TSMC, the Taiwanese manufacturer of custom chips, will start industrial production of semiconductors this second half with N3 process technology, the most advanced in the entire industry in terms of density, power consumption and global performance of the chips manufactured. Last year, TSMC led the global semiconductor market with 26% of total output (excluding memory) and expects to continue to grow this, thanks to its expansion in the 5G and high-performance computing (HPC) market. Only Samsung overshadows it, despite the fact that it still has problems in the production of very high-performance chips, because the third party in contention, Intel, is still anchored in its manufacturing technology for more than five years.
Last Wednesday, on the occasion of the general assembly and the presentation of the 2021 report, the president of the board of TSMC, Mark Liu, and his CEO, C.C. Wei, were extremely confident in the future of their company, thanks to its reliable and continuously expanding production of advanced semiconductors. This situation at TSMC contrasts with that of the rest of the industry, which faces numerous challenges in terms of production efficiency, supply of materials and production lines, and talent and skilled labor, leading to fears that it will continue to be unable to meet its commitments to deliver semiconductors to its customers, including next year.
In the 2021 report, an increase in TSMC’s billing last year of 24.9% compared to 2020 and constant increases in billing for twelve consecutive years are made official. At the assembly, Liu reiterated his goal of increasing revenue this year by 30% compared to 2021. TSMC is entering a period of high structural growth, thanks to strong demand for 5G networks and mobiles and high-speed applications. performance, which trigger the need for much more advanced chips, Liu said.
TSMC is set on producing highly competitive chips over the next three years, he announced at its annual meeting.
The mass production of the N5 chip manufacturing process (colloquially referred to as 5 nanometer chips, although this measurement now has little to do with the physical distance between transistors) is entering its second year and already represents 19% of all sales. of wafers from the Taiwanese company. The N4P process technology, an improved version of the N5, was introduced last October and is expected to go into mass production this coming semester. The N4P and N4X, the extension of the N4P, achieve a performance improvement of 11% and 13%, respectively, compared to the N5. The N4X is expected to go into mass production in the first half of next year and will be the first to focus on high-performance computing chips, primarily for data centers.
The N3 chip generation comes into play
C. C. Wei confirmed that by the end of the year chips with N3 technology will begin to be manufactured in large quantities and that the improved version, the N3E, will go into production a year later, with which the so-called 3-nanometer chips will progressively displace the current 5 nanometers for the most demanding applications. The CEO of TSMC also assured that the development of the next chip manufacturing technology, the N2, continues on its planned course.
For the next ten years, the semiconductor industry has very good opportunities, thanks to digital transformation and robust demand for electric cars and HPC applications, as well as mobile networks. Wei acknowledged, however, that demand for smartphones, computers and consumer electronics products is slowing. This has made the company adjust its product portfolio towards those that offer higher performance, with more features and lower power consumption due to the higher integration density of its chips. With the N3 process technology, it is clear that the possibilities of the chips will be greater than with the N5 and the N4P and N4X.
Mark Liu, left, and C. C Wei, CEOs of TSMC.
The good expectations proclaimed by both directors of TSMC and the continuity of the semiconductor integration curve in the coming years and the company’s roadmap, which can be seen printed in the annual report, does not prevent some concern by the increase in the costs of raw materials, the construction of manufacturing plants, production lines and the lack of talent and skilled labor, as well as supply problems of all kinds.
Samsung negotiates with ASML the rapid delivery of advanced semiconductor manufacturing lines, at any price, to supply demand
TSMC may not have as much trouble supplying chips to its customers as the rest of the industry, but it is also affected. The supply chain for a chip is very long and with the wafer made, or even cut into small pieces or chips, which is what is done in Taiwan, it is only the first step. The pandemic on the island, which has already caused more than two million people to be infected by Covid in the last two months, is another concern.
So much so that TSMC rented a hotel for its annual meeting and asked its hundreds of shareholders to follow it recorded and broadcast live from separate rooms and distributed in very small groups, while TSMC managers spoke in an empty room. TSMC, like all of Taiwan, takes Covid as seriously or more seriously than mainland China, which has allowed its chip production to continue more or less normally in the last two years.
Higher-than-expected costs in the US
Mark Liu acknowledged at the meeting, for the first time in public, that the construction costs of his chip manufacturing plant in the United States are higher than expected and that hiring the right personnel is a challenge. The location of the plant in Arizona, in the southern United States, does not make construction work any easier either, because it is suffocatingly hot.
Liu wanted to play down the problem of recruitment in the United States, noting that finding engineers and technicians in the United States “is more difficult than in Taiwan, but we are achieving it and for now we have the people we need.” In the case of Arizona, the fact that Intel and Samsung are also building plants may make the supply chain more competitive for years to come, he noted. In Arizona, you have the advantage of abundant land and water, an extremely scarce resource in California, where the large US chip plants used to be located.
Image of the TSMC plant in Phoenix, Arizona, in September last year.
In the plant that TSMC is building north of the state capital, in Phoenix, the initially planned investment is 12,000 million dollars. Chips will begin to be manufactured in 2024, with N5 technology, and TSMC estimates that at a rate of 20,000 wafers per month. By the time the US plant is complete, the company will have its N3 process technology up and running in Taiwan. TSMC chips that are made in the United States will be used primarily for components used in smartphones and computers.
It is common for the most advanced semiconductor factories to be built first in the companies’ home country and replicated abroad years later. This is what Intel has always done: the plant it is building in Oregon will have state-of-the-art technology when it comes online and the one it is now expanding in Ireland and in a few years in Germany, thanks in part to subsidies from the European Union, they will use the process technology tested a few years earlier by Intel in the United States. Samsung does the same: the most advanced plants are in South Korea and then it has others in China and the United States.
Liu reiterated his commitment to building chip manufacturing plants outside Taiwan, as he will do in Japan thanks to his agreement with Sony or is doing in the United States, but at the same time defended the importance of expanding chip production in Taiwan. “There is talk of geopolitical tensions around Taiwan and whether it is an ideal area to have so many plants,” he conceded, but added that “politically, if there is good and deliberate international cooperation, Taiwan is a very stable place to manufacture chips. ”. Regarding making chips in Europe, Liu said that the possibility is being evaluated, “although it is very far from having concrete plans.”
Liu dismissed the prospect of the United States allying with Japan and South Korea to challenge his company’s dominance in advanced chipmaking, as had been speculated by President Joe Biden’s visit to Japan and a Samsung factory in South Korea. The advanced manufacturing of a chip is not something that can be achieved by imitating others, but by allocating huge amounts of money and talent over many years, Liu said, although, he clarified, rivals cannot be taken to the lightly
Tour of the head of Samsung in Europe
Precisely, Lee Jae-Yong, heir to the Samsung conglomerate and vice president of the South Korean company, is visiting the Netherlands these days, after requesting permission to go abroad to the authorities of his country, for “reasons of national interest”, since it is under judicial surveillance. The main reason is to meet with Peter Wennink, CEO of ASML, the only company capable of building production lines for semiconductors with extreme ultraviolet (EUV) technology, essential to achieve such precise and tiny photolithography.
The president of Samsung comes with the checkbook in his pocket, willing to sign any amount as long as ASML quickly supplies him with chip production lines. But Wennink had already warned him that the demand for his products far exceeds his production capacity, even though it has increased markedly in recent months and years. ASML is also feeling a lot of pressure lately from Washington and Beijing to also sell them equipment. TSMC is also one of ASML’s big customers, but apparently it was the first to pass on orders for the next few years long ago while others, like Intel, have woken up later. At the same time, China is pressing for ASML not to pay any attention to Washington’s dictates.
ASML has more than 5,000 suppliers from around the world of highly sophisticated components, who also have problems sourcing raw materials and shipping finished products to ASML. And increasing the production of a component is not an easy task either, because it takes a lot of time and allocates highly qualified personnel. Each ASML EUV production line is priced at $150 million per unit and requires several months to assemble, in addition to moving several planes with material.
Samsung announced at the end of May a plan of 356,000 million dollars to invest in the manufacture of semiconductors over the next five years and thus supply the insatiable demand that is expected in the next decade. Samsung also manufactures two different types of semiconductors: those for memory, such as DRAM or NAND flash, which are highly integrated, highly repetitive designs, and those for more or less custom components, such as processors or controllers. Your investment possibilities are thus broader and more diversified; Although it is always about making chips, the cycles of supply and demand are not the same.
Samsung’s main rivals for the first type of chips are its compatriot SK Hynix, the American Micron or the union of Toshiba and Western Digital, now called Kioxia. For the second type, the rival is above all TSMC, since Intel mainly manufactures standardized microprocessors and other companies, many of them European, are mainly targeting chips for vertical sectors such as automotive or digital signal processors (DSP) and are smaller. .
Find a reliable alternative to TSMC
To lessen their reliance on TSMC, US companies like Qualcomm and Broadcom have tried to shift orders for their most sophisticated chip designs to Samsung and Intel, with relative success because TSMC’s expertise is hard to come by. A recent example is the new M2 processor designed by Apple and exclusively manufactured by TSMC, like its predecessor M1, and which is putting the venerable Intel designs to shame with its x86 architecture. AMD is also gaining a lot of share from Intel with its x86 processor designs made by TSMC and GlobalFoundries, a former shareholder. And NVidia is making a name for itself in the high-performance server processor market, with its designs made by TSMC competing with those of AMD and Intel. It is logical that both Apple and NVidia wanted to have an alternative supplier to TSMC, for now difficult.
For Samsung, growing in the high-performance chip market is critical, because that’s where the biggest growth is expected and where profit margins are highest. But it is also the case of Intel, which in recent years has hardly grown its turnover while that of TSMC, Samsung and other smaller manufacturers has skyrocketed. For the United States, getting Intel off the ground is a priority, because it would grow the entire US semiconductor ecosystem.
The growing rivalry between the three big semiconductor manufacturers (TSMC, Samsung and Intel) raises, however, growing doubts about profitability in the medium term. While annual investments in semiconductor manufacturing lines were around 40,000 million dollars between 2012 and 2017 and 60,000 million during the following three years, in 2021 they exceeded 90,000 million and for this year they are expected to touch 105,000 million dollars.
It is true that the global turnover of semiconductors will be almost 650,000 million dollars this year, 100,000 million more than last year, but it is also not clear that this rate of growth in turnover can be maintained when the current investments in manufacturing lines enter at full capacity within two or three years.
It is reasonable to think that supply will begin to exceed demand in the medium term and prices will fall accordingly. But it’s also clear that TSMC’s optimism for the next two or three years is justified. The problem, after all, is to glimpse what will happen after 2025, a horizon that is too far away for current politicians and executives, who are rightly concerned about the current situation at all levels.