Editor’s note: This article comes from WeChat public account “Semiconductor Industry Watch” (ID: icbank), author Qiu Liting, 36 krypton released with authorization.
Recently, news that the United States wants to build a fab in the country has been frequent. At first, people with knowledge of the situation came out. Trump administration officials are negotiating with TSMC to build a plant in the United States.According to the semiconductor industry observation, TSMC has announced its plans to build factories in the United States.According to TSMC’s announcement, the plant will produce 5nm chips with a planned production capacity of 20,000 pieces per month.
This will result in more than 1,600 high-tech professional direct jobs, and thousands of indirect jobs in semiconductors.According to the plan, the plant will start construction in 2021, with the goal of starting production in 2024, and capital expenditures from 2021 to 2029 will be as high as $ 12 billion.
In fact, not only TSMC, Intel and GlobalFoundries are also potential partners for the US to seek controllable wafer foundry.
According to a recent report by the Wall Street Journal, last month, Intel CEO Bob Swan wrote to officials of the Ministry of Defense that his company is ready to cooperate with the Pentagon to establish a commercial foundry.Laurie Kelly, a spokesman for GlobalFoundries, said the company is ready to work with the industry and the US government to “ensure that the US has the manufacturing capabilities needed to provide semiconductors for its safest and most sensitive technologies.”
It can be seen that the Trump administration is prepared to fight this time, but can they really reshape the current foundry pattern?
Wafer foundry status
Wafer foundry is a vital part of the semiconductor industry chain, and the level of manufacturing process directly determines the advanced level of the semiconductor industry.Before the emergence of TSMC, basically all semiconductor manufacturers had their own fabs, just like Samsung and Intel. If they continue to develop according to this model, it will undoubtedly be a huge challenge for most new chip entrants, becauseThe investment in the wafer production plant is not a small amount.
A decision by TSMC founder Zhang Zhongmou changed the trend of the entire semiconductor industry. His idea is to solve the “last centimeter” of the road in the popular language today.
At that time, most semiconductor companies regarded wafer manufacturing as a sideline business, because their main business was not wafer manufacturing, but designing and selling their own products, so their foundry services were not professional.Zhang Zhongmou, who has rich experience, takes this into consideration. Taking advantage of Taiwan’s progressive manufacturing advantages, he has begun to develop the “bad land” of wafer foundry.
According to Wikipedia data, in 1986, Zhang Zhongmou established the first professional wafer body manufacturing company: Taiwan Integrated Circuit Manufacturing Company (TSMC), and served as chairman.In 1988, TSMC ushered in a great turnaround and won the opportunity to work for Intel.Since then, TSMC has continued to expand in size and has made amazing profits. It has become Taiwan’s most profitable company.
The process from the establishment of TSMC to the continuous expansion of the scale is also the process of the successful independence of the professional chip foundry model from the semiconductor industry.The success of TSMC has prompted the rise of fabless semiconductor Fabless.On the contrary, the rise of Fabless has also increased the development and competition of foundry.
Now, three decades later, wafer foundry has become an extremely important business in the semiconductor industry.According to the latest data analysis of the Tuo Bong Industrial Research Institute of Jibang Consulting, the top ten wafer foundries in the first quarter of 2020 are: TSMC, Samsung, GF, UMC, SMIC, Gaota Semiconductor, World Advanced, PowerJaeger-LeCoultre, Huahong Semiconductor and Eastern Hi-Tech.
As can be seen from the table, TSMC is still a must-have, with a market share of more than 50%. At present, TSMC has mass-produced 7nm. For 5nm technology progress, TSMC said at the financial report meeting on April 16 this year that 5nmMass production will begin in the second half of 2020, and research and development of 3nm and 2nm are in progress.
The second-ranked Samsung is also about 16%. Samsung 7nm has started mass production in February this year.At the same time, overweight investment and deployment of 5nm process technology is expected, and it is expected that the 5nm EUV production line will be completed by the end of June, and the production of 5nm will begin as soon as the end of this year.
Domestic SMIC and Huahong Semiconductor are also catching up. At present, SMIC has begun mass production of 14nm, and has successfully carried out OEM production for Huawei’s HiSilicon Kirin 710A chip. Although there is still a certain gap with the head manufacturers, it has achieved “Breakthrough from 0 to 1. “At the same time, in terms of Huahong Semiconductor, Zhao Honghang, the chief engineer of Huahong Group, pointed out that its 14nm FinFET process has been completed across the board, and the SRAM yield rate is 25%. It will be promoted rapidly in 2020.
Nowadays, the United States is beginning to attract wafer foundries. Will the pattern be rewritten under the current wafer foundry situation?Based on this, we first consider the two invited US manufacturers.
GF seeks revival
In the semiconductor industry, the phenomenon of “the boss eats meat, the second eats soup, and the third drink northwest” is everywhere. The rule of winner-take-all prevails in the field of wafer foundry.And GF is the “third”.
Ranking of the world’s top ten pure wafer foundries in 2018 (unit: million US dollars)
Looking back at the 2018 data, GF also occupies the second position on the list. In just over a year, GF, which had originally ranked second, has been more than doubled by Samsung in market share in the IC industry.With the rapid development of the environment and the trend of global fabs benefiting from the IC industry, GF’s performance is undoubtedly disappointing.
GF was split from AMD. In 2008, AMD announced its fabless plan and decided to split the semiconductor manufacturing business into a new company.In 2009, GF announced the establishment, quickly acquired a chartered semiconductor company, and set up a new factory in New York.
In order to catch up with the pace of TSMC and obtain intellectual property rights and related technologies, in 2014, GF acquired most of IBM’s semiconductor business. While gaining advantages, it also acquired production capacity and talents.In 2016, GF obtained a 14nm 14LPP FinFET process authorization from Samsung Electronics and developed a 12nm 12LP node based on this.
For fabs, maintaining process leadership is a core competence, so integrated circuit manufacturing companies are constantly investing heavily in purchasing expensive equipment and research and development of new technologies.However, GF has suffered a great loss in the advanced process. To ensure the technological advancement, GF invested like snowballing in R & D, but it has always been unable to control R & D investment and output, resulting in increasingly heavy debt.
Looking back at the financial status of GF in these years, the performance is very general. Some data indicate that from 2011 to 2017, GF has a total loss of 6.94 billion US dollars in 7 years, with an average annual loss of nearly 1 billion US dollars in 2014 and 2015.The three years and 2017 were the most severe losses, with a total loss of more than 4 billion US dollars in three years.
On August 28, 2018, GF announced that it would stop investing in 7nm process research and development and instead focus on the existing 14 / 12nm FinFET process and 22 / 12nm FD-SOI process.With the news, some people have begun to question whether AMD’s decision to cooperate with TSMC’s 7 nm process product is related to the move. After all, AMD has always been GF’s main customer. GF’s withdrawal from advanced process research and development has undoubtedly made TSMC smile.AMD orders are firmly established in the global foundry throne.
For GF, another loss caused by the withdrawal from advanced process research and development is IBM. In November 2018, according to foreign media TechSpot, TSMC and IBM signed an agreement to produce large server chips for the latter.IBM itself also has a bitter to say. After all, in order to sell the fab to GF, it also posted 1.5 billion US dollars. One of the reasons for this move may be to maintain a long-term cooperative relationship with GF.
After announcing the withdrawal from advanced process R & D, GF has made many changes, including the sale of multiple fabs and related businesses. The fab project that invested nearly 10 billion US dollars in Chengdu has also fallen into stagnation.GF’s work has always been negative news such as layoffs, factory sales, project closures, and company sales.
GF’s toss and go seems to have not stopped. As mentioned earlier, GF is discussing with the government to build an advanced foundry in the United States.Relevant reports show that spokesperson Erica McGill said in an email: “GlobalFoundries is discussing with the US government how to ensure US technological leadership through semiconductor manufacturing in the US.”
However, imagination is beautiful, and reality is cruel.For GF, the abandonment of advanced processes has made it lose two customers, AMD and IBM. If you want to rebuild the old one, you must come up with advanced technology, at least not fall on TSMC, which is obviously impossible for the current GF.
Not to mention surpassing TSMC, as far as research and development of advanced processes are concerned, according to the data, it currently requires more than US $ 4 billion to build a 12-inch 32 / 28nm production line, and the investment in the 12-inch 14nm production line is as high as US $ 10 billion.For more advanced technology, the investment cost is unimaginable, and the depreciation cost is also amazing. The GF that has been losing money for years will inevitably be very difficult.
Even if GF has considerable capital with the support of the federal government, research and development will take some time. Moreover, it has spent a lot of time and money to enter Albany, and it has not become a leading foundry.
Intel keeps on giving
Another manufacturer is Intel. As another US-based manufacturer, Trump administration officials are also negotiating with Intel in the hope that they can build factories in the United States.Many people think that Intel is only responsible for its own chip production, in fact, Intel has also done wafer foundry business.
In 2008, Intel opened its foundry business. By 2013, Intel officially opened its foundry business. Its customers include Achronix, LG, Netronome, Spreadtrum and other companies. However, by 2018, there are rumors that Intel will close chip foundry.The business lasted only a few years.
At that time, some foreign media pointed out that after Altera was acquired, Intel’s foundry business hardly left any well-known chip design companies. Altera was Intel’s largest early foundry customer.After a lot of calculations, Intel spent nearly ten years entering the wafer foundry market, but in the end it became a dream.
There is an old saying that the opponent is the one who knows himself best.Zhang Zhongmou, chairman of TSMC, once commented on Intel’s efforts in foundry foundry. He said: “Intel is not a professional wafer foundry, but just extends its feet to test the water temperature in the pool. I believe it will find that the water is very cold.” Zhang ZhongmouHe even quoted the Battle of Stalingrad. He said that the battle was a matter of life and death for the Soviet Union, but not for Germany. Therefore, the German army of 300,000 was wiped out.
Intel seems to always stay in the OEM business. At the 2017 manufacturing day, Intel once said that advanced processes are the fastest growing part of the chip OEM field (23 billion US dollars), accounting for almost half of the chip OEM revenue.Therefore, at the meeting, it not only exposed the details of 10nm, but also revealed the 22FFL entry-level FinFET node, and made it clear that it is establishing a foundry franchise.
But this so-called “foundry franchise” has no splash in the next few years. Seekingalpha once pointed out that Intel did say that “22FFL was originally developed for its own purposes”. Obviously, itIt does not seem to be a good consciousness of becoming a foundry, and always prioritizes its products.
Not only that, Intel also performed poorly at 10nm.In 2015, it was stated that 10nm was postponed from 2016 to the second half of 2017.In 2017, it was postponed to the second half of 2018.In 2018, postponed to 2019, it is no wonder that the outside world called it “toothpaste factory”.
Needless to say, according to Moore’s Law, the three-year delay in the process will greatly change the competitive power of products provided by foundries.Not to mention that this will cause problems for customers who promise to join the node.This move not only lost the original customer but also the potential customer lost interest in it.
TSMC has performed extremely well in this regard. It successfully mass-produced 10nm in 2016 (Note: Intel and TSMC have different understandings of the process), and achieved mass production of 7nm in 2017. At present, TSMC’s 5nm process is also about to be mass-produced.The development of the more advanced 3nm process is also very smooth, and it is expected to start production in 2022.
As a foundry, TSMC produces chips for many customers with many different designs, which requires flexibility and originality.And Intel’s factories are mainly designed for their own products. However, they can’t even effectively produce their own 10nm chip designs.
For ten years, Intel has been operating this business as a sideline, or as a regulatory buffer for its main processor business. For Intel, wafer foundry may have its strategic importance, but it has never been regarded as the main core.It is this mentality that Intel’s foundry business has not performed very well.
The invitation of the federal government has become an important opportunity for Intel to restart its foundry business. From the outside, Intel seems to be ready.The first is the previously criticized 10nm, Intel has achieved formal mass production in 2019, and its CEO Bob Swan even called the production of 10nm “good”.Secondly, Intel gave a complete R & D process, and guaranteed to maintain the process rhythm and the ten-year roadmap for the update every two years.
Intel even promised to regain its technological leadership in the 5nm process. The next step seems to be obvious. Intel is ready to recommit to the foundry market. The invitation of the US federal government may just be a smooth ride.
But as simiwiki pointed out in the report, if Intel wants to completely change its identity and become a strong competitor of TSMC, it will be an extremely difficult task.The main points are as follows:
1. Because foundry and IDM are two very different concepts.Foundry for others requires not only different skills, but also different fab settings.
2. Most chip design companies are profit-centric, and Intel’s foundry costs will be prohibitive if they cannot be resolved.
3. Intel’s production capacity is also a very important issue. After all, Intel still has to produce chips for itself, so it can only use the remaining production capacity for other chip design companies.
4. Intel is an IDM manufacturer, so companies that previously worked with Intel are far less complementary than competitors. Let competitors such as AMD or Nvidia also be potential customers to choose Intel OEM. Not only will Intel not agree, for theseAs far as opponents are concerned, there are also concerns about product secrets leaking to their counterparts.TSMC has not been involved in the field of chip design so far, and there are also considerations in this regard.
Nonetheless, we do not know Intel ’s actual production capacity, nor do we know whether it has controlled the corresponding costs. After all, Intel still has the right to speak in terms of advanced processes.
Is TSMC the best partner?
In fact, for the United States, the introduction of TSMC may be the best solution at the moment.
As we mentioned earlier, TSMC has become the world’s leading foundry, and they also have a large customer base.Its 5nm production line is also launched this year.They also plan to start trial production of 3nm next year.With customers and technology, this is undoubtedly the best partner for the United States.
But as early as last year, the United States had plans to allow TSMC to build factories in its own country. However, TSMC gently rejected the request on the grounds of high costs in order to maintain a neutral attitude.Bloomberg columnist Tim Culpan commented, “TSMC cannot always watch the fire from the other side, even if Liu Deyin just wants to make chips, he will eventually have to make a choice.”
According to TSMC’s 2019 annual report, last year the U.S. market accounted for about 59% of TSMC’s revenue, compared with 19% for Chinese customers.Under such a general trend, TSMC does not have many options, so Gao Canming believes that it is very likely that TSMC will build a new plant in the United States in the foreseeable future.The development of the U.S. domestic wafer fab also requires a certain amount of time and capital investment.
From the latest report in the Wall Street Journal, TSMC seems to have made a decision.But the well-known semiconductor analyst Lu Xingzhi said recently that he did not approve of TSMC’s visit to the U.S. to set up factories for three main reasons:
1. There is no semiconductor downstream packaging, testing, PCB, module, and assembly industry chain in the United States (Hon Hai Wisconsin’s investment has not yet been affected), do we still need to process the wafers, and then spend huge freight to send to various countries for assembly?The most powerful thing in the United States is design and R & D. Do you not want to become a kingdom of technology product manufacturing?
2. The United States also lacks the efficiency required for foundry manufacturing, and can it work and develop three shifts 24 hours a day (not in Europe)?Ten years ago, I estimated that the cost of wafer manufacturing at Intel ’s US plant was more than twice that of TSMC (mainly yield, capacity utilization, semiconductor upstream and downstream clustering benefits, and work efficiency differences).
Watch TSMC Wafertech know it without expanding the factory, which is why even Intel has to run overseas factories to find ways to reduce costs. Now AMD relies on TSMC’s advanced and low-cost process to grab the share of Intel CPUs. Globalfoundries Fishkill New York plant can not continue to operates reason.
3. If TSMC goes to the United States to produce core technology semiconductors, in addition to doubts about the future development of Taiwan’s semiconductors, when a major event occurs at TSMC, the U.S. government only needs to take care of the TSMC U.S. factory, so all U.S. customers are taken hostage.If necessary, TSMC must not be fooled.
Write at the end
It has been reported that if it is counted from now on, it may take five to ten years to establish a leading fab in the United States and may be able to barely run.Considering that Intel ’s failure in the foundry business is not due to lack of technology, perhaps, after a decade of investment and with the help of TSMC or with the cooperation of both parties, a leading foundry may be established in the United States.But no one will know what will happen in Taiwan in ten years.
The OEM business has experienced rapid growth for more than 30 years, and is currently showing the situation of being strong.Talent, advanced technology and production capacity have always been the core business of the fab.Especially for TSMC, in the war of advanced processes, TSMC can be said to be invincible in the world.Such a “vicious circle” of high investment and high return has strengthened TSMC’s strength, and the difficulty of catching up with latecomers has become increasingly greater.Although Samsung, which has earned a lot of money for storage in the past two years, has invested heavily in the fab, even if it does, there is still a long way to go.
But then again, although the outbreak hit everyone by surprise, semiconductors are a highly globalized industry.So far, there has been a clear division of labor.If the United States really wants to establish a self-sufficient supply chain, it is necessary to move the entire industry chain back.This includes not only foundry, but also packaging and testing.After all, producing wafers in the United States and shipping them to other countries for packaging is tantamount to doing more.
Packaging and testing is much easier than building a foundry, which is why it shows that they are low-level, low-margin businesses. In a developed country like the United States, where labor costs are high, packaging and testing is obviously not cost-effective, Will greatly increase the cost of the chip.Therefore, the follow-up may be that the chip is still packaged overseas.
At least for now, it is not easy for the United States to reshape the foundry pattern.