The picture comes from Canva can draw

At present, TSMC’s status as the global foundry leader seems to be unshakable.

Thirteen years ago, Zhang Zhongmou came back from the financial crisis and laid the foundation for TSMC’s rise on the key 28nm process node. In 2015, TSMC’s 16nm process technology was successfully mass-produced, and it defeated Samsung’s 14nm duel. In the next five years, TSMC made great strides, and the gap with its competitors was getting wider and wider. In 2020, defeating Samsung again in the 5nm competition has completely established TSMC’s invincible market position in the field of advanced process wafer foundry.

And winning these thrilling technology competitions, of course, has brought huge returns to TSMC. Recently, TSMC released its financial report for the fourth quarter of 2020 as of December 30. From the performance of the financial report, the steady growth of revenue and profit proves that TSMC’s advanced process technology has won wider market recognition.

It also seems to show that TSMC has once again entered a two-way virtuous circle of technology and capital brought by advanced process leadership. However, is the truth really as good as it seems?

Harvest the best quarterly report of the decade

In terms of comprehensive performance, TSMC achieved its best quarterly report in a decade in the new quarter. In dollar terms, TSMC’s revenue in Q4 2020 reached $12.68 billion, a year-on-year increase of 22%. Net profit reached 5.008 billion US dollars, a year-on-year increase of 31.7%. Revenue and profit have set new records.

Even more surprising to investors is that its fourth-quarter profit margins were exceptionally bright. The gross profit rate reached 54%, and the net profit rate reached an astonishing 39.5%, which also created a new historical record. At the same time, if compared with other technology giants, this level of net profit margin is even much higher than Apple’s 21%, and its ability to attract money is terrifying.

The key reason why TSMC was able to achieve such an excellent performance in the fourth quarter is that its advanced process technology of 16 nanometers and below contributed 62% of the revenue. After all, for TSMC, compared with mature process technology, the price of advanced process technology is higher, and the profit margin is also larger.

Among them, the contribution of 5nm is particularly prominent, and the proportion of revenue has soared from 8% in the third quarter to 20% in the fourth quarter. Now TSMC has no real rivals on the 5nm process node. Globally, only Samsung has mastered the mass production capacity of 5nm process technology except TSMC. Samsung’s mass production scale can only reach 1/5 of TSMC’s, and the product performance and cost price are not dominant. Therefore, it is certain that TSMC’s 5nm will still be able to show its prowess in the foundry market for a long time in the future.

From this point of view, there is nothing wrong with TSMC’s logic of pursuing its own core advantages on the basis of leading advanced manufacturing processes. However, when the advanced process technology is very close to the physical limit, TSMC’s continued insistence on further increasing investment in R&D and production of advanced processes has actually become a risky behavior.

Continuing to overweight advanced processes has become an adventure

At present, the performance and power consumption returns of chips brought by advanced manufacturing processes are being significantly reduced. In recent months, smartphones equipped with 5nm process technology SOC have been launched one after another. Judging from the actual performance of these mobile phones, whether it is TSMC’s 5nm FinFET process or Samsung’s 5nm LPE process, the performance and power consumption improvements have failed to meet market expectations.

In terms of TSMC, the actual performance improvement of 5nm is a little stretched. Taking Apple’s A-series processors as an example, also based on TSMC’s 7nm process, the A13 processor has a 20% increase in CPU performance and a 20% increase in GPU performance compared to the A12 processor; while the A14 based on TSMC’s 5nm process has improved CPU performance compared to the A13. It is about 16.7%, and the GPU performance improvement is about 8.3%. That is to say, on Apple’s A-series processors, the progress brought by the progress of the 5nm process may not be comparable to Apple’s own optimization and upgrade of the processor architecture.

Although some media people speculate that this is due to the low yield rate at the beginning of 5nm, Apple A14 shielded some cores. However, the Kirin 9000, which also uses TSMC’s 5nm process, also has a big difference in power consumption control compared to the official data.

In terms of Samsung, the problem of power consumption overturning is more prominent. According to the actual measurement of Xiaomi mobile phones by the well-known digital blogger Geekerwan, the Snapdragon 888 processor using Samsung’s 5nm LPE process is compared with the previous generation Snapdragon 865 processor. dramatically drop. And the Exynos 1080 chip, which also uses the 5nm LPE process, also pulls the hips in terms of energy efficiency.

The actual performance of the 5nm advanced process technology is generally not satisfactory. For products using the 5nm process at the current stage, its marketing value may far exceed the practical value of the advanced process itself.

What is even more disturbing is that TSMC continues to increase its R&D investment in the next-generation process node 3nm process when the practical value of the current TSMC 5nm process technology is in question. At the recent earnings conference, TSMC management announced that it plans to significantly increase annual capital expenditures from $17 billion in 2020 to $25 billion to $28 billion in 2021, an increase of 45% to 63%, of which about 80% will be used For the 3nm process research and development, this means that TSMC will have more than $15 billion in capital expenditures this year to invest in the 3nm process.

According to TSMC’s previously announced plans, their 3nm process is planned to be trial-produced this year and mass-produced in 2022. In other words, according to TSMC’s plan, we can see some products equipped with TSMC’s 3nm process in the market next year.

This is still in line with the rhythm of TSMC’s advanced process upgrades in recent years. However, from the actual performance of the product, the high price has not perfectly achieved the expected effect. In other words, TSMC is now likely to have hit a bottleneck between capital investment and technology implementation. TSMC, which ignores this bottleneck and is eager to realize the 3nm advanced process technology, has actually fallen into an extreme technology adventure. .

Extreme technology adventure, a blessing or a curse?

The reason why TSMC is currently facing technical risks has a high risk. In addition to the obvious reduction in the return of advanced process technology, it is also affected by many other risk factors, of which three are more obvious.

First, the cost will further increase. This can be felt from the fact that TSMC’s capital expenditure may soar by 45% to 63% this year. In terms of equipment cost, the process below 7nm is inseparable from the EUV lithography machine. The entire 3nm process must use the EUV lithography process, and the number of layers has been increased from 14 layers in the 5nm process to 20 layers. Each additional layer is huge. the cost of. In addition, TSMC previously disclosed that its 3nm process will continue to use mature fin field effect (FinFET) transistors, but some necessary technological upgrades will inevitably lead to further increases in costs.

Second, the production energy consumption is higher. It seems that large power consumption should be included in the category of cost increase, but TSMC’s power consumption problem is not as simple as the increase in electricity bills. As early as 2015, Zhang Zhongmou pointed out that the only factors that plagued TSMC’s development were power shortages and power outages. By 2019, TSMC’s power consumption has soared to 14.33 billion kWh, and one-third of the increased power consumption in Taiwan in the past five years has been occupied by TSMC. TSMC will use more and more EUV lithography machines to build 5nm and 3nm production lines, and the energy conversion rate of EUV is only about 0.02%, so the power supply in Taiwan cannot afford the further development of TSMC, which has already became a real problem.

Third, the technical difficulty has been upgraded. The frequent quantum effects and similar micro-scale problems of advanced processes below 7nm are actually difficult to truly overcome. Although TSMC and Samsung have alleviated these problems through some effective technical means. However, from the perspective of the 5nm process product yield and specific performance and power consumption improvement performance of the two, the negative impact still exists, and there are doubts about the extent to which subsequent optimization can be made.

TSMC must have made relevant plans to deal with these risks, but it is undeniable that if TSMC’s dazzling performance is used, it will truly understand the risks it faces. So it’s hard not to worry, will TSMC’s next development path really be as smooth as most people imagine?

The Links:   LJ320U27 FF450R17ME4_B11