//
Recently, American Museum of Computer History released a shorthand for their interview with Jiang Shangyi in March this year.
According to the introduction, Jiang Shangyi was born in Chongqing, China in 1946, and his father moved to Taiwan, China in 1948. According to Jiang Shangyi's description, his growth process in Taiwan was very crowded and difficult. However, he was still able to study in electrical engineering at Taiwan University in Taiwan. In 1969, he entered Princeton University in the United States. In 1974, Jiang Shangyi transferred to , Stanford University, to study for a doctorate degree. After reading
, Jiang Shangyi worked in ITT, TI and HP Labs, and then served as the president of TSMC's R&D, leading the TSMC technical team to become the leading position of the wafer foundry giant. Jiang Shangyi retired from TSMC in 2006, but was asked to return in 2009. He wants to focus on two major projects: transistor "leading" and advanced packaging. He left TSMC in 2015. In 2016, he was invited to join the board of directors of SMIC and started his career at home for several years.
According to the Museum of Computer History, this oral history is a very valuable insight into TSMC's development and how it can dominate the semiconductor foundry business. Anyone interested in the semiconductor industry dynamics must read it.
The following is Jiang Shangyi's oral summary:
1 Career
When asked why I chose to work in the semiconductor industry, Jiang Shangyi responded: "I chose semiconductor because I happened to have a system, and some people served as chair professors, most of them from American universities. When I was a senior, we happened to have two lecturers in the semiconductor field. So I took some semiconductor courses. At that time, our possible choices were control, power and communications, and semiconductors. Computers had just started at that time, all of these fields. I was engaged in semiconductors because I thought my own interests were more in the direction of physics."
was because I was interested in physics. Jiang Shangyi chose semiconductors, communications and information theory, which he believed involved more mathematics. Then, he went to Princeton.
"Princeton is very famous, I'm very happy. But when I got there, I found that Princeton is actually very, very academic. At that time, I thought they were top in physics, mathematics, and philosophy. If you were an EE student, you could benefit a lot. If you were a semiconductor major, they would send you to the physics department in the first year, and then you came back the next year. For information theory, they sent you to study mathematics and then repeat the same situation. After going to Princeton, I was very shocked. It was very academic in most fields," said Jiang Shangyi. After
, Jiang Shangyi said that he started to apply to real engineering schools, including MIT , Berkeley , Stanford University, Caltech . Finally, he was admitted to Berkeley and Stanford, and not to MIT and Caltech.
As mentioned above, Jiang Shangyi finally went to Stanford University to pursue his PhD, studying under Dr. Gerald L. Pearson, an old senior at Bell Labs , who worked with Shockley and those people.
According to Wikipedia , Gerald L. Pearson was born in , Oregon . He received his bachelor's degree in mathematics and physics from Willamette University and his master's degree in physics from Stanford University. Since 1929, he has worked as a research physicist at Bell Laboratory, and his early work on thermistor has obtained 13 thermistor patents. After in World War II, he was a member of William Shockley's group, and his experimental results are crucial to developing semiconductor behavior models. In 1946, according to Shockley's suggestion, he applied voltage to a droplet of ethylene borate (glycol borate :gu) placed on the PN junction, looking for transistors.
1954 His work on silicon rectifiers with Daryl Chapin and Calvin Southern Fuller led to the birth of the first practical photovoltaic cell ( solar cell ).In 1960, he retired early from Bell and worked as a professor of electrical engineering at Stanford University, establishing the compound semiconductor research project.
According to Jiang Shangyi, his doctoral thesis at Stanford was about gallium arsenide , which was a very promising material at the time.
2 Early work experience
When he graduated from his Ph.D. in 1974, it was during the economic recession, so Jiang Shangyi began his working career at ITT, engaged in research on GaAs lasers. According to reports, this product is an LED for optical fiber communication . ITT has a fiber optic communication program that requires a light source, so my first job was to study this light source for them. At that time, its working team was about five or six people.
After working at ITT for a year and a half, Jiang Shangyi jumped to TI, which was mainly based on two considerations: on the one hand, Jiang Shangyi believes that he is more like studying silicon; on the other hand, he hopes that the first job should allow him to see further, rather than being limited to such a small team, and TI is the choice he made based on the above two considerations.
After going to TI, Jiang Shangyi's initial work was to study silicon solar cell . At that time, affected by the "oil embargo" crisis, the US government hoped to explore a new path forward and use it as an alternative energy source. According to Jiang Shangyi, at TI, they made silicon solar cells. They also set a very radical goal at the time, but it never achieved it.
Two years after studying solar cells, Jiang Shangyi turned to studying NMOS and CMOS. At that time, research on CMOS had just begun, and Jiang Shangyi's team also tried to learn how to make CMOS. According to Jiang Shangyi, it was very interesting at the time that you could develop a complete technology with an engineer. I could run some SUPREM simulations to simulate, and then I diffuse, put my own wafer on the quartz boat, push it into the furnace, pull it out and see what it does.
"At that time, we were studying the 5 micron process, and we could achieve it too. Then when I left TI, I was studying the 1 micron CMOS technology," said Jiang Shangyi. He also emphasized that during this experience at TI, he learned a lot of semiconductor-related technologies and knowledge.
From Jiang Shangyi's introduction, we know that he worked at TI for about four and a half years, from 1976 to 1980. "At that time, Zhang Zhongmou, who worked at TI, was about 10 levels higher than me," Jiang Shangyi teased. He also said that during his time at TI, he did not really know Zhang Zhongmou.
After leaving TI, Jiang Shangyi joined HP Labs and worked there for 17 years. According to reports, during his work at HP Labs, Jiang developed some different technologies in different projects, but most of them were related to CMOS. "I've worked in CCD projects, I've worked in CMOS projects, I've also worked in bipolar projects. I've been appointed as the project manager to develop the first generation bipolar products they call HP-25. I think they're still using HP-25 in production, mainly for instruments. The instrument still uses bipolar because it's high speed and not very large," said Jiang Shangyi. In his opinion, during his time at HP Labs, he happened to be seized on the opportunity of the IC industry. Because at that time, every company wanted to have its own IC technology and build its own wafer fab.
Jiang Shangyi also mentioned that when HP , especially in the later stage, they began to realize that the above practice (each company has its own IC technology and wafer fab) was meaningless, so they began to integrate all IC technologies. Jiang Shangyi also realized that HP had planned to phase out ICs, so he believed that it was always under pressure to integrate later in HP's career - budgets were cut every year and would never hire new people.
Finally, HP Labs finally closed.
Jiang Shangyi quipped: "In retrospect, this is also our own fault. For example, we said we started developing CMOS with 0.5 micron. But we never completed it. We would also say that 0.5 micron is too old and no longer advanced, let's work on 0.35." It is precisely because we always want to study the most advanced technology, but we never accomplish anything, and we never transferred any technology to make it.This ultimately led to the end of HP Labs. This is also the reason why Jiang Shangyi believes that the laboratory is no longer popular with the company.
"In terms of work, HP is really a very suitable place, and its culture is also very good and comfortable. But looking back, for IC technology, you really need scale, and it is a large scale. Small scale cannot survive because it cannot be competitive." Jiang Shangyi said.
"Start from the very basics, because the equipment is too expensive. For example, if you want to study 16nm technology today, you need 300 different devices, each with its own throughput. Some devices can make 300 wafers in an hour. Another device can only produce 3 wafers in an hour. This means that when working, a device is vacant 90% of the time," Jiang Shangyi continued.
According to Jiang Shangyi, if the above situation occurs, your equipment cost, your depreciation, and your equipment efficiency will become very, very low, so your cost will be very high. For 3 wafers per hour, you buy another device, and suddenly, your throughput goes up to 6 wafers per hour. So if you look at the cost of capital per 1,000 wafers versus the total capacity curve. That's it.
At the beginning, it was very, very high because many devices were idle. As your capacity grows bigger, it will drop. So, when it comes to knee, this knee point is about 15,000 to 20,000 wafers per month. So those who want to make new technology, leading technology fabs today need to produce at least 30,000 wafers per month to be competitive. Because even at that time, your entire factory cost 55% is depreciation, even if you reach the plateau (plateau).
So if the fab capacity is much smaller than 30,000 wafers per month, you will find that you are not completely competitive. Your costs will be much higher than your competitors. Additionally, if your company has a small capacity, you will have to pay more and higher prices when you purchase the equipment from Applied Materials Company . We know the gap could be 15% or more. And Samsung or TSMC pays for the same equipment, they can cost 15% or 20% less than small wafer fabs.
So in all these things, economies of scale really play a very critical role. Most importantly, now if you want to develop leading technology with
, you need at least about $2 billion in budget. If you have $2 billion, your income needs to be about $40 billion.
According to Jiang Shangyi, the only ones that can afford to develop leading processes now are TSMC, Intel and Samsung, because they all have more than $40 billion in revenue and are fourth-placed less than $10 billion, so they can't afford it.
He also recalled that in the later stage, HP actually realized the financial problems facing HP Labs, and they also planned to do wafer foundry. They also visited TSMC, but found that the gap with TSMC was too big. This made HP fully understand that it was too late.
3 TSMC's career
In Jiang Shangyi's decades of working life, TSMC's experience is undoubtedly the most worthy of writing.
Jiang Shangyi also emphasized in the interview that he was not very familiar with Zhang Zhongmou, who was the chairman of TSMC at the time, and at best he had met him a few times in some meetings. But one day in 1996, after returning home from get off work, the 50-year-old received a call, and the other end of the conversation was Zhang Zhongmou.
On the phone, Zhang Zhongmou told Jiang Shangyi: "We have a vacancy in R&D VP now, you come to work with us. This is your salary, this is your job, this is the signing bonus." (Original text: We have an opening for VP of RD. You come to work with us. Here, this is the salary, and this is your responsibility, and this is a signing bonus)
"I don't need an interview, but I was directly provided with a job opportunity," Jiang Shangyi said. He also said that after hearing Zhang Zhongmou say these words, his response was: "No need, thank you very much, I never thought of going to Taiwan.” (Original text: Oh, no, thank you very much, but I never thought about going to TAIwan).
Jiang Shangyi said that the reason he said this was because in his opinion at that time, TSMC was still a very small company, and the job was not very stable. Moreover, he was carrying a mortgage at the time and needed to support his children's studies. Therefore, he could not stand such an attempt. "If something happened, it would be difficult for me to find a new job," Jiang Shangyi said.
But after half a year of negotiations, Jiang Shangyi still joined TSMC. He was also frank that the reason why he accepted TSMC's offer was related to the TSMC stock that Zhang Zhongmou provided him with at that time (the stock was awarded immediately). "Based on the stock price of TSMC at that time, if I worked at HP at that time, I worked at 65 at HP according to my salary at that time In the year, the salary income obtained is not as high as that of TSMC stocks, Jiang Shangyi said.
"From a financial perspective, there is no risk for me to join TSMC, so I accepted this offer," Jiang Shangyi said.
After joining TSMC, Jiang Shangyi also reported directly to Zhang Zhongmou. During his time in Taiwan, he often went home after 9 pm because his family was not in Taiwan, which gave him enough time to devote himself to work. "But I found that Zhang Zhongmou got off work on time at six o'clock every day," Jiang Shangyi joked.
As Jiang Shangyi said, when he joined TSMC, TSMC's technology had no status in the world. Even if they wanted to invest in papers to IEDM, the other party didn't even look at it, and then responded contemptuously: "TSMC's? oh! ” (Original text: From TSMC? Oh). He also pointed out that when he joined TSMC, the later foundry giant had only 120 R&D personnel, most of whom had very shallow experience.
Three months after joining TSMC, Jiang Shangyi became the fourth person in charge of the "0.25 micron project" that TSMC was developing at the time. Previously, they replaced three R&D VPs and also removed three project managers. After taking charge of this project, Jiang Shangyi led the team to confirm five major technical issues that need to be solved.
"I used the knowledge I learned in HP to solve technical problems," Jiang Shangyi admitted. "I did not steal any documents and trade secrets from my former boss," Jiang Shangyi continued. Although there are still many various problems in
, under Jiang's leadership, TSMC finally released the 0.25 micron process. When the 0.18 micron process, the team he led encountered some new problems. "In metal interconnection, you have metal, you have dielectric, and you use dielectric to isolate your metal wires. In the past, all dielectrics were silica , which is a very good insulator. But then people began to realize that I wanted to reduce the capacitor , so I had to reduce the K value . "Jiang Shangyi pointed out.
This finally promoted the "low k" concept that became famous in the industry.
Jiang Shangyi said that at that time, "low k" had begun to become an idea. The first generation of "low k" was called FSG. They put some powder in SIO2, and the K value would drop a little; so we used FSG at 0.25 microns; but after 0.18 microns, someone proposed a clever idea - to make some products that are very similar to FSG, but use spin technology, without having to undergo chemical vapor deposition. Spin technology has two advantages, namely lower cost and more planarization. This ultimately promoted the birth of HSQ. In addition, at this time, IBM also had a spin "low" called SILK k” materials, including Samsung, UMC and ST, also joined IBM's material R&D alliance.
Next, at 0.13 micron, Jiang Shangyi led the TSMC team to achieve another breakthrough.
According to reports, at 0.13 micron, people began to explore the transition from aluminum interconnection to copper interconnection. As a leader in copper technology, IBM has accumulated more than ten years of R&D.However, TSMC had no relevant experience before, but they still insisted on going to copper interconnection without hesitation. They eventually became the first fab in the world to realize copper interconnection and "low k" production. According to him, the reason for IBM's failure is related to their spin selection on "low k".
"The 0.13 micron process is a very important node for TSMC, because we took the lead in mass production of 'copper interconnect' and 'low k', and the industry has also begun to pay attention to TSMC from this time. This can be said to be a key turning point for TSMC," said Jiang Shangyi. In the interview, he also mentioned that the investment of employees is invaluable for TSMC to surpass. Because they work in three shifts and keep developing 24 hours a day. However, R&D in the United States only works eight hours a day. This is called "Faster learning cycle".
Jiang Shangyi also said that the reason why Taiwanese engineers can work in this way is related to the culture of Asians themselves. He said that Asians have experienced very difficult years, so they have a very high desire to make money. And willing to sacrifice your privacy and private life to have financial protection. (Original words: I think the culture. Asians are more hungry, because we had a tough life. So, to make money is more important to us. People are willing to sacrificetheir own privacy, their private life in order to have financial security.)
In his opinion, it is precisely this 24-hour uninterrupted work that is one of the reasons that made TSMC successful. He also gave an example, saying that in Taiwan, if the equipment is broken, even at two o'clock in the morning, they will ask the equipment engineer to maintain it, and the engineer will not complain, and their families will not complain. But if it is in the United States, the equipment can only be repaired the next morning.
Jiang Shangyi also shared an interesting story when he was at TSMC.
In 1999, TSMC opened a fab called WaferTech in Oregon, but the fab performed far behind TSMC's fab, which gave TSMC a headache. So one morning they decided to get the vice president in charge of manufacturing in Taiwan to get a list of 20 people, ask him to call everyone, call them to the office, each of whom stays for ten minutes respectively. The initials were confused because they had never spoken to the vice president and had no idea what they wanted to do. In the end, it turned out that TSMC wanted to send some people from Taiwan to the United States. In the end, this 20-person group pushed WaferTech on track.
4 Advanced Packaging Story
As we all know, Apple and TSMC can be said to be the most representative pair of partners in the past decade. The two of them have also fulfilled each other over the years of development. But Jiang Shangyi said that when Apple first contacted TSMC, it was just a test of the waters, and the American giants only submitted one product. Later, Apple tied itself tightly with TSMC. According to Jiang Shangyi, the reason why this happens is closely related to TSMC's layout in advanced packaging.
And it all starts in 2009. Because in that year, Zhang Zhongmou, who had retired for four years, returned to TSMC as CEO, and then he also wanted to bring Jiang Shangyi, who retired in 2006, back to TSMC. But when accepting Zhang's invitation, Jiang Shangyi said that he wanted to launch two plans, the first of which was transistors to lead. "We have been lagging behind Intel, and I want to catch up with Intel," Jiang Shangyi told Zhang Zhongmou. "The second plan I want to launch is advanced packaging," Jiang Shangyi continued to tell Zhang Zhongmou.
But after hearing this statement, Zhang Zhongmou said, "It was too late for you to join TSMC. In fact, since the first day of creating TSMC, we have considered whether we can do packaging, but we finally gave up. The main reason is that the latter has relatively low technical content and the profit of is also low." Jiang Shangyi responded: "No, the packaging I mentioned is not a traditional packaging like wire-making. What I want to do is advanced packaging."
Jiang Shangyi pointed out that at that time, he really could not think of a better term than advanced packaging to describe the kind of packaging he envisioned. But now, advanced packaging is already familiar to everyone."If you look at the technical diagram, on silicon wafer , we follow mol's law , and the progress is very significant. But if you look at the packaging in the PC board, if you look at the PCB, the metal pitch is 110 microns, which has been maintained for almost 20 years." Jiang Shangyi continued.
He said that in the past, people spent very little on packaging research and development. In addition, they spend more energy on reducing costs. Thanks to the advancement of silicon wafer , each upgrade technology will achieve performance improvements, which has made the entire industry satisfied in the past, so they did not pay much attention to packaging. But now, we are starting to see that encapsulation can become a bottleneck in some cases. As Moore's Law approaches its limit, we need to solve this bottleneck.
"Graphics chip giant Nvidia is our customer. They used to have a GPU with 8 DRAMs. You need to send a lot of signals back and forth between GPU and DRAM. If you look at this GPU and DRAM, the gap between them is so big. Why are they so far apart? Because the metal wires are very wide. If they are too close, you can't connect all of these metal wires. Because of this, people are willing to pay about 30% of the speed and about 60% of the power consumption to drive these wires." Jiang Shangyi gave an example. He also said that if a silicon wafer is used instead of a PCB, the GPU and DRAM can be placed side by side, so that its performance will be very similar to being on the same silicon wafer. After doing a simulation, this is almost equivalent to benefiting from two generations of technology.
However, Jiang Shangyi was also frank that at that time, only the graphics chips were designed like this, which meant that chips with the same needs were a minority.
After listening to Jiang Shangyi’s story, Zhang Zhongmou deeply agreed and asked: “How many resources do you need me to provide.” Jiang Shangyi responded: "I need an additional 400 engineers and about $100 million in equipment." After getting Zhang Zhongmou's decision, Jiang Shangyi began to recruit people. About a year later, he found Doug Yu, a person with very good skills, to promote the entire project.
So another year later, TSMC finally developed a technology called "silicon interposer". In the "silicon interposer", TSMC uses solder bumps instead of wire bonding and places the interconnections from CPU to DRAM on silicon. During this implementation process, TSMC does not even have to use very advanced silicon technology, but uses three generations of old technologies. But unexpectedly, the effect is very good.
While developing technology, TSMC began to look for customers. When they told Nvidia about this idea, the GPU giant responded, "Oh, this is awesome!". But Jiang Shangyi also said that they had never used it.
Another day later, Nvidia's vice president of OEM and outsourcing told Jiang Shangyi: "You don't know how we cooperate with you, how we cooperate with packaging companies, and how we cooperate with OSAT." “We can let OSAT do anything, but you’re too inflexible, you ask for 30% of the profit, they only ask for 5%,” the vice president continued. "I won't work with you to develop packaging unless you transfer your technology to OSAT and I work with them," the vice president told Jiang Shangyi.
In the view of this VP, working with OSAT, they can tell the latter to keep inventory when I have a problem with my budget. I can also tell them when to hold it and when to release it. In other words, OSAT will completely follow Nvidia's instructions, but they think TSMC has never done so. In summary, the reason why Nvidia does not want to use TSMC's advanced packaging is because they think TSMC's profit is too high, and they want TSMC to use OSAT's profit to do these things for them.
is exactly this, which made Jiang Shangyi realize that any customer using brand new technology is high risk. What if it doesn't work? It will cause the entire company to go bankrupt, and the person who makes the decision to choose new technology will definitely be fired. Such a risk system also makes TSMC unable to find customers. Later, after unremitting efforts, TSMC convinced Xilinx to use their advanced packaging.However, Jiang Shangyi said that Xilinx's way of using it is far from what he expected.
"My original intention of developing the technology was that I thought it would solve the performance bottleneck problem, but Xilinx just wanted to connect four dies together so that they could sell them as the next generation and they could sell them at very, very good prices. So they used advanced packaging, packaged the 4 dies together, making it a very big die. In my opinion, my innovation was not used in good places," said Jiang Shangyi.
According to reports, Xilinx uses TSMC's first generation CoWoS technology. In this generation of technology, TSMC has only one customer, and their monthly orders are 50 wafers, which puts a lot of pressure on Jiang Shangyi. But afterwards, , Qualcomm, appeared. Jiang Shangyi recalled that during a meal with Qualcomm VP, they talked about TSMC's advanced packaging technology. Later, the Qualcomm VP said: "If you sell that technology to me, I will only pay one cent for each square millimeter ."
It was this sentence that made Jiang Shangyi suddenly realize it. After returning, he asked Doug Yu to calculate how much CoWoS cost the company. The final result is - seven cents per square millimeter. "So that's why we can't sell it." Jiang Shangyi murmured. He went on to say, let's develop some technology that costs only a cent, and to achieve this, you can sacrifice performance. Therefore, the second-generation technology known as InFO was launched, which is also TSMC's first advanced packaging technology to be sold well. InFO is also the reason why Apple is fascinated by TSMC.
According to Jiang Shangyi, the reason why TSMC did not receive Apple's orders in the early stage was because Samsung provided them with a packaging solution - wire bonded DRAM on the top of the CPU and AP, which TSMC could not do at the beginning. But later, after the InFO interview, TSMC snatched Apple over.
"It can be said that one sentence saved our lives", Jiang Shangyi emphasized. Jiang Shangyi continued to say that when they proposed the advanced packaging plan at that time, they did not have CPU customers, and they were not very concerned about it. But now, AI has become a craze, and related chips can also benefit from these packaging designs of TSMC, which is also the reason why advanced packaging is becoming more and more popular.
5 Why don’t there be 18-inch wafers?
In the interview, Jiang Shangyi also explained the reasons for the failure of the 18-inch wafer.
He said that in 2013, the 450mm wafer was very popular, and Intel was also making great efforts. According to him, the wafer’s 6-inch direction was driven by IBM, while Intel was the biggest behind the wafer’s 8-inch and 12-inch directions, so they wanted to drive the 450mm wafer debut again. So
In the eyes of many people, the reason why Intel wants to do this is to increase productivity. However, Jiang Shangyi said that this is just a game of the giant - a game where a big guy wants to use the little guy, which is also very obvious. Because if you want to go into production of 18 inches, then all the equipment suppliers first need to upgrade their equipment to support 18 inches, and fabs will no longer make the most advanced technology on 12 inches. This means that on 18 inches, there will be no more small players, and they will be automatically eliminated. Second, these small players don’t need such a large capacity, and they can’t afford it. Therefore, Jiang Shangyi believes that this is a means to eliminate small players. After
, Intel began to implement his plan. Intel is also working very hard to get TSMC and Samsung to join forces. Intel has also started spending billions of dollars to prepare 450mm wafers. And at the time, TSMC was still actively pushing for 12 inches because they thought it would work well for them. But likewise, TSMC is starting to become very radical. At a certain investor conference, Zhang Zhongmou himself even gave the TSMC 450mm roadmap. Samsung, on the other hand, is very quiet. However, this cannot stop the 450mm wafer from being inflamed at all.
But one day in March 2013, Jiang Shangyi told Zhang Zhongmou: "I don't think we should promote these 450mm wafers.Because in the past, our competitors were UMC, which were much smaller than TSMC, and we advocated that 450mm could lead them. But now, we only have two competitors - Intel and Samsung, both of which are bigger than us, so this push hasn't helped us at all, but it will hurt us. "
Jiang Shangyi gave an example and said that he didn't know how many R&D engineers Intel had, but TSMC had 6,000 R&D personnel. But he knew that Intel's R&D budget was much larger than TSMC. If Intel had 8,000 R&D personnel, it would mean that the ratio of R&D personnel on both sides was 8 to 6. But if TSMC started 450mm R&D, it might constrain 3,000 engineers, leaving TSMC with 3,000 available engineers, but Intel still had 5,000. Competing with 5,000 to 3,000 will put TSMC much more pressure, which will not help the company at all.
It was Jiang Shangyi's statement that made Zhang Zhongmou suddenly enlightened. He was also very grateful to Jiang Shangyi, and then asked: "What can we do? Jiang Shangyi responded: "Mike Splinter (then CEO of Applied Materials) is here today, you may want to talk to him." It is understood that at that time, almost all equipment companies actually opposed the 18-inch wafer plan, except ASML. Because in the eyes of the Dutch giants, their lithography machines work the same regardless of the size of the wafer. But this also strengthened Zhang Zhongmou's determination to give up this plan.
By 2013 SEMICON West, Intel once again invited TSMC and Samsung to join, hoping that they would promote 18-inch wafers together. However, TSMC responded that our primary goal is to develop advanced technologies, which ultimately led to the final failure of the 18-inch wafer project. Since then, no one has ever mentioned 450mm wafers.
6 Intel failed to discuss
When asked why TSMC has made great progress now, Intel has been struggling to make it difficult. Jiang Shangyi responded: "In my personal opinion, especially from the perspective of R&D, we did not really do anything special or excellent, but we did not make any major mistakes. I think when we started, UMC was a real strong competitor of TSMC and the competition was very fierce. But UMC made a mistake in the 0.13 micron process. I also believe that Intel also made some mistakes, but TSMC did not make any major mistakes," Jiang Shangyi said.
He further pointed out that Intel and TSMC have two different cultures. Intel decided to do everything, "copy exact".
This is the most important principle in their research and development and manufacturing. What does this mean to them? This means that they are developing this technology in research and development, using this equipment, using this formula. They check it all out very thoroughly, make sure everything is great, and then go make it and you never change it. You just need to follow their guidance. It's good not to make any changes, because your risk is much lower. But the problem is that in a year, the new equipment will be more efficient. At this point, TSMC will try it, but Intel won't. Then, this causes TSMC to start to cost less than Intel.
Another obvious reason is Intel's system. They can sell wafers for $20,000 per chip because their CPU chips are very expensive. But TSMC cannot sell the wafer for $20,000, it can only sell for $4,000. So TSMC must work hard to reduce costs.
"I really respect Intel, and I think they are most willing to take very high risks. In every generation of technology, they are willing to take risks to do new things. In many key areas, such as high K metal gates, strain engineering, FinFETs, etc., Intel is always the first to adopt it. Then TSMC will adopt it in the next generation," said Jiang Shangyi.
So, on every node, Intel's performance is better than TSMC. When Jiang Shangyi was still at TSMC, he often told his colleagues: "We are behind Intel." He also pointed out: "Don't be happy because you released 10nm before Intel. In fact, TSMC's 10nm is more like Intel's 14nm." Jiang Shangyi reiterated that TSMC usually waits until Intel adopts new technologies before introducing it in its next generation.This not only includes design rules, TSMC also lags behind Intel in transistor performance.
”When I was at TSMC, I launched an initiative called Advanced Transistor Leadership, and we wanted to catch up and beat Intel in transistor performance, but that project failed. I don’t think anyone except TSMC’s group knew that. So, the biggest regret of my entire career is that we didn’t catch up with Intel. But now on the surface, TSMC can do 5nm production now, and Intel is still at 10nm. It’s certain that Intel has indeed fallen in some way, and Intel is really a bit behind, too,” Jiang Shangyi said in an interview. However, he also pointed out: "TSMC claims that they were ahead of Intel three years ago, but this is not the case."
7 was written at the end
09 after returning to TSMC, Jiang Shangyi stayed until 2013 and then went to the mainland to work. This is another story, and we will not repeat it here. But there is no doubt that from Jiang Shangyi's work experience, we have seen many changes in the industry and some milestones in TSMC's dominance process.
For example, I mentioned earlier that when Jiang Shangyi first went to TSMC, submitting articles to journals would be underestimated. But Jiang Shangyi said that when he left TSMC, when those journals received TSMC submissions, they would say, "If this paper belongs to TSMC, it will inevitably be accepted." (Original text: If this paper is from TSMC, it will be accepted.). Because at that time, TSMC had become a global technology leader.
Jiang Shangyi also pointed out that the semiconductor industry can reach today is the result of many people's efforts.
*Disclaimer: This article is original by the author. The content of the article is the author's personal opinion. The reprint of the Semiconductor Industry Observer is only to convey a different view, and does not mean that the Semiconductor Industry Observer agrees or supports this view. If you have any objections, please contact the Semiconductor Industry Observer.
Today is the 3128th content shared by Semiconductor Industry Observation for you, welcome to follow.
wafer｜Integrated circuit｜Equipment｜Auto chip｜Storage｜TSMC｜AI｜Packaging
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-