인텔의 파운드리 서비스: 2세대 EUV 리소그래피 및 전략

인텔은 2025년까지 첨단 칩 생산을 위해 2세대 EUV 리소그래피를 활용하여 반도체 파운드리 서비스를 선도하는 것을 목표로 하고 있습니다.

### 인텔의 파운드리 전략과 2세대 EUV 리소그래피
* 🏭  인텔의 파운드리 전환과 반도체 업계에서의 어려움을 파운드리 전략과 2세대 EUV 리소그래피에 초점을 맞춰 논의합니다.
* 💰  인텔의 파운드리 역량과 비즈니스 모델을 살펴보고, 다양한 IP를 개방하고 고객과 협력하여 경쟁력 있는 칩을 만들기로 한 결정에 중점을 두고 살펴봅니다.
* 🌐  인텔의 2세대 EUV 리소그래피가 갖는 정치적, 기술적 의미를 살펴보고, 더 작은 규모로 패턴을 안정화하고 그리는 데 있어 도전 과제와 잠재적 장애물을 알아봅니다.
* 🔍  인텔의 2세대 EUV 양산에 대한 경쟁사들의 반응과 반도체 산업에 미치는 영향에 대해 논의하며, 각자의 기술 발전을 추구하면서 데이터를 관찰하고 수집하는 것이 갖는 전략적 의미를 강조합니다.

Let's begin. The world's top battlefield, the semiconductor industry. Among them, I would like to hear about the foundry. In fact, because the content of semiconductor foundry is too difficult, no one talks about it anywhere. So, we have invited Professor Kwon Seok-jung from Sungkyunkwan University, the only person in Korea who can explain this. Hello. Hello. These days, the semiconductor industry is really hot. Yes, among them, AI semiconductors seem to be the hottest. That's right. Then, the biggest war between countries seems to be happening not in the foundry, right? That's correct. Then, among the foundries, isn't the most anticipated and least known, the dinosaur Intel? Will Intel, the dinosaur, reappear by genetically modifying genes, as if it were found in a fossil or Jurassic Park? It seems like this is the biggest concern here. First, Intel.

When it comes to talking about the foundry, how should we start the conversation? Now that the person in charge has given a public metaphor, if we receive the public metaphor again, I will explain a little more. In a way, Intel can be seen as a fossilized public giant, somewhat close to a huge dinosaur, like a Brachiosaurus. In fact, until the early 2010s, companies that demonstrated the maturity of technology in the so-called front-end process, commonly referred to as 'pairing,' were like that. However, strangely enough, around the early 2010s, as the well-known Moore's Law slowed down, not only did it become technically difficult, but also at a point where costs were too high, leading to a situation where they had to go hand in hand at that point.

Intel's dominance began to weaken from that time, and at that time, Intel did not neglect its foundry. However, now, in Intel's foundry, in fact, TSMC has taken over, to some extent, since then. TSMC, of course, was naturally one of the most influential companies in the world as a foundry specialist even before the 2010s. If we only consider the foundry volume that Intel produces in-house, Intel was actually closer to being a larger manufacturing company. However, now, from the 2010s, Intel has spent almost 10 years producing various logic semiconductors, including GPUs and CPUs, in its foundries. In fact, these chips worked well in the server market, but in the B2C market, which we commonly use, such as PCs, tablets, or,

Smartphones and various logic chip markets have started to lag behind. For example, in the AP market, Qualcomm, Samsung, and later even Apple with Apple Silicon chips have experienced this. In the simple PC CPU market, AMD has started to fall behind. As these events repeat, it seems that Intel, unable to exert proper influence in the sub-10-nanometer technology process, has ultimately faced the limits of such technology and manufacturing expertise. The reason for this change can be attributed to various factors. However, even within Intel, as a technology company, they naturally conduct advanced technology development, possess various expertise, and test various candidates. But when it comes to mass production, it seems that their own capacity...

Mainly focusing on improving the performance tailored to the customer, or creating new innovations, it seems like we haven't received much attention at the right moment. So, as these aspects accumulate, one of the most important things we are looking at in the 2010s is AI. Especially in AI semiconductors, what I think is most important now. Of course, NVIDIA is making GPUs, but Intel seems to have missed many timings to catch up on various AI-specific memory parts that can go well with GPUs. As Singa became CEO, Intel's structure began to change little by little. In my opinion, one of the most important signals of improvement in that structure that we should pay attention to now is in 2021.

There was a showcase by Intel, and if we go back further, around 2018 or 2019, Intel implied a gradual shift to the TSMC-style business model. Recently, Intel finally declared that they will properly introduce their new foundry service to the market. The abbreviation for Intel Foundry Services will be IFS. IFS will now publicly announce external foundries. In my opinion, when a vertically integrated company operates well, everything goes well from A to Z. So, if the design is good, the foundry works well, and if the foundry works well, the design also works well. However, if one of the two goes wrong, no matter how well the design is done, it cannot be made properly, leading to failure. Therefore, it is necessary to do a good design in advance for this business to succeed.

In the past, Intel failed to catch up with others, so they failed together. Actually, when it comes to going below 10 nanometers, they almost couldn't even produce chips. But Intel, they have their own foundry. Well, anyway, TSMC seems to be much better at foundry. Now, suddenly, they are saying to other customers, Oh, we will open our foundry to you. Then, will TSMC and Intel go together? So, before separating Intel's foundry, people are curious about how to evaluate Intel's foundry capabilities. Therefore, if we simply think about this foundry, people think it's just about outsourcing production processes well. But TSMC is different. Now, what kind of business model is Intel aiming for? Of course, we can do the whole process well. We can do patterning well. We can do lithography well. We can do all of that well. But, that's not all.

We will open up various IPs that we have accumulated while making various chips so far. This is very important. For example, if I make a modem chip, there can be various IPs suitable for that modem chip. For example, how to handle data banks, input-output channels, what materials to use, how to set process parameters, etc. These are very important know-hows that are integrated into IPs. Normally, we try not to open this to the outside. Especially now, companies like TSMC, which mainly focus on foundries, if they have obtained IPs from a specific customer through business relationships, they will not easily open them to other customers or companies, even if they are not in a competitive industry. However, unlike Intel, which cooperates with customers to create IPs, as you mentioned earlier in the conversation.

Until A, we tried everything. Yes, there are quite a few IPs fought there. Also, if you look at the IPs accumulated by Intel, when you gather them all together and create it, there is a nuance that Intel's products themselves are not that great. However, each detailed IP that makes up this is quite competitive. But within Intel, when they looked at their own products, they didn't think it was anything special. But when they actually show it to customers, from the customer's perspective, it could be eye-opening or even surprising. So, Intel originally did this in a way that only those who are already familiar with it or only those customers who are already familiar with it would use it. But now they are saying that they will completely open it up. We have this kind of thing. And we have already made a list of customer companies to do this. So, for example, what commonly known automated chip design tools are making what, what system is it.

Cadence와 같은 기업들은 새로운 칩 디자인을 할 때 인텔의 지적 재산권(IP)을 활용하여 고객에게 이러한 성능을 제공할 수 있는 칩을 제작할 수 있다는 것을 보여줄 수 있습니다. 이런 ED 업체는 반도체 설계를 하는 회사로, 자동화된 설계 소프트웨어 업체입니다. Cadence나 Synopsys와 같은 회사들은 원래 설계 프로그램을 제공하는 회사인데, 이제는 고객들에 제공할 수 있는 포트폴리오가 확장되었습니다. 인텔 파운드리를 사용할 때, 이러한 설계 측면에서도 지원합니다. 그렇죠, 단순히 파운드리의 문제를 넘어서는 것입니다. TSMC와 인텔의 다른 점은, TSMC는 IP를 보유하고 있지만, 그 IP는 TSMC의 고객사와의 협업을 통해 축적된

So, there is actually no need to reduce this IP customer, right? But now, they are going to open this up to customers. Why are they pushing it to this extent? Ultimately, it's because Intel understood the successful model of TSMC. The successful model of TSMC is ultimately because a healthy ecosystem has been formed where customers can make various products across diverse layers. Of course, TSMC also has technological competitiveness in the process, but Intel recognized the strength that comes from a diverse ecosystem that can be obtained from various customer ecosystems. Then Intel thought like this. Let's gather customers around us. But the customers originally used TSMC, so there is no motivation to go to Intel and do this, right? So Intel is throwing a carrot. Use ours.

Let's do it together. We have many babies like this, we often talk about such nonsense, you know. I didn't know what you would like, so I prepared everything. That's why please choose from here. And what's fun is not just choosing randomly, but you chose IP. You have a discerning eye. Our professional and fair engineers will partner with you on this. We will also take on the role of a design house together. Oh, originally, we did it ourselves, so it wasn't divided, it was all in there. The characteristic of TSMC is that TSMC does foundry work, but now customers design chips. Between the person designing the chip and the person doing the process, it is difficult to create a connection in a way. They are completely different areas, so there always needs to be an interface. I mentioned this last time, but there needs to be a point of contact like a design house. But that

The reason why a design house is needed is because TSMC and customers live in different worlds anyway. Intel doesn't need that. Now Intel has people who know how to do the process and Intel has experience in designing. So, I've been saying this until now, but that has been a disadvantage for Intel because they have been doing their own business for a long time. But by approaching customers more thoroughly, it is actually turning into an advantage. So, now we don't need to have a design house externally, of course, we don't need two. Now, we will take on the role of a design house and we will thoroughly tailor it to what customers want. Even in the past, Intel used to say, What chips do customers want? and then they would adjust the specs.

Rather than what Intel wants, they have been going in the direction they have tried, and then they spend more money. Or rather, there was something that resonated more with Intel's position than the demands of the customers. Now, Intel's actions or plans being shown thoroughly go towards customer-oriented minds. In a way, Intel has really come out with a clear direction in a subtle way. In a way, this time is different. Of course, many are skeptical about this part. Will Intel be able to transplant the tsmc model to its own? Will they really open up all the IP accumulated over decades? Without any compensation, rather, opening up like that could be seen as holding customers hostage. Once you start using it, you have no choice but to continue using it, or it may be a bit predetermined here.

On the surface, it may seem like creating a very diversified ecosystem with TS, but in the end, some people are now making negative predictions that it could eventually go back to recreating something like the Intel Empire. Since making chips internally has been one of their main business models so far, there are still many oppositions internally. In other words, there are those who have negative views on whether the resources used internally can be adequately shared with external channels. At least, I think we need to wait for another 2 to 3 years to see how well Intel can share its IP with these external customers. Therefore, it is really irrelevant to Intel, completely different companies from other countries making chips in completely different fields, should bring services to Intel and produce proper chips.

We should consider owning it. But it feels a bit like an open feeling in a wealthy neighborhood. That wealthy house has been collecting various things in the warehouse for a long time, not even knowing what's inside the warehouse. Oh, in the past, they bought something expensive but rarely used it. However, now that the door is open, you can see things scattered around. Wow, there's this, and that too. So, if you want to use that, combining it with this would be nice. It gives a feeling that such service is possible. It could be a different strategy from what has been growing around the concept of consignment foundry. So, I think about how Intel Foundry Services will be launched in what form and strategy.

I would like to ask for some explanations. Ultimately, as I mentioned earlier, TSMC will refer to such a business model a lot. However, before discussing such strategies, it is necessary to mention a few concerns that people are worried about now. First, although it's okay for Intel to open up the foundry, the process technology that can combine various IPs and make chips is really good. So, even if you didn't build things well in the past, you might say sorry for this kind of comment, but you probably didn't miss the opportunity. However, now that you are open, even if you use IPs, can you really do a proper sub-10nm process like Meteor Rain released last year?

When looking at the chips, in the end, it doesn't seem to be calculated in the benchmark score, or you need to use a tremendous amount of electricity to achieve the performance you mentioned. In the end, we also have to create fairness in this way to build a house like you do, in order to bring out the performance, we have to suffer a lot in a very different way, or the performance doesn't come out properly, right? Doubts arise. Yes, regarding this, one of Intel's strategies also acknowledges such criticism, and we also acknowledge that we have not made the chips more powerful enough so far. So, in a way, we acknowledge it. What does Intel acknowledge? We acknowledge that we are entering the foundry now and we acknowledge that it is still a newcomer. So, naturally, Intel's first goal now is until 2030, catching up with Samsung. So, what I mean is, catching up with Samsung alone.

That being said, according to TSMC, it is still difficult. However, Samsung Foundry, currently in second place, seems to think that Intel will try to catch up through these strategies. But, as mentioned, Samsung is not a company to just sit still. In fact, Samsung has been accumulating a significant amount of technology and know-how in the foundry business for nearly 10 years. So, how will Intel narrow this gap in this situation, or even catch up with Samsung, or possibly position themselves to catch up with TSMC? Intel is boldly throwing a decisive move in this strategy. What is it? They say they will be the first to mass-produce the second-generation EUV lithography that TSMC and Samsung are not currently using. People were greatly shocked by this news. Why were they shocked?

If you have received it, Intel actually has the first generation of EUV, which we commonly refer to as around 0.33 numeric. We call it NA. The company that has a lot of EUV lithography of the ASML NXE 3000 series, which has such numerical features, is not ASML. And that ASML, their new product, no one has tried it yet. It's still for R&D purposes. Now, gradually, one or two units will be tested by TSMC and Samsung. For example, in the case of Intel, as far as I know, the first generation of EUV, they haven't accumulated much over the past five years. It's a single-digit number. TSMC is at the level of around 110 to 120 units, and Samsung is between 60 and 65 units. So, compared to TSMC and Samsung, Intel's EUV is considered to be closer to the R grade. Of course, about half of them are entering mass production levels, but still.

One of the very important reasons for discussing EUV lithography like this is that eventually, EUV equipment is known as very advanced equipment, but it is equally demanding and prone to failure. Therefore, operating it for four hours is by no means easy. It costs about 1.5 trillion won per unit, and the second generation now costs around 4.8 trillion to 5 trillion won. Of course, the price varies slightly according to the specifications of the ordering company. The reason I am saying this is that even equipment like UV lithography is from the previous generation. Now, ARFI lithography equipment is very expensive and demanding. It is even more difficult to handle than that, and the factory is very sensitive. It is very sensitive to changes in surroundings such as vibrations, humidity, and temperature, and it is really delicate. The operators are the same as yesterday and today.

It's different. Data fluctuation is very severe, so it takes a lot of time to stabilize it. Therefore, in the process, a lot of know-how is actually accumulated. However, Intel seems to think that the process has not accumulated enough. Well, earlier, if we compare it to the EUV holding, TSMC and Intel have accumulated about ten times more aging than Intel, and Samsung has accumulated about six times more aging than Intel. But in a situation where such know-how has not been sufficiently accumulated, without fully producing and utilizing the first generation, how can they skip all of that and go straight to the second generation? Intel says they will not just produce the second generation, but will try to bring forward the mass production of the second generation as much as possible. So, just a few weeks ago.

The first equipment of the second generation was delivered to Intel by ASML, and it has been installed in my Intel fab. Now, we are going through a lot of trial and error to obtain test data. From Intel's perspective, they are planning to start mass production as early as the first half of 2025. There is only one year left. Oh, but that one year is a very short time. Typically, when the first equipment comes in, for the first six months, we can't do anything but continue testing. Thousands or tens of thousands of wafers are discarded, and an enormous number of personnel work 24 hours a day, 365 days a year, taking turns to create data, stabilize it, and verify if the light is coming out correctly. First, we need to confirm if the light is coming out properly, and second, if the light that comes out is properly guided to the wafers, and whether the stages of the wafers or masks are properly placed there.

Truly, such tremendous inertia, now, how do we see it? It's called the Entai inertia device. Many tests need to be done to ensure that such devices operate properly. In fact, overcoming such issues is one of the challenges that Intel must face now. Intel guarantees that if they really use the second generation equipment first, so by 2025, we can see the 18A level, which is the usual 1.8nm, and the 18 Angstrom level node. Then, in 2026, they will be the first to enter the 14 Angstrom level. This way, they might give a signal to the market. Intel is not dead yet. Also, Intel can finally produce the best single-pattern in nano-underlayer that TSMC or Samsung have not been able to achieve yet, even at the mass production level. For this, Intel can use the second generation EUV first and is working on it. This way, we have IP and process technology.

Now, we are not much different from the two companies in front of us, but rather, we want to send a signal that we are somewhat ahead. This is Intel's hopeful circuit sound. But if I may say, in reality, Intel did not have much presence even in the EUV era, and Samsung and TSMC, who are good at making chips, are facing some difficulties in catching up in the nano field. However, Intel just said they will do 1.8 nanometers with just buying one equipment, so if that works, why can't they do 3 nanometers? And if they can't do it, how did they achieve 1.8 nanometers in just a year? Here, one should not misunderstand a little in the market. Intel saying they will do the 2nd generation does not mean they won't do the 1st generation. They will continue with both the 2nd and 1st generations. While doing the 1st generation, Intel will now be able to compete with things like Intel 3, Intel 5, Intel 4, which are like TSMC's 5 nanometers, 4 nanometers, 3 nanometers processes.

Those chips will continue to be made, but those chips will be turned towards the customer's internal foundry side. Of course, after the first generation is sufficiently depreciated, they must extract sufficient value. Once it stabilizes to some extent, it will be used to create a customer base. However, the huge costs involved in making such chips as objects will be redefined as internal chip-making costs. They will try to redefine their strategy by using the knowledge gained there, transplanting it to this generation, or deploying trained professionals to this generation. However, the problem now is that the equipment is much more expensive than expected, as I mentioned earlier, and Intel is now just.

A total of six second-generation reason equipment came in batches. Even if the price of just one of these is around 3 trillion won, now, it's not just about putting in one or two equipment, but a variety of processing equipment needs to be included. It will be about 1,500 units to 1,600 units per one. So, now, the cost of just one fab will easily exceed 20 to 30 trillion won. From Intel's perspective, to utilize someone at this level, at least three more of these need to be erased. Then, from Intel's perspective, they need to invest nearly 100 trillion won at most. However, one of the dilemmas for Intel is that they cannot do this with their money alone. They will talk to the U.S. government. When they say they will reshore and grow semiconductors in the U.S., there is no one else but us, right? So, shouldn't we receive more subsidies from them?

Please, Intel is aiming to receive at least $10 billion in subsidies from the U.S. federal government, and Intel continues to pressure. Since industrial policies are being implemented in this way, if you want to definitely increase manufacturing more in the U.S., there is only a lack of chip laws. Please make Chips Act 2. Please make Chips Act 3. So now, officials from the Biden administration's Commerce Department, like Raimondo, are going to Intel events and actually supporting, saying, Yes, we need Chips Act 2. The problem is that the Biden administration's term is only about six months left. Of course, now President Biden may be re-elected as the 47th President of the United States, or Trump may return to the White House again. In this situation now, whether Biden or Trump

Many people are skeptical about how long the U.S. federal government can continue to pour almost an equal amount of government-level subsidies to a specific company like China. The biggest difference between China and the United States is that both countries are major powers, and both governments want to promote this industry. However, in China, they can push through top-down policies, somewhat like a single-party dictatorship. But in the United States, it's different. The administration changes every four years, there are midterm elections, and there are tremendous lobbyists. Therefore, it is uncertain whether it is possible to keep channeling subsidies to a specific company on a one-time basis. However, if this continues, for example, in rounds 2 or 3, accumulating up to 10 billion dollars, 20 billion dollars, or even 50 billion dollars, 100 billion dollars, or 1 trillion dollars, this is a government-led.

In another dimension, it will be a very burdensome task. So, in fact, from Intel's perspective now, in a way, when the government is willing to provide subsidies, we should try to receive them as much as possible and invest it in KPX smoothly. Let's enter fixed costs anyway, let's do it as soon as possible when the cost is relatively low, because as time goes by, equipment, land prices, labor costs, etc., will all go up. Therefore, we are also considering maximizing the utilization of these government subsidy policies as part of our strategy. I understand. In the process of Intel's takeover, there will be some financial issues, and we will receive some support from the U.S. government. In fact, the first option is the U.S., and the second is Korea, so from the U.S. government's perspective, even if TSMC and Samsung build factories in the U.S., Intel may want to focus on Intel.

I don't have a good sense of money, but I think ASML really sent this generation to Intel first. I think it was influenced by the power of the U.S. government or Intel's dominance, knowingly or unknowingly. Yes, that's correct. However, will Intel, which struggled with the first generation and couldn't do well with the first Eve, be able to succeed with the second generation and create 1.8L? We need to consider these factors or potential obstacles. What are the key points we should focus on? Well, first of all, I will talk about it from a political context, and then I will talk about it from a technological context. In terms of the political context, obviously, due to the power struggle between the U.S. and China, companies or countries around the world that are putting some effort into global semiconductors have no choice but to keep an eye on the two countries.

In a situation like this, especially when a powerful country like the United States is trying to foster its semiconductor industry, it is not just about attracting factories. For example, if such factories are attracted, various equipment, as mentioned earlier, needs to be installed. Companies that are good at producing such equipment include American companies like Applied Materials or Lam Research. However, even companies like ASML from the Netherlands now need to invest more here. One of the things mentioned in the United States is not just to stop at delivering equipment. What they mean is for the United States to further expand ASML's R&D center. So, now, it's not just about simple maintenance, but coming here and, led by Intel.

The semiconductor manufacturing industry in the United States is now talking about stepping up to the next level, entering the gear cluster, and making more investments. Even in these devices now, companies like TSMC and Samsung, such foundry companies, are inevitably making a lot of investments in the United States. This may simply be targeting the U.S. market, but in a way, it can also be seen as pressure that is not visible to the U.S. government. However, this cannot be ignored because the United States cannot simply exert pressure by force alone. The United States also has some technological strategic points that can influence the key demands of these companies. For example, people think that companies like Intel, ASML, and AMD have developed all technologies.

Many technologies have origins in the United States. For example, the light source we commonly talk about, the light source was initially developed using a method called laser-induced plasma or laser-produced plasma, also known as LP method. Before these specific areas were created, it was necessary to test what optical characteristics these areas actually had. To do this, a method to create light was needed. Where was this light created? It was created at a synchrotron accelerator. However, the country that has excelled the most in operating and accumulating such accelerator technology is indeed the United States. Therefore, the United States has national labs such as Brookhaven National Lab, Lawrence Berkeley National Lab, and Lawrence Livermore Lab. Among them, there is the Brookhaven National Lab.

In the same place, there is a lot of discussion about synchronous debate. There are technologies developed there, such as light sources, commonly known optical technologies, stage technologies, stepper technologies, and mask technologies. If you follow the seeds of these technologies, some are held by IBM, some by universities, and some by research institutes. These are now coming together one by one and are being incorporated into a significant portion of the lithography scanners produced by ASML. Therefore, from the perspective of the United States, if they say, You can export the equipment you made, but do not use the technologies we developed, ASML may find it difficult to complete the equipment. So, from the viewpoint of ASML, some pressure from the U.S. government might be seen as...

Politically and technically, we need to consider some functioning aspects. Another issue that arises as we transition to the 2nd generation now is that what was difficult to stabilize even in the 1st generation equipment becomes even more challenging to stabilize in the 2nd generation. One of the biggest reasons is that, as I mentioned earlier, the cost of numerical aperture increases. However, the advantage of the increase in numerical aperture is that we can slightly increase the resolution. In other words, for example, what used to be produced at 13 or 12 nanometers in the 1st generation can now be made even smaller, at 10 or 9 nanometers. But there is a cost to this. What is different in this cost is ultimately that as the cost of the new aperture increases, it leads to more photon-induced stochastic effects.

When we talk about probabilistic paths or events, unwanted side effects can occur. Now, when such things happen, what becomes a problem is, for example, in controlling lithography, what we need to do within the PR method is to accurately notify the photochemical reactions as much as we want. However, as time goes on, it becomes increasingly difficult to control this when the next generation comes. This means that problems may arise, such as occurrences in areas where photochemical reactions should not occur or not occurring in areas where they should. Also, in the process of creating these photochemical reactions, if the energy of the photons is too high, they may disturb the molecules in the chemical substance called PR. However, just disturbing them is not a problem, but while doing so, unwanted secondary electrons, radicals, or similar things may come out.

Starting, they can just walk around and listen quietly. It's okay if they just listen quietly, but when they walk and generate heat every time, it can cause some thermal deformation. So, there may be some changes in the quality of the pattern from 13 nanometers to 9 nanometers. When it becomes 9 nanometers instead of 13 nanometers, a much rougher pattern emerges. For example, patterns that should not be connected are connected, and patterns that should be connected are disconnected. When these issues occur, inevitably, there will be a decrease in yield. Therefore, now with Intel's 2nd generation, there is a high possibility of introducing the 1st grade, or as mentioned earlier, 18-longstrom or more precisely, 14-longstrom 2nd generation. Although reducing the patterns may face difficulties, stabilizing and drawing these patterns consistently.

Making it consistently in high demand will be even more difficult. So, to solve these technical challenges, traces of researching such fundamental issues surrounding EU lithography within Intel should be shown. Of course, it could be that they are not disclosed yet, but in fact, such traces are not clearly visible. However, from Intel's perspective, probably from this year, they will start presenting little by little that they also have this information, not just keeping it to themselves, at conferences related to lithography like SPIE or NGL. Even if customers see such data, ultimately, regardless of that, Intel will gradually start presenting, saying, This is a common technological issue faced by the industry, and here is Intel's solution, within a year.

It seems difficult to overcome. Of course, it will not be too late to judge after the data comes out later. Therefore, there will be companies that are interestingly watching Intel using second-generation equipment first. Well, TSMC will be the most interesting to watch first, and second will be Samsung. In particular, TSMC collects data dominatingly well on specific equipment or architectures or materials that no one has tried before. For example, Samsung said, Now we will not use finFETs from now on, we will go to gate-all-around. And from now on, we will change the conditions of EUV a bit. However, as soon as such talks come out, TSMC can immediately call ASML. What is this about? Is there any information that we don't know, ASML?

I ask them, but if I ask, they shouldn't tell me. They should just know by looking at my expression. Because TSMC is the biggest voice. Um, what happens daily is like this. Earlier, I mentioned that the difference between introducing EUV is roughly 2 to 1, right? So, what ASML is going to do in Korea is to establish a center worth 200 billion and another one worth 4 trillion. Then people would think, Oh, they are investing billions of won. At the same time, ASML will build a 1.2 trillion and a 1.5 trillion center in Taiwan. It's inevitable. They have much bigger customers, and they need to produce the most data. So, ASML has to focus on two things. First, these big players have to keep buying their equipment. Second, they only make equipment, so they cannot test how their equipment actually works inside the fab and what kind of results it produces. Ah, they cannot test the actual back-test data. They must work closely with these big leaders.

It can be correct. ASML's data is very valuable because with that data, we can reflect on it when making equipment for the next generation. However, if TSMC does not cooperate on such data, or if ASML keeps excluding them, then ASML is really doing something big. That's why we call ASML super. From ASML's perspective, they have no choice but to be very cautious of TSMC. However, for example, if Samsung or Intel launches the 2nd generation first or uses some new materials, then TSMC will start to feel pressured, watching interestingly and starting to pressure ASML. What kind of know-how did you give them? Did you collaborate with them? They might think, What happened? Come and report back. Then, things start to heat up, and they start talking about it all in Taiwan. This has nothing to do with TSMC.

There is no need to worry about it. They might say things like that. So, now, at the headquarters of TSMC in Taiwan, competitors who have similar levels of technology are always faced with what choices they make. Especially this time, Intel's second-generation EUV mass production is starting right away, and it's even going into six devices. It will be very interesting to watch and they will continue to constantly monitor this data. Of course, Samsung will also be monitoring it, and especially at Samsung, there are many people who used to work at Intel, and vice versa, there are also people who used to work at Samsung and are now at Intel. In fact, since these people have some knowledge of what is happening at Intel, they are analyzing what technological limitations and advantages can be obtained in this way. Ultimately, however, all the answers are at Intel.

When the chips produced by the second-generation EU lithography come out, you reverse-engineer the chips, cross-section them, analyze each element one by one, and look at them through imaging. At that moment, it becomes evident if it's just a pig or not. In my opinion, Intel probably wants to show only the best chips, but once mass production starts, Intel may not be able to control that. Right? Once mass production starts, the answers will come out, and it will be possible to estimate the yield to some extent. Well, I see. Actually, Intel being the first to go is threatening, but the equipment is very expensive, and new paradigms and materials will come out, so from the perspective of competitors, it might be a good strategy to watch and collect some data while going their own way with 3 nanometers and then transitioning like that. It was quite impressive to say that this could be a good strategy. Semiconductor part 2.

[반도체 특집 1부] 진짜로 ‘왕의 귀환’일까?

복잡하고 어려운 지식 컨텐츠를 쉽고 재미있게!
압도적 지식 컨텐츠 압권팀 입니다.
재미있게 봐주세요😁

👇 각종 문의는 아래 연락처로 부탁드립니다 👇


압권 Apkwon

반도체 파운드리 산업에서 인텔의 역할

한때 반도체 파운드리 업계에서 독보적이었던 인텔은 무어의 법칙의 둔화와 높은 생산 비용으로 인해 특히 소비자 시장에서 영향력이 감소하는 문제에 직면해 있습니다.

파운드리에서 다양한 로직 반도체, 특히 AI 전용 메모리 부품을 생산하는 방향으로 전환한 것이 구조조정과 미래 성공에 결정적인 역할을 했습니다.

인텔은 인텔 파운드리 서비스(IFS)를 도입하면서 파운드리 시장에서의 입지를 개선하고 효과적인 설계 및 생산을 통해 성공을 보장하기 위해 TSMC 스타일의 비즈니스 모델로 나아가고 있습니다.

인텔의 파운드리 비즈니스 모델 및 IP 개방

인텔은 이제 파운드리를 다른 고객에게 개방하고 축적된 IP를 제공할 계획이며, 이는 IP에 대한 TSMC의 보다 보호적인 접근 방식과 대조적입니다.

인텔은 자사의 IP를 제공하고 설계 파트너로서 역할을 함으로써 다양한 생태계를 조성하고 TSMC의 고객을 유치하는 것을 목표로 하고 있습니다.

인텔의 전략은 칩 설계와 제조 공정을 연결하여 파운드리 사업에서 독특한 접근 방식을 제공하는 것입니다.

인텔의 고객 중심 접근 방식으로의 전환

인텔은 이전의 자체 중심 전략과 달리 고객의 요구에 맞게 제품을 맞춤화하는 디자인 하우스의 역할을 맡는 고객 지향적 접근 방식으로 전환하고 있습니다.

인텔이 TSMC의 모델을 복제하고 외부 고객과 IP를 효과적으로 공유할 수 있을지에 대한 우려와 회의론이 있습니다.

인텔의 목표는 2030년까지 삼성을 따라잡는 것이며, 이는 업계 경쟁에 대한 초점의 전환을 의미합니다.

2세대 EUV 노광 기술을 통한 인텔의 대담한 움직임

인텔은 TSMC와 삼성을 따라잡기 위해 2세대 EUV 리소그래피를 최초로 양산하는 등 과감한 행보를 보이고 있습니다.

인텔은 경쟁사보다 훨씬 적은 수의 EUV 유닛을 보유하고 있지만 2025년 상반기까지 대량 생산을 시작할 계획입니다.

인텔은 2025년 18A 수준, 2026년 14앵스트롬 수준에 도달하여 나노 언더레이어 생산에서 강점을 보여줄 것을 목표로 하고 있습니다.

인텔의 반도체 생산 및 정부 보조금 계획

8나노미터를 달성하여 TSMC의 나노 공정과 경쟁하는 것을 목표로 하고 있습니다.

인텔은 2세대 칩 생산에 수조 원을 투자할 계획이며, 미국 정부에 100억 달러의 보조금을 받기 위해 압력을 가하고 있습니다.

인텔과 같은 특정 기업에 대한 정부 보조금의 지속 가능성에 대한 우려가 있지만, 인텔은 이러한 정책의 혜택을 극대화하여 반도체 생산에 투자하는 데 집중하고 있습니다.

미국의 전력 투쟁이 ASML 및 반도체 산업에 미치는 영향

ASML과 인텔의 관계는 글로벌 반도체 산업에서 미국과 중국 간의 권력 투쟁의 영향을 받았을 수 있습니다.

미국 정부가 ASML의 R&D 센터를 확장하고 주요 기술 개발에 대한 통제를 강화하려는 움직임은 ASML의 향후 장비 생산에 잠재적인 장애물이 될 수 있습니다.

미국 정부의 압력은 ASML이 장비를 완성하고 글로벌 반도체 시장을 공략하는 데 상당한 영향을 미칠 수 있습니다.

EU 리소그래피용 2세대 장비 안정화의 과제

2세대 장비의 수치 조리개가 증가하면 해상도가 높아지지만 광자에 의한 확률적 효과도 더 많이 발생합니다.

이로 인해 광화학 반응을 제어하기 어렵고 원치 않는 부작용이 발생하여 궁극적으로 생산되는 패턴의 품질과 일관성에 영향을 미칠 수 있습니다.

인텔은 이러한 근본적인 문제에 대한 연구를 점진적으로 공개하고 리소그래피 컨퍼런스에서 해결책을 제시할 것으로 예상됩니다.

인텔의 2세대 장비 도입은 리소그래피의 새로운 기술적 과제를 해결하려는 다른 기업, 특히 TSMC와 삼성이 면밀히 주시할 것입니다.

ASML의 전략적 움직임과 TSMC 및 인텔에 미치는 영향

ASML은 TSMC와 인텔 같은 대형 고객을 지원하기 위해 한국과 대만에 수십억 달러 규모의 센터를 설립할 계획입니다.

이 회사는 차세대 장비에 필요한 귀중한 데이터를 수집하기 위해 이러한 주요 업체들과 긴밀한 관계를 유지하는 데 중점을 두고 있습니다.

EUV 리소그래피 생산 분야에서 TSMC와 인텔의 경쟁이 치열해지고 있으며, 인텔의 초기 대량 생산이 경쟁사에 위협이 되고 있다는 기술 분석이 나오고 있습니다.

이러한 위협에도 불구하고 TSMC와 같은 경쟁업체는 자체적으로 전환하기 전에 데이터를 관찰하고 수집하는 한편, 인텔의 대량 생산을 면밀히 모니터링하여 잠재적인 인사이트를 얻을 수 있습니다.