficulty running native built-in applications stably.

The demise of Windows RT was doomed, with Windows RT devices disappearing from the market around 2015, but it paved the way for what followed, as Steven Sinofsky, then head of Windows, detailed. Microsoft has done a lot of work to define the Hardware Abstraction layer (HAL), ACPI firmware, and basic class drivers for the Arm version of Windows so that the operating system can be installed and run as expected on a variety of narrowly standardized ARM hardware, just as it would on a fully standardized x86 PC.

In 2017, Windows 10 first appeared on ARM devices with support for 32-bit x86 application conversion, and while this version of Windows on ARM is more of a tech demo, it does get closer to what it needs to be to be successful with Windows on Arm: A direct replacement for the x86 version of Windows, the two versions are essentially indistinguishable to non-technical users.

The next major advance comes in 2020, when Microsoft announced a preview of 64-bit Intel Application Translation for ARM PCS, and with subsequent improvements, there are still some compatibility gaps in the current ARM version of Windows, especially in terms of external accessories and dedicated software. But the vast majority of productivity apps and even games can now run smoothly on the ARM version of Windows without user or developer intervention.

Today, the PC device equipped with Qualcomm Snapdragon X Lite is about to be listed on a large scale, which may be since 2012, ARM has the biggest harm to x86, although Apple's M series chips have appeared before, but the closed ecology of macOS determines its limited influence, but Qualcomm and even more ARM processor manufacturers are different. They will have the most direct impact on x86 Windows, eating the iron rice bowl for more than two decades, and now it seems that it is not so strong.

In addition, ARM is not limited to the consumer side, it also launched an attack on the server side of x86, in the past few years, the number of ARM servers is rapidly expanding, according to Gartner data, about 77,000 in 2020, in 2021 to 252,100. It is expected to be 540,400 units in 2022. But 934,600 units are expected to ship in 2023, 1.71 million units in 2024, 2.54 million units in 2025, and 3.2 million units in 2026.

Obviously, ARM is eating x86 server increment, which is a huge bad news for Intel and AMD, ARM may not become the mainstream of the server market in the future, but there is no doubt that it will occupy its own acres and become the choice of many OEM manufacturers.

From the consumer end to the commercial end, the two architectures of ARM and x86 are forming a competitive trend.

Hynix's reversal

Hynix was born into a rich family, but he wasn't really the chosen one.

Hynix was born in March 2001 when Hyundai Electronics merged with LG Semiconductor. In August 2001, Hynix was officially independent from the Hyundai Group as a memory semiconductor specialist, but the company was already heavily indebted and on the verge of bankruptcy.

For Hynix at the end of 2002, the external is the global economic slowdown, the semiconductor industry recession, trade friction, Micron acquisition, internal is the lack of technology and experience, organizational management after the merger, lack of funds and so on. At this time, the United States, the European Union and Japan also imposed countervailing duties on Hynix's DRAM products, and Hynix's memory business was struggling.

In order to save the company, the management of Hynix implemented four major initiatives, namely technological innovation, business adjustment, strengthening partnerships and introducing more financing, including the successful development of 0.15 micron process, significantly streamlining its business, disposal of some non-core businesses and assets, and the strategic partnership with ST for NAND flash memory development.

After a brief recovery in 2004, Hynix ran into a new dilemma: Around 2007, a global semiconductor oversupply caused DRAM prices to plummet, which also triggered the worst recession in history, which hit all DRAM vendors hard.

While Hynix continues to decline, South Korea's SK Group has stepped up. In July 2011, SK Telecom, the flagship company of SK Group, submitted a letter of intent to acquire Hynix. Through the acquisition of Hynix, SK Group added a third growth pillar - semiconductor, in addition to the two growth pillars of energy and chemical industry and information and communication technology, greatly strengthening SK's global business capabilities.

After the acquisition, SK began to invest a lot of money in Hynix. Hynix returned to profit in the second quarter of 2012, and in 2013, 2014, 2015 and 2017, SK Hynix's sales and operating profit hit a record high, and Hynix has now become one of SK's biggest cash cows.

But for Hynix, which was acquired by SK, Samsung is a mountain it cannot get around.

From the table above, we can also see that from 2016 to 2019, Samsung has maintained a share of about 45% in the DRAM market most of the time, which is far ahead. It is not only DRAM, but also has a strong dominance in the NAND market, and Samsung's countercyclical investment. Once a number of memory factories including Elpida and Qimonda, how to get rid of Samsung's influence in this field is the biggest problem faced by Hynix.

Hynix chose the corner overtaking, it cooperated with AMD to develop HBM, although at the beginning of the release did not make too much waves, but with the AI explosion at the end of 2022, HBM quickly became the fastest growing product in the memory market in recent years, and Hynix took the lead in mass production of HBM3 in October 2021, far more than Samsung in this technology. It has also firmly grasped the heart of Nvidia and become the largest supplier in the HBM market at present.

However, Samsung has made a mistake in its judgment on the technical route, and its choice of TC NCF is far less stable than Hynix's MR-MUF, according to analysts, Samsung's HBM3 chip production yield is about 10% to 20%, while SK Hynix's HBM3 yield can reach 60% to 70%.

According to Merits Securities data, in the first quarter of this year, SK Hynix accounted for 59% of the HBM market share, while Samsung Electronics accounted for 37%, if Samsung can not take the time to solve their own problems in HBM technology, the name of the future memory hegemon may give way to another manufacturer.

Samsung has been proud of the memory market for more than 30 years, but now it has fallen on a small HBM, and the pattern of the DRAM market has the chance to be rewritten, not so much because of fate, it is better to say that it does not really grasp the wind of the semiconductor market.

But for Hynix, which was acquired by SK, Samsung is a mountain it cannot get around.

Write at the end

Who can control the future of the semiconductor market? Intel didn't do it with processors, Samsung didn't do it with memory, and in other segments of semiconductors, no one else can do it.

For example, the most critical lithography machine in semiconductor equipment, when Nikon and Canon, the two Japanese manufacturers, mastered the lithography market in the last century, I am afraid that they did not expect that due to their own judgment of the infiltration technology, the Dutch ASML will replace them in the future with the momentum of crushing.

Another example is the automotive computing chip market, which used to be firmly grasped by traditional manufacturers such as NXP, Renesas and TI, and they reached a seemingly unbreakable alliance with domestic auto manufacturers, but the rise of new energy vehicles easily tore open the gap on this line of defense, Qualcomm, Nvidia and even more manufacturers' chips appeared in the car. The dominance of established players no longer exists.

Although we cannot predict which semiconductor manufacturer will rise and become the new hegemon in the future, only one thing can be confirmed that sticking to the rules will only lead to the end of decline, and only those manufacturers who are committed to innovation and constantly try to really laugh at the end.