Overtake TSMC in five years?
Overtake TSMC in five years?

Overtake TSMC in five years?

When Samsung finally solves the 3nm yield problem, it will remember the afternoon it unveiled the world's first 14nm mobile processor.

In June last year, Samsung announced the official production of 3nm chips, becoming the only semiconductor manufacturer in the world to adopt the GAA (Gate-All-Around) architecture transistor technology and provide 3nm process foundry services.

Some people say that Samsung once again went ahead of Intel and Taiwan Semiconductor to mass produce 3nm, but also once again proved the strength of the Korean semiconductor industry, 4nm and 5nm a small failure will not hinder the overall situation.

However, after the simple good news was released, the bad news came one after another: some media reported that the production yield was only too low, and the initial batch production of the chip was small in scale, and no big customers were willing to pay for it in a short time. The so-called mass production is equivalent to a loss of money, compared with its old rival Taiwan Semiconductor Manufacturing Co., LTD.

In the face of external doubts, Samsung has been very low profile, until recently said that the improved 3nm process has reached 60% to 70% of the product yield, in the latest roadmap, Samsung plans to mass produce the second generation of 3nm process (3GAP) next year, and 2nm process (SF2) in 2025.

图片无替代文字
Samsung's future technological path development

But this has also led to more concerns, Taiwan Semiconductor Manufacturing is about to mass produce 3nm Apple chips, Samsung has been slow to come up with a plan to mass produce 3nm chips, its sincerity to meet customer expectations? Can the second generation 3nm process really beat the TSMC N3? Will big customers like Qualcomm, AMD, and Nvidia really change their minds on the 3nm node?

It is the most prudent and reasonable choice for Samsung semiconductor division to continue to improve the existing 4nm technology and skip the battle of 3nm in 2023, but to avoid this approach completely lost the courage of the original counter-cyclical investment.

Samsung 14nm - The beginning of the corner overtaking

"Samsung's advanced 14-nanometer FinFET process technology is undoubtedly the most advanced logic process technology in the industry," said Samsung's Executive Vice President of Sales and Marketing for Systems LSI. "We expect the production of our 14-nanometer mobile application processor to further enhance the performance of flagship smartphones, thereby positively influencing the development of the mobile industry."

Reading between the lines of the press release, the Samsung division leader's pride was evident in the fact that while the likes of Qualcomm, Apple, and Mediatek were still using TSMC's 20nm technology, their Exynos processor was able to beat their flagship mobile processors by a year, both in performance and energy consumption ratio.

It is natural that Samsung s 14nm technology was the source of his confidence, but the technology was hard won.

Also as the four Asian tigers, Singapore, China's Taiwan and South Korea, when the semiconductor industry is booming, they all choose to vigorously introduce and local support, a large number of American and European semiconductor manufacturers flocked to set up factories, after undertaking this part of the industry, some excellent local manufacturers also take advantage of the situation, such as Taiwan Semiconductor Manufacturing Company of Taiwan, Samsung Electronics of South Korea, It's the best of a handful of local manufacturers.

While Samsung Electronics prospered by betting big on memory, Taiwan Semiconductor Technology Co., LTD. (TSMC) surged ahead in wafer manufacturing. The two companies took two different paths in their early days.

With the bankruptcy of Germany's Qimonda and Japan's Elpida, Samsung Semiconductor has hit the ceiling in the memory market. Just at the time of the rise of smartphones, Taiwan Semiconductor manufacturers, led by Taiwan Semiconductor Manufacturing Co., are shining, relying on wafer manufacturing to make a lot of money. How can you not be impressed?

In 2005, when the global storage market plunged, Samsung took advantage of the trend to reduce losses in the semiconductor division. In this year, Samsung officially started the wafer foundry business.

However, Samsung has always been half way behind Taiwan Semiconductor Manufacturing Co., LTD., which is not only the first manufacturer to use immersion lithography to produce 90nm chips, but also the first manufacturer in the world to introduce 28nm general process technology. In the key node of 28nm, Taiwan Semiconductor has occupied the first place in the market for many consecutive years. Some of the better-known high-end processors in the early days of smartphones, such as Qualcomm's Snapdragon 800 and Mediatek's MT6589T, use Taiwan Semiconductor's 28nm process.

Although Samsung has its own fabs, it is far worse than Taiwan Semiconductor Manufacturing Co., LTD., which was originally a contract manufacturer. In 2005, Samsung took the order of CDMA chips from Qualcomm from 90nm, but at this time, Taiwan Semiconductor has already started the trial production of 65nm chips. From 2005 to 2009, the revenue of Samsung's contract manufacturing business never exceeded $400 million, while that of Taiwan Semiconductor Manufacturing was already close to $10 billion.

However, Samsung's contract manufacturing division was not without success. During this period, Samsung turned to Jobs to make various chips for Apple, starting with the flash memory of the iPod Shuffle in 2005, including the iPhone launched in 2007, which was equipped with the Samsung S5L8900 and used Samsung's 90nm manufacturing process. Three years later, on iPhone 4, although Apple introduced the first self-developed chip A4, its core layout was very similar to that of Samsung S5PC110, and 45nm technology was adopted from Samsung, including the later A5, A5X, A6, A6X and A7, which were all contracted by Samsung."

图片无替代文字
Aplle of “M1‘

The global popularity of iPhone not only made Apple set one record after another, but also made Samsung's ranking in the wafer foundry market soaring, rising rapidly from a dozen in 2007 to the fourth in 2012. The foundry revenue was 3.439 billion dollars, only lagging behind Taiwan Semiconductor Manufacturing Co., LTD., and United Microelectronics Co., LTD., becoming a pivotal role in the foundry market. There would be no Samsung wafer foundry business today.

As a huge consumer electronics group, Samsung Electronics not only manufactures chips for Apple, but also supplies flash memory, storage, screens and other components for Apple. Samsung's mobile phones are also competing with iPhone in the smartphone market. Samsung, both in front of and behind the scenes, is not the most ideal partner. Apple is also wary of HTC's failure to supply AMOLED screens to Samsung.

Moreover, at that time, Taiwan Semiconductor Manufacturing Co., the leader in wafer contract manufacturing, not only first developed the 28nm process in 2011, but also rapidly advanced to 20nm process. Apple, in pursuit of performance and power consumption, hit it off with Taiwan Semiconductor. The news sent Samsung's share price soaring

Samsung is not completely unaware of the outside world's news, since 2011, Apple and Taiwan Semiconductor has been in a secret relationship, all day long, as a contract manufacturer of Apple chips, Samsung after knowing this news, also in the reception of stock analysts put out a stern message: as long as Taiwan Semiconductor dare to make (Apple chips), it will dare to Sue.

In the face of these difficulties, Taiwan Semiconductor Manufacturing Co., the world's largest foundry, has gone to great lengths to woo Apple.

In order to solve the problem of insufficient production capacity after receiving orders from Apple, Taiwan Semiconductor Manufacturing Co., LTD invested 9 billion US dollars to build new and expanded production lines. The construction of Taiwan Semiconductor Manufacturing Co., LTD. 15 plant in Taichung began on July 16, 2010, and was completed in early 2012. In the first half of 2013 and the second half of 2013, the company began to expand Zhuke 12 plant and Nanke 14 plant, among which Zhuke 12 plant opened an "Apple-exclusive" 20nm A8 chip R&D production line, and dispatched a large number of production line engineers from Nanke 14 plant for cooperation.

In response to the chip IP dispute between Samsung and Apple, TSMC sent a "One Team" of nearly 100 people to Apple's headquarters in the United States at the end of 2011 to help Apple solve the design and patent certification problems of A6 chip. It also presented two versions of A8 plan for Apple to choose before 2014, so as to minimize the risk of IP litigation between Apple and Samsung.

In the end, the silk screen printing of A8 chip in iPhone 6 and iPhone 6 Plus appeared in Apple's press conference in 2014 was quietly changed from Samsung to TSMC. TSMC finally squeezed out Samsung and enjoyed the hundreds of millions of mobile phone chip orders alone.

Samsung saw that it could not retain Apple any more, and the lawsuit could not contain TSMC in the short term. So it had another plan, which was to accept the suggestion of Liang Mengsong, a veteran R&D veteran of TSMC, to leap over the 20nm and 16nm nodes after 28nm and sprint to 14nm nodes. Under the condition that the size of traditional planar field effect tubes was reduced to the limit. To fully promote the FinFET technology, the field effect transistor to achieve three-dimensional, impossible, and finally realize the curve overtaking.

In fact, Samsung has been planning for wafer contract manufacturing for a long time. In the top management decision-making meeting in 2009, Samsung secretly approved a plan to Kill Taiwan (kill Taiwan), and one of the targets of this plan is Taiwan Semiconductor Manufacturing Co., LTD.

According to media leaks, since 2009, Samsung has poached 50 middle and high-level R&D executives of Taiwan Semiconductor Manufacturing Company. In order to strengthen the strength of wafer foundry, Samsung has also locked the talent of the world's top semiconductor companies. A person familiar with Samsung pointed out that "Samsung will first make a three-year contract to recruit employees, ensuring that the salary will be at least twice of the original company. We will renegotiate the contract depending on the performance."

The arrival of a large number of semiconductor talents and the influx of massive research and development funds finally allowed Samsung to complete the curve that others thought was almost impossible to overtake. When TSMC was still excited about the mass production of 20nm in early 2014, Samsung made its own big move -- the end of 2014 officially announced the production of 14nm FinFET process.

When the Exynos 7420 based on 14nm was released and officially launched, not only Taiwan Semiconductor and Apple, which had just reached cooperation, but also all the wafer foundries were stunned. Everyone was still struggling on 20nm, but the South Korean chaebol, which had not been in business for a long time, produced 14nm nearly half a year in advance. Could it be that Intel and Taiwan Semiconductor made up for Samsung in advance?

The DRAM industry has long been familiar with this. Samsung Semiconductor used counter-cyclical investment to overtake the red Sea corner of the DRAM market and launched the world's first 64Mb DRAM in 1992. It also used counter-cyclical investment to survive Qimonda and Elpida.

Not only that, starting with Samsung's 14nm, the most brutal shakeout in wafer fabrication history has begun: if you can fix FinFET, you'll get hot and dirty, if you can't fix it, you won't use 28nm and older processes.

Samsung 5nm - Radical resulting in dispirit

At least a generation ahead of TSMC's 20nm technology, Samsung has managed to attract the attention of a number of Fabless manufacturers, including Apple.

In 2014, the founder of Taiwan Semiconductor Manufacturing Co., Ltd. came out again and prepared the Nighthawn plan, known as "one hundred thousand young liver, one liver to save Taiwan". Elite engineers were selected to speed up the research and development of 10nm manufacturing process in the way of three shifts in 24 hours.

However, the lag in FinFET has become a foregone conclusion. What is more fatal is that the 20nm launched in 2015, the flagship Qualcomm Snapdragon 810 processor, has completely exposed the shortcomings in the process, whether it is heating or power consumption, has reached a shocking point. Other than Samsung Galaxy S6 series with Exynos 7420 and Note 5, all major Android manufacturers had crashes this year.

And Apple also began to make its own calculation: although it broke up with Samsung, in retrospect, Taiwan Semiconductor is not as good as I thought. If the price is right, it is not impossible to make up with Samsung.

Therefore, for the iPhone 6s series, Apple used two processors for the first time. The A9 processor used 14nm technology of Samsung and 16nm technology of Taiwan Semiconductor Manufacturing Co., LTD., which is the same reason that the iPhone 12 used multiple screen suppliers in the future. It not only ensured the stability of supply, but also took the initiative in bargaining in this way. Minimize costs.

However, in iPhone 6s, the difference in performance and power consumption between the different chips of Samsung and Taiwan Semiconductor still caused a problem. Many users and websites reported that the battery life of the iPhone 6s version of Samsung was significantly weaker than that of Taiwan Semiconductor, which led to a rush to return Samsung and buy Taiwan Semiconductor that year. There are still plenty of people out there who can tell the difference between chip versions.

Is Samsung's 14nm process better than TSMC's 16nm process? Most people would probably prefer the latter, and Apple is no exception, having firmly sided with Taiwan Semiconductor as the sole supplier of its processors.

However, by this time, Samsung has thoroughly proved its strength in wafer foundry through 14nm, and part of the performance and power gap can be completely erased by relatively cheap foundry price. Meanwhile, processor manufacturers such as Qualcomm also offered an olive branch to Samsung in 2016, handing over its flagship processor Snapdragon 820 to Samsung Foundry, and a large number of manufacturers began to recognize Samsung's wafer foundry.

As for why TSMC's 16nm technology is no less or even better than Samsung's 14nm technology, the answer emerged in 2017: In this year, Qualcomm Snapdragon 835 uses Samsung 10nm technology, while Apple A10 at the same time still uses Taiwan Semiconductor 16nm technology, in addition to Apple is proud of the single core running score, energy consumption and multi-core performance are inferior to Snapdragon 835, Taiwan Semiconductor is indeed behind half a body position.

In fact, it is not difficult to find that when xxnm becomes a digital code, the best way to test the process is the actual OEM products. After 2014, the mobile phone processor began a big iteration a year, and the feedback from consumers made the advantages and disadvantages of the new process no hiding, and the tussle between Samsung and Taiwan Semiconductor has gradually moved from the background to the stage.

Since 10nm, Samsung and TSMC have been in a fierce competition to catch up with each other. In September 2018, Apple's A12 chip was introduced, using TSMC's 7nm technology; In December 2018, Qualcomm Snapdragon 855 debuted, using Samsung 7nm technology, the two chips in terms of energy consumption is not much of a gap, the two companies in 7nm almost a tie.

To secure an edge, Samsung has also aggressively introduced the EUV process at 7nm to improve production efficiency, once again ahead of Taiwan Semiconductor's first-generation 7nm DUV process.

But at the key point of 5nm, Samsung's contract manufacturing division made a fatal mistake, completely ruining the good situation that had been formed since 2015, and once again falling behind Taiwan Semiconductor Manufacturing.

What kind of mistake? That is, compared with 7nm, Samsung's 5nm is not a complete upgrade from one node to another, but an upgrade on the basis of the original 7nm. Many people pointed out that Samsung is selling sheep's head to sell dog meat, marking 6nm as 5nm, and forcibly changing its name to catch up with the progress of Taiwan Semiconductor.

In Samsung's roadmap, the 14-10-7-3 key nodes are clearly demarcated. Samsung intended to adopt a new GAA structure on 3nm to once again surpass TSMC and take the lead in technology.

图片无替代文字
Samsung roadmap

We can also get a glimpse of the specific technical details speculated from the outside. In terms of transistor density, the density of TSMC N5 process is 173.1 MTr/mm2, while that of Samsung 5LPE process is 126.5 MTr/mm2. Although they are in the same order of magnitude, there is a big gap between them. It proves from the side that Samsung's 5nm is not a complete iteration.

图片无替代文字

Some people may question, Samsung in the early stage of FinFET is still able to chase TSMC hard, took away a lot of TSMC's original customers, in the key nodes such as 10nm and 7nm, but how to suddenly drop the team at 5nm?

The reason can be seen in the market share, from 2019 to 2022, Taiwan Semiconductor has been steadily occupied half of the contract manufacturing market, while Samsung, although the second place, has been slow to narrow the gap with Taiwan Semiconductor, as of the fourth quarter of last year, the share of 58.5%, Samsung Electronics accounted for 15.8%. The market share gap was 42.7 per cent, up from 40.6 per cent in the previous quarter, and the gap has not narrowed.

图片无替代文字

In terms of R&D investment, the gap between Samsung and Taiwan Semiconductor is even more obvious. While Taiwan Semiconductor has increased its investment for several years in a row, Samsung has tightened its investment in wafer foundry, which is only about one third of that of Taiwan Semiconductor.

图片无替代文字

Add to this the fact that Samsung has a very limited customer base. TSMC has long had stable large customers such as Apple, Broadcom, Hisse, AMD, Mediatek, Nvidia, Qualcomm and Intel, while Samsung's production capacity relies on its own Exynos chip digestion, other production capacity is mainly allocated to Qualcomm, and other customers such as Nvidia, IBM, Intel and so on account for a very small proportion.

It seems that Samsung has boundless scenery in 14nm-7nm. As the only advanced process foundry, it attracts a large number of customers to place orders. In fact, it is walking a tightrope, catching up with TSMC in advanced process with about 30% of revenue and R&D funds.

In addition, the loss of Apple, the largest consumer electronics customer, means tens of billions of dollars a year down the drain. It is almost a delusion for Samsung to turn over TSMC again after 2015, and 3nm with GAA is more like the Merogu scheme proposed by Wei Yan of Shu-Han. In front of the mature 4nm and 5nm of TSMC, Only the Samsung engineers themselves know how much advantage there is in yield and transistor density.

Three - sided countercyclical failure

Although the semiconductor business of Samsung is huge, it is mainly divided into three main departments, which are memory, system LSI and wafer foundry. Memory is the embodiment of the efforts of two generations of semiconductor talents in South Korea. Since the 1990s, it has maintained the position of the bull of the memory industry, and it is also the department that Samsung is proud of to play the anti-cycle magic.

But now the storage market is still in the cold winter, advanced process setbacks, Exynos was their flagship into the cold palace, can be said to be a three-sided attack, the three semiconductor business at the same time encountered a big problem, a little careless handling, will cause a chain reaction, you know, even the most good at the storage business, Samsung is not alone, Hynix, Micron and domestic storage are still looking at the market, and if they slow down investment, the market will change hands in a blink of an eye. As for Exynos chip, its only high point goes back eight years, and it won't even be a flagship mobile chip anymore.

2023 has been an unprecedented dilemma for Samsung's semiconductor sector for more than a decade. Now, if Samsung wants to pick up the stick of counter-cyclical investment, it has to weigh the priorities of the three sectors and the priorities of the situation. Even if Samsung is a big business, it can't say how much it will spend arbitrarily.

On May 24 last year, Samsung Electronics said IT would invest 450 trillion won ($360 billion) over the next five years to accelerate the development of semiconductors, biopharmaceuticals, IT and other next-generation technologies in response to growing economic and supply shocks.

Many people may think that Samsung's wafer foundry business can finally be strong, but not so fast, the semiconductor division of 360 billion dollars, and how much of the wafer foundry money, spread over five years, whether more than Taiwan Semiconductor Manufacturing and Intel's research and development investment in wafer foundry is still open to question.

Given Samsung's previous investment in the memory business and the wafer foundry business, the situation may not be as simple as many people think, and the gap with TSMC will not be so easy to close.

图片无替代文字

In a recent speech, a senior executive of Samsung's wafer manufacturing division admitted that Taiwan Semiconductor is far ahead of Samsung in chip manufacturing and that it will take five years to catch up with and overtake Taiwan Semiconductor, as if recognizing the reality that memory is the eldest son of the company and Samsung is the second.

But the problem is, he also believes that Samsung's current GAA adoption is critical. When TSMC moves to 2nm, Samsung's lag will change, and he expects TSMC to have trouble developing GAA because of the new technology.

Samsung's semiconductor division is being too optimistic, assuming its rivals make mistakes.

图片无替代文字
Click on the image to learn more
More Models(chips),Please try to contact me
You can contact us for any chip model that is difficult to find on the market
Email: [email protected]
Joan Compte

at NCCM Company – Next Level Nonwoven Roll Technologies

1 年

Great post! I'm looking forward to seeing how you overtake TSMC in five years. It would be great to connect on Linkedin so we can stay in touch. Could you please send me an invitation so we can connect? I'm interested in meeting you at Metec2023. I look forward to hearing from you soon.

要查看或添加评论,请登录

Ajay Guo的更多文章

其他会员也浏览了