This is the agenda for the presentations this afternoon. A small strategic overview by myself, market trends, Chris. Wafer level assembly, Peter, and Christoph, Die Attach, Shihun Packaging, and Bart, plating, followed by a Q&A. Next slide, please. Let's hurry up a bit. Hybrid bonding, that's where you all came for. Well, I hope you also came to see the other Besi products, but anyway, we're in the midst of a unique transition in this industry, where we see the hybrid bonding being adopted as the next mainstream technology for interconnect of single devices, but also organized in a chiplet architecture for dimensions connected to below 5 nanometer chip design.
That is in the past 12 months since you were here, already set, not anymore a question of whether this will come to the mainstream, but the question is only how fast and for which applications. Some of that we will explain a bit more in Peter's presentation, but that's, of course, key. Then you've also seen the first high volume TCB chip to wafer. To make that more clear, again, we have entered into the TCB world about 12 years ago, focused on stacking memories at that time, building memory packs, and now continued with a new challenge is on connecting CPUs as well.
We see also in industry, again, at an inflection point in the importance of semiconductors for our society, is, not only because of the technology enabling many more applications and faster, and supporting a competitive environment for our economies around the world, but it has also become ever more critical. That has put more focus onto, what is happening in this industry, where it is happening. As you all know, more and more, the dependency on Asia is becoming too risky, so the world is developing plans to establish more onshore, new technology developments, both in Europe and in the U.S.
That combination is helping us tremendously in understanding where this industry will lead, and especially the back end, which so far has been, you could say, separately organized, but through this hybrid bonding, becoming ever more integrated into the front end of semiconductor development. Next slide. We also see that demonstrated, I would say nearly every week. We have articles in major newspapers, also many conferences. The world is looking for a way to do 2 things. First of all, to move this industry in the next technology generation, where advanced packaging, and especially hybrid bonding and chiplet architecture, is seen as to be the major enabler of that. Number 2 is how the world will reorganize in order to develop over time, less dependency on the Asia-Pacific region, and certainly to have control over technology onshore.
BESI is in the midst of that. It is both our customers, also in the partnership with Applied Materials, ever better positioned in being part of that community, having the opportunities for the years to come. Next slide. If we look at more down to earth, what's happening, we are still in the midst of a downturn. A downturn which started at the beginning of last year. Some people argue we should be somewhere in the trough, and things should become better from now on. Who knows? This industry is highly unpredictable, it could still go on for some time. If you look at basically how we've managed so far through this downturn, you all know the numbers we released end of April, things have not gone, let's say, too bad for BESI.
Revenue has declined, however, margins have gone up. We've spent enough time, I think, on that, in explaining our numbers after the release end of April. In the second half of this year, what will happen? Of course, the big question: will the tide turn? There are some indications in China, which we also mentioned already end of April, which indicates some increased demand, but still on a relatively isolated basis. You can't say there's an overall recovery, certainly not outside China. There are some industry segments which still grow positively, like automotive. Basically, 2023 is a down year. If we look at the statistics, typically, the down cycle lasts six to eight quarters.
Somewhere towards the end of this year, beginning of next year, we should see a next cycle take off, which is predicted by all the independent analysts in this industry, TechInsights, Yole, whoever you want to follow. Important, even more important is how are we positioned for this next cycle? It's fair to say, and we'll spend some time in the presentations, we have developed not only for the latest technology, but also for all of our existing products, next generation improvements, applications, and that should give us an opportunity to benefit again, better from the next upcycle like we've done in many cycles in the past. Those are key points to realize. In addition to that is our cost structure.
As you know, our margins are definitely at the higher end of the spectrum, That means, or that's only possible when you have a constant focus on cost. In every downturn, we reassess our manufacturing organization, our supply chain organization. Through the COVID period, we've had to add many additional suppliers, which were qualified because others were not able to deliver, and so we have a higher selection capability, which is also on a full swing. From a cost point of view, we also anticipate to be in a better position, despite that we have, of course, this inflation analog. This inflation works on both ends, on the top line, but also on the cost structure. Far, we are able to manage that in a positive way, but that is also something to look at it carefully. Next slide.
Here we see a bit of history, 2006, the first year this company was in the Besi family, 2005. There you see, when we announced to the world in 2005, we've built the best in the world, flip chip, etc. There you see we go, but this was the banking crisis. Okay, it's a bit of an excuse. Anyway, in all the years you see the swings, 40% down, 50% down, 50% up. You also see gradually improvement of margins, gross margins, now well above 60%. That is, again, a combination of focus on the winners in the three segments, communication, data, and automotive, and at the same time, improving on an ongoing basis, our cost structure and our operating model.
If you take the midpoint of the guidance and you take the last 12 months, we've already fallen off a cliff of EUR 750 to about EUR 600. If you look at past, that also is a trend which you can recognize. We're not worried about that. Just that you know, it's a wonderful cyclical industry. Next. We see here a bit of, yeah, let's say, comparisons in efficiency, profit efficiency over cycles, both in revenue, orders, gross margin, and then baseline operating expenses. Always keep your expenses down, is the message. A consistent line which doesn't need much more detail. Next. We also see that here in market shares, a gradual improvement of Besi's market share overall.
For some of you, who do not know yet, we're not that much interested in market share. We're interested in margin. Still, if you have the right margin, you will also see an increase in market share. In the addressable markets, strong increase, again, 22 over 21. Die attach, very consistent and very strong. Packaging also. In total, and we see here the revenue split, 79% is die attach, 21 packaging and plating. It's always around 80, 20. All in all, a very positive development in the last upcycle. Next. If we take out of the total back-end markets, the addressable markets, and especially the advanced package placement, for years, we've been focused on this advanced packaging. That's also the history of Besi from the start.
If you focus on that consistently, it simply offers you the best growth opportunities and also the margin potential. This is details which we share each time, and the data was released in the last 2, 3 weeks from TechInsights. Next slide, please. What's important here is that those areas where we are focused, so the advanced packaging in particular, is where the growth is, and that's not surprising. If you do that consistently, you also increase your share overall. We can see in the years to come, the big opportunities are the growth in the advanced die attach arena. Next slide. Important is, of course, shareholder returns. All of you know that over all the years, we have always maintained a very healthy balance sheet, simply because margin-driven.
Anything above 20% of revenue, we distribute to shareholders in form of dividends or share buybacks. Simply to keep a balance sheet safe, you see here the threshold of 20%, so that we can weather any storm, but also are prepared for any strategic step to finance that on our own balance sheet. Over time, a tremendous return to shareholders over all the years. Also you see here the development of net cash and gross cash over the past five, four years and a quarter. Next. We see that the capital allocation then to shareholders, it's amazing, EUR 1.6 billion since 2011 in dividends and share repurchases, and still maintaining that 20% at a minimum in net cash, 20% of revenue. Next.
Our model remains very much intact in this downturn. What we shared last year, the billion plus, plus. More details to follow. Although our revenue has declined in this down cycle from EUR 750 to EUR 600, if you take the payments, we still maintain to our billion plus, plus revenue model. That means in addressable markets, where you saw in the last numbers, just below 40%, that it should be slightly above 40%. Our margins maintained well above 60%. For gross margin, 60%-64%, operating 30%-45%. Headcount split, Asia, Europe, 80/20, not much change. The Scope 1 and 2. Emissions, 62% reduction, and global energy needs 75% from renewable sources. That's our business model. Next slide, please. That leads us to the next chapter.
Please share with us your insights on where this industry is heading.
Thanks a lot, Richard. Hello, everybody. Good afternoon, and hello as well to those of you listening in remotely. My name is Chris Scanlan, I'm SVP Technology at BESI. Just a brief introduction for those of you who I have not met. I've been in the industry for about 27 years now, working all the time on advanced technology development. I have about 70 patents, all related to advanced packaging. I've been with BESI now for about 2.5 years. I came to BESI to really help change the world with hybrid bonding and all these advanced Internet technologies. I'm excited to talk to you about that today. With that, let's move on to the first slide.
We're fortunate to be in an industry that's growing rapidly, and we have some projections from Gartner and other firms that we're actually expected to see higher growth rate in the future or the next decade or so. We're moving from an era where we had Moore's Law helping us out on cost reduction year by year as we increased the number of transistors in a chip, and at the same time, reduce cost per transistor. We're now kind of seeing the end of that trend. The other thing that's happening is we're having a trend towards big data and AI, generating data and information for us automatically. It's no longer us as humans relying on ourselves to create the information and data. It's our machines that we're building doing that.
There's simply more data, more information has to be processed or transmitted. This is a positive trend for our industry. Next. That said, yeah, we're in a cyclical downturn. We had at least two really great years, and now we're paying the price with a downturn. You can interpret the chart as well as I. Maybe we're at the bottom, not sure, but Richard said, we're experienced in managing these downturns, and what we're doing now is laying the foundation for the next upturn with the technology and the services that we're developing. Next. This is seen in this data from TechInsights, the semi CapEx overall market. This is front-end CapEx, but also including back-end, is expected to decrease about 17% in 2023.
The key points is that, you know, the biggest percentage decreases in memory, logic, and foundry is not decreasing as much. If you look over time, over the last five years, fast logic and foundry has actually grown faster than the memory and analog methods. We're more exposed, I would say, to that foundry and logic portion of the market. Next. If you look at just what's been announced in terms of big investments and fab capacity worldwide with these leading players, we see, you know, linearized over the announcement period, which one of them about EUR 80 billion-EUR 85 billion in annual investments each year on new fab capacity. We're being supported by government initiatives, CHIPS Acts, various different regions, as well as indirect investments from people like Apple and so on, to further increase that number.
We see a positive outlook for the long-term future. Next. Of course, we do anticipate a downturn in the assembly equipment market this year of about 23%, maybe 22%, depending on the analyst. A rebound is projected in 2024, 2025. This year, you can see that the downturn looks to be at a higher level certainly compared to 2019. Next. Looking specifically at the advanced packaging markets, Besi is really focused in terms of our product strategies. We certainly are R&D spending on advanced packaging. This is data from Yole Development that recently was published showing a 40% CAGR in advanced packaging revenue. This is not the equipment revenue, but the assembly revenue associated with advanced packaging.
They define high-end performance packaging as those things like, fan-out, CoWoS, hybrid bonding, and those kind of things. These are all the areas that we're playing in. This is very good for our future. Next. The end markets that we're serving principally are mobile internet, computing, automotive, and industrial. In all these areas, there are applications that are driving continuous growth. We'll talk about some of them today, cameras. We'll talk about generative AI and over again, on the networking side. We'll talk a little bit about auto, vehicle, electrification and automation. All these markets are continuing to be important to Besi. Computing is now the largest market, which is a change from last year. Next.
Now, I want to talk a little bit about chiplets, because there's one trend in the last couple of years that has really brought focus to our R&D efforts at Besi, and it's chiplets. What are chiplets? Chiplets are the idea of taking an SoC, which year by year, we used to cram more and more stuff into the SoC, more functionality with smaller transistors and reduce costs. That capability is simply slowing because it's getting harder to reduce the size of transistors. It's getting actually very difficult to implement a new node and keep the same cost per transistor. In fact, transistor cost is going up on a per transistor basis.
Because of that, there is a trend now towards splitting this SoC functionality that used to be one chip into multiple chips, in such a way that we can use these most advanced nodes only for those functions that really can benefit from those nodes, because some of them can't. Like SRAM, as an example, does not shrink anymore as you implement new nodes. Only this analog functionality, as you can see from this chart. Therefore, by using chiplets, we can reduce the cost of the overall system by combining the most advanced nodes with trailing nodes, maybe even devices from different fabs, all in the same package. When we do that, we need to have a very high-density interconnect to put them back together inside the package. This is one example from Intel called Ponte Vecchio.
This is a device GPU for data center and supercomputing that has 47 active chiplets, which is still kind of the best example, actually implemented in the market today. I'll talk a little bit now about what kind of interconnects. Next. The overall market for chiplets is up until now, been relatively small, but this data from TechSearch estimates that the CAGR for chiplet packages will be about 100% over the next 5 years. It's really, again, driven by things like this Amazon device, that used to be a single chip, are now split up, taking out functions that aren't really benefiting from this most advanced node and reassembling them into a package. The way that we do that assembly can be various, but we anticipate this to be prevalent and really very commonplace in servers and CPUs initially.
This big jump in their data, according to TechSearch, they anticipate a transition as well in mobile. Next. How do you put these triplet packages together? These things can be very complex. That package I showed you from Intel, again, had 47 chips. Inside that package is many different interconnect technologies at play, not just one, and they all complement each other. In this example, this is kind of a generic example. We might start with hybrid bonding, this salmon-colored stack, 2 chips or multiple chips directly connected using hybrid bonding, copper-copper interconnect, providing the highest density interconnects. Even if we have that, we then have to take that stack of chips and assemble to something else. Oftentimes, that might be done using chip-to-wafer TCB, taking these chiplets and then further attaching them to interposer.
That interposer might have other things on it. It might have HBM, it might have other chiplets, all assembled using flip chip or TCB. We have to encapsulate, right? We have to make that encapsulation around the chiplets, planarize the whole thing to make it thermally efficient. That interposer has to be assembled onto a substrate. That also takes a die attach machine that is capable to handle a very large component like our Evo. Basically, at Vestige, our strategy is to provide a full solution to allow customers to assemble these complex packages with less equipment. Next. The most important technology that we are working on today really is hybrid bonding. Hybrid bonding is the third major kind of interconnect in our industry's history. We started with wire bonding, and then we introduced flip chip.
When I started in 1994, I was working on flip chip at Motorola, and it was really transformational because before that, we had wire bond pads on the edge of the chip, and designers thought that way, right? The people that were designing the chip, the circuit, they knew they had to put the I/O pads on the edge because that's where they could be connected. When we said, "Hey, you know, you can put the contacts anywhere you want on the chip," that was really difficult for them to grasp at first. Some of the early flip chip even had bumps on the peripheral pads because the designers couldn't think differently.
Hybrid bonding is even more transformational because what it's going to allow us to do is design in three dimensions with much higher contact density than what's been possible in the past. What is hybrid bonding? It's basically a direct copper-to-copper interconnect between chips without any solder or other kind of wiring. We can do it at extremely fine pitches. Today, 6-9 micron pitch, in the future, down to 1 micron pitch and below. What this provides is the highest bandwidth, lowest power, and highest efficiency in our connect that's possible. This, in turn, is going to allow new architectures, new ways of thinking about design that are not possible or have not been possible until today. Next. Some of the applications that we see. First, high-performance computing.
We see, you know, this 3D V-Cache that we'll talk about, but in the future, many other implementations or ways of architecting high-performance computing systems. Then as a next step, stacking high-bandwidth memory. Peter will talk about that in some detail. There's the application processors for phones and also opportunities in displays, photonics. Next. Last year, we talked about AMD's roadmap, and they, of course, came out with 3D V-Cache last year. Just to remind you, the picture on the, on the left is a picture of the 3D V-Cache architecture that AMD introduced with hybrid bonding. Fundamentally, it takes a CPU chip, which is the bottom chip. On the back of that CPU chip, they mount a SRAM device, which basically triples the SRAM capacity of the chip.
I just wanted to remind everybody, and again, that when you're counting bonding steps, I know a lot of you are counting bonding steps, it's not just 1 hybrid bonding step, it's 4. Why is it 4? We mount the SRAM to the logic device using hybrid bonding. That's obvious, but there's 2 other spacer chips that also have to be mounted. They're just blank silicon, but to our bonder, they look the same UPH and everything. There's a phantom bonding step that we don't talk about much that is used as kind of a temporary bonding step. 4 bonding steps for each 1 of these small chiplets. AMD can take that small chiplet, they can combine it in many different ways. In the middle part, this is the new implementation of their gaming processor. You can see there's 2 CCD chiplets.
In this case, only one is hybrid bonded, the other not. I just want to stop there and point out something that's really brilliant about what they've done. This is a completely new technology. Hybrid bonding die wafer had never been done, so you might think AMD is taking a huge risk. The way that they did this is they implemented the design in such a way that they can actually use the same chip with or without the extra SRAM. This allowed them to qualify, their technology, the manufacturing, process, and improve the maturity. Now that they've done that, they're now introducing products like MI300. MI300 is a GPU-CPU combination. It's used for, supercomputing applications, and this is really architected around hybrid bonding. There's no other way to build it. It proves that hybrid bonding is mature.
AMD is not the only one thinking about this. AMD has proven the feasibility in production, and there are many companies now looking at designing systems like this. Next. We talked about capital intensity last year. What's new? AMD has now implemented another generation with even more process steps, called the Genoa-X. We're now up to 50 placement steps in a single package. Next. Another key development that we see is a lot of interest in chiplets, around standardization of the chiplet interfaces. There are multiple organizations working on this. The most mature is called the UCIe. You can see the companies that are engaged in UCIe, but all these companies are working on ways to standardize the chip-to-chip interface, both the physical interface and the interface in terms of software and IP. Next. We've had a few questions about the 2.5D interposer as well.
This is actually a pretty small market today. You can see the market size of 2022. But a 2.5D interposer, it's sometimes also known as CoWoS, involves placing flip chip die or other kinds of die onto an interposer and then further assembling that interposer onto a package substrate. The chip-on-wafer, that's the part where we place a chip die onto the wafer, and then the on-substrate is taking that interposer and further attaching it to an underlying substrate. We expect this market to grow rapidly. This is new data from Yole, just from May, showing a rapid growth rate in this step. BESI does support multiple process steps in this flow, but as I said up until today, it's been a relatively small market. Next.
Yeah, in the mobile space, there are several different parts of the mobile phone and wearable devices that we're interested in tracking. Christoph mentioned a little bit about advanced camera technologies. There are new form factors, new types of cameras that we're working on. There's pro cameras under display panels. In the mobile arena and the basically the application processor side, there's a lot more AI capabilities being built into the application processor, and that will continue in the future. We're looking towards a chiplet architecture for the AP. This is one possible arrangement using hybrid bonding. This will be a major inflection for us in the future. Instead of having 1 die in a base package, having 3, 4, 5, 6 die in a base package instead.
We've all seen some news, I think yesterday, about AR, VR chipset or headsets. This is also an area where BESI is engaged. We can't talk too much about the details, but this could be an interesting market going forward. Next. In automotive, yeah, the two major trends are electrification, the powertrain, and automation, so autonomous driving and ADAS. We simply see a very high growth rate in both of those areas. In the case of electrification, we anticipate an increased growth rate as we have government regulations kicking in many different areas. By 2040, we think, according to forecast, 60% of the global fleet will be for new cars, will be electric. This is all driving higher number of chips and content per car. Next.
One of the key areas in the powertrain side is trends in power semiconductors, moving from IGBTs to silicon carbide or gallium nitride. On the EV powertrain side, silicon carbide is becoming prevalent. Silicon carbide also used for high power charging, for other kinds of power conversion. Gallium nitride is in huge share already in things like commercial portable chargers, but also now being used on onboard chargers, maybe even in the drivetrain in the future. Both of these are growing rapidly, require different kinds of assembly, especially in die attachments, where we have moved from soft solder, which is still growing, but introducing diffusion bonding and sinter bonding as higher reliability of technologies for these power devices. Even in packaging and plating areas, we have customized solutions, high reliability solutions for the automotive market. Next.
This brings me to the summary of the market outlook section. As I mentioned, really, the move to chiplet architectures is driving a lot of our R&D and innovation at BESI, particularly in the area of hybrid bonding. We see applications and related developments in all the end markets, including HPC, mobile, and in automotive, as well as the automotive power aspects. With that, I think I'd like to wrap up my section. Thank you.
Thanks, Chris. Any questions in between? We have a question and answer at the end. Now it's your turn, please.
Thank you very much. Welcome, everybody here in the room and also online. My name is Peter Wiedner. I have been working for Besi for over 19 years, with one interruption in between. Actually, a little bit more than one year ago, I took over the responsibility for our wafer-level product portfolio. Let's get started. Our wafer-level product portfolio is actually what you see on this screen here. You have seen that the scheme of this new chiplet architecture in Chris' presentation just before, and we can offer a lot of production steps within that. The real wafer-level steps in that are, on one hand, for sure, the hybrid bonding step, number one, which is the enabler for this chiplet technology.
Secondly, the thermal compression, which is also coming along not only in old memory package types, but also in these new chiplet package types. Third of all, we also do provide machinery for the bridge attach. Having that said, it is in a way that the two enabling technologies, first and foremost, for sure, the hybrid and also the TC, these are the enablers, while actually the bridge attach is more a supporting step. For that reason, I do want to concentrate for this year's presentation, solely on the hybrid and the thermal compression.
Let's get before we get into the hybrid and the thermal compression, let's look at this chart, because we are very often getting questions about, well, isn't hybrid taking over the TC technology or other way around, and how is this related to each other? I think that's a very nice chart where you can see that.
You have not only, by the way, the hybrid, but also the whole development, you know, from the flip chip arena that Chris also mentioned in the beginning, over the development of the C4 flip chip, which is the deep purple one here, then the TCB technologies in a chip to substrate configuration, over to then the more modern style of the TCB with chip to wafer, which is then coming close to the new chiplet architecture as well, and finally going into hybrid. That's on one hand, really, a historic development of the flip chip technology overall. At the same time, you can also see on this chart when to use which technology, so to speak, because it really heavily depends on the number of interconnects that you need per square place.
That, for that, with that along, for sure, you're coming to the bump pitch, to the pitch between the interconnects. You see that only when you, when you get actually in the arena where you have interconnect pitches below 10 microns, so that's the green area, you only can use hybrid anymore. You don't have a chance to use TCB, because simply with melting the metal of these TCB studs in between, it's too narrow to the neighboring, and it will interconnect. You would more or less connect all the bumps that you have underneath, and so only a copper-to-copper interconnect without any other additional metal, enables you to go down to these low bump pitches. That's the reason why that's not one or the other. They are coexisting. You need both. For the really high-density interconnects, you need the hybrid.
As Chris already mentioned, I want to emphasize it once more, also in this new chiplet architecture, if you then have a hybrid stack and you want to connect it, let's say, to a carrier or to a weight, or a carrier substrate, where you typically don't have that high in-intensity of interconnects anymore, then you can switch over and use typically a TCB. Still, because these are high-end packages, it's still a lot of interconnects. You see one work in there, in the light blue area, fluxless, so in the left portion of that, between 10 and 20 microns. That's definitely an arena where TCB can play. However, it's a more advanced TCB compared to the, to the history, because the pitches are getting so small that you, on one hand, need a high accuracy.
That's clear, obviously, to meet the counterpart, but also that you cannot use flux, which is a supporting material during the bonding anymore, because you cannot clean it out afterwards, because the interconnects are too close to each other. For that reason, you then need actually to have a fluxless TCB. That's exactly this arena between 10 and 20 microns pitch. That's precisely what we are targeting with our next generation, between as you can see. Let's stop here. What's this? I get six different signatures here. Doesn't matter. Let's go straight into the hybrid, actually, into the introduction slide. Last year, in the same forum here, we were talking about our back then, newest hybrid machine with 200 nanometer, with a dual gantry.
Providing also a good speed for this very complex process between 1,500 and 2,000 UPH, that's still there. Since then, there is 1 very important thing came in addition to that. Since then, we are doing high volume production, and also the yield of this production, we could bring up to the expected level of 99.9% and above. That's really an add-on since last year. I'll get to more details in a few slides from now. For this year, our main focus points are really developing and providing the first machines to the market on the 100 nanometer. That's a big topic of our today's and this year's activities.
Also, because last year we have started actually with logic devices, we are heavily engaging this year specifically with the memory devices for high-bandwidth memory. More to that later on. We don't do that alone, actually. We are doing that together with AMAT in a joint development that we have started back in 2019 already, which is very crucial, because in the hybrid bonding, it's not only about the bonding machine, it's also a lot about the surface preparation of the material that is getting into the machine. Only if everything comes together, you get a good bond, and we get a good yield. For that reason, it is really crucial, and we have established that relationship, and it's a very good relationship. We are expanding on that every year.
Out of that, we already get a lot of benefits. On one hand, we now can offer, together with AMAT, not only a single stand-alone bonder, which we still do for sure, but in addition, we can also provide to the market a fully integrated system, which is also including all the preparation steps that are needed. To really make it easier for the customer to control all the individual crucial steps that are needed to get a good bond. That's number one. That's the integrated system that you see in the left part of the slide. The other part is we have set up together in Singapore, actually, our center of excellence, where we also have this type of equipment over there.
All both of our teams are working there, and where we can do, actually, on one hand, development work on the process together. For example, the memory high-bandwidth, the high-bandwidth memories is one thing, as I already mentioned for this year. On the other hand, actually, we are also, because we have the full capability there, we are doing also the customer demos in our center of excellence in Singapore. Actually, all relevant customers who want to use hybrid in the near future are definitely, they have all visited us, so we are in contact with all of the leading customers of this world via our center there. That's the reason why this partnership is really crucial and important to us.
Now, let's look at the development of hybrid since last year, since we have met last year in the same, for the same reason here. Last year, in fact, at that time, we were just about to install bonders in Taiwan for high volume production. On the upper side of the slide, you see the first products that have been introduced by them. Actually today, we can say, well, we have finished that. The first high volume production factory in Taiwan is set up all with our machines. Actually, the end customer of our customer has launched his products, and you see a few examples. That's not all, but a few examples on top, and Chris has already shown it, I don't need to repeat it twice.
They started with the Ryzen, but they also expanding step by step. Besides this setup, which is a standalone bonder setup, we have also managed to finish the development together with AMAT of this integrated system that I have mentioned before, and also the first shipments of this joint integrated system have been happening actually last year, towards the end of last year, to these first integrated lines, also to lead customers all over the world. That's our joint activities. As already mentioned earlier today, we are not stopping there because, especially via, you know, the lead customer, AMD, or the first adopter, AMD, we are already getting the signals that the next generations, because the design thinking is changing.
That was the example that Chris was making back in the flip chip. Also the designers now start to change, and they want to see the opportunities, the new things that they can do. Out of that, we immediately got the question: Well, I think the next year we will need even more accurate machine for 100 nanometers. As this was on our roadmap, also shown last year, we have executed on that. Actually, we will ship the first of the kind machine, so the first alpha machine of that, to our customer at the end of this month, so that he can already start with that machine to install it and also do, you know, the sampling for his end customer for this next generation type of chiplet products.
I would say a lot of things have been going on since we met last time, but there is also for sure a look what's ahead of us. What is ahead? Yeah, well, on the market introduction, if you look at that, especially now about this time, you know, they start with the MI300, that Chris has explained a little bit in more detail what it means. For their really high-end products. You know, with also the input of the next 100 nanometer that is needed, we also did get a confirmation, actually, about the next ramping scenarios of the two leading logic customers of this world. which are planned actually for 2024, all over the year, starting in the beginning of the year.
We got the confirmation for both of them, and that is really a good outlook for us. For us, actually, we have not only, you know, the cluster tool, but we will also ship additional tools this year and the 100 nanometer machine. What is also very important for us this year is that while last year, you know, there was this lead customer we were very focused on, this year, together with AMAT, we have broadened our scope, and we were really shipping a lot of machines, to, you know, these first type of machines, you know, in the R&D facilities of all these other leading customers, logic and memory.
That helps us to see that it's not only Taiwan and the U.S. starting with hybrid, but that really also Korea is following, and on the logic and on the memory as well. We are getting more insight into their programs, and we can now see much more that their programs are firm, and they are going for it and not only asking for information. That's really crucial to us. Actually, we have widened our customer portfolio this year very much, which is helping us then for the future years to also expand into their productions.
In order to do that, also one important topic, which is on the right bottom, is our fully-fledged, also redesigned 100 nanometer machine release, because this year we are keeping alpha machines in order to support leading customers, and from next year, end of Q onwards, we have the serial solution for that ready for all the customers in the program. That's for the n.xt year, and that was very much focused on logic. Let's have a look at hybrid and memory.
As I said, we are already in contact with all the memory customers as well. It is getting more and more visible that actually, for the high bandwidth memory generation 4, the HBM4, that will be the generation where these customers and these companies will introduce hybrid as a bonding step into their product portfolio. There are actually 2 reasons, and one you can see actually quite nicely on the right upper side. You know, all these AI calculations and all the high power computing does not only need high computing power. The higher the computing power, the more memory need you need in order to connect to these processors. It's not only the amount of memory, it's also the speed that you can access the memory.
What they can do with the HBM4 is that they, at the same space, which is limited and standardized, they can not only make a 8-stack memory or a 12-stack memory, because the connections really need no place. They can make up to a 16-stack memory, so that means you have more memory in the same space. Plus, as the connections are really pure copper connections and are electrically superior, they can get a higher bandwidth out of it. That's their driving forces, why they are going into that direction. They have a very clear reason why they need it, and for that reason, their programs, they all say that between 2025, 2026, depending on the customer, that's where they want to start launching them.
We have the R&D now. We have quoted actually from the old one statement here, which is from an actual report, that according to their forecast, 36% in 2028 will be memory of this HBM memory will be hybrid bonded. That's a little bit about the memory, the reason. What are we doing for that now? Next slide, please. Thank you. What are we doing for that? Well, number 1, we have started to intensify our process development activities in our center of excellence, which is on the top row of this slide, some of our test vehicles. We have started with our own test vehicles to develop, you know, the processes that are needed for this memory stacking. That's one activity.
We are not only doing that on our own. We also actually have engagement with the leading memory customers, and we have started to work together with them in our center of excellence. That's on the process side. On the machine side, actually, we started already last year to 1 memory customer who was kind of an early bird, and he already wanted to have a demo machine last year, which he has still today and in his R&D lab for his own research. Actually, in Q3 of this year, we will definitely ship cluster tools to 2 of the leading customers for their first adoption and their first, you know, testing of this memory.
That's what you see here on this chart. Overall, if I sum it up for the hybrid, it has started in logic last year. We were more in the R&D stage. In the meantime, we are in the HVM stage, and we have the outlook for next year as the next big ramp year for the logic. Now this year, we are starting together with the memory makers in the memory R&D stage, which will lead us to an outlook of volume ramping as a forecast from our customers roughly about 25, 26. With that, having that said, let's go to the last hybrid slide, which is, you know, the graph that we also showed last year, for sure, with updated numbers.
Where we can see as the logic is much more firm now, we could rise the low case numbers to higher levels because low case is, for us, so to speak, the case where we say, "Well, only logic will adapt to hybrid and none of the others." Now, with a much firm outlook, we could raise that number a bit. What we also did not have on the radar screen, well, we had it in our mind, but not on this slide last year, is also a third bullet, which is the application processors for mobile devices, which is a little bit farther out than the memory.
It's the classic high power computing first, then the memory, but also the logic devices, not the logic, the application processors for the handheld devices will get down the road because also there, the computing power is increasing and increasing, because not all the AI calculations can be done in a cloud. Something needs to be done with edge computing as well. For that reason, also, their performance needs to increase, and especially also in the combination that they are running on battery. Low power consumption is also a good argument there. That will drive also these gentlemen down the hybrid road, but they have a little bit more time for that. In order to do that, currently, we got some signs that it could be around 72, 2072, 2027.
Out of all these three categories, we can see that our outlook is still great for the hybrids. Actually, it's more on the rising chart. For sure, it's always coming in waves. You know, it's the first R&D, then it's the first HVM wave. It's not straight. It's always a little bit cyclical, but it's definitely coming. We are currently seeing that our outlook, what we think we can follow, is the mid case in this chart. For sure, there is always some room for improvement.
You have heard, I think, no, you might hear from Christoph that also, you know, in the CMOS arena, there is hybrid bonding, which, you know, which might get from a wafer to wafer to a die-to-wafer bonding, which is then our machine compared to other machines in the world. That will give them, would give additional, you know, uptick to a high case, if that is, if that is happening. Anyway, we are in a good way for the mid case. With that, let's start to move over to thermal compression. Once again, thermal compression, in our case, for our new machine, Richard said we are doing that since 12 years. That's correct. With our new machine, we are clearly targeting actually the chiplet architecture.
That means, I have explained that on the other slide, on the, on one of the beginning slides a bit. We need high accuracy, we need fluxless capability, and what I did not mention, we need much the capability to handle much bigger die sizes than in the old times, where it was typically memory dies with a fairly small or medium die size. Because if you have already a package of hybrid bonded devices, that's then bigger than in the past, the memory device was. We are now currently in the stage of releasing our first machine, and we have a clear roadmap to even higher accuracy and fluxless, and that will be the focus. That looks like this.
That exactly right now, we are starting, actually, you have seen actually a full production-worthy equipment for the people who are here on the floor. That's finished now. Last year, we were in the R&D. It's now finished, and we are starting actually with our lead customer in now in the second half of the year to adopt. We will start at the end of this year, also with memory customers. As I said, roadmap for fluxless and accuracy included and not long out, but already within the next 3 quarters. That is actually in order to complement in the chiplet architecture, our hybrid offering. With that, I can come to the conclusion. With our wafer level portfolio, we are really targeting, you know, all the steps that are needed in this new chiplet era.
For sure, number 1, with hybrids, where we have made our inroad, and we are clearly the market leader today. Secondly, with the TC, which is starting as we speak here. Third, which I did mention, but did not go into details, also with supporting technologies like the chip attach, like the bridge attach.
Thank you very much. I would hand over to Christoph.
Good afternoon, everybody. I want to welcome you here in Radfeld. Briefly introduce myself, I am Christoph Scheiring. I'm actually Austrian and coming from here. I have grown up here in that facility and have meanwhile moved to Switzerland and running the facility there. I have 20+ years of experience in semiconductor and in Besi. I have a technical background and, yeah, I am responsible for the mainstream die attach, and this is what I'm talking about now. Die attach, what does that mean? It's about 70%-80% of the Besi revenue, with a market share around 40%, as Richard has presented, steadily going up over the last years.
Even more important is our market share in the so-called advanced die attach segment. There we have 75%, and this is the segment where we really do the advanced packaging with high accuracy machines. The die attach portion is an interesting market because it's growing above average, with a 12.2% CAGR, as predicted by TechInsights, and the drivers behind are clearly the mobile. I talked about it a little bit already in these mobile cameras. That is a main driver for us, automotive, as well as computer. This is a snapshot of the die attach product portfolio of the product lines I'm responsible for. The bread and butter machine for die attach is clearly the one on the top left side, the Multi-Module Attach or 2200 evo, as we call it.
This is, I explained it a little bit already, a very versatile machine that is in very high volumes at this moment, always used when it comes to handling of non-standard processes, non-standard formats. On the right-hand side, we see the other high running platform, the die bonding platform. This is the high-end die bond, mainstream die bonder in the market. The next one in line is the flip chip, where we have an offering in the mass reflow flip chip market for both, for the strip-based formats as well as for the wafer formats. You have seen that downstairs, and I'll also talk about it in a minute.
Last, but certainly not least, the soft solder platform, which is a dedicated machine for the power electronic markets, where we do modules as well as discretes. Here I'm trying to explain a little bit our positioning in the end user market. It's clearly the mobile that is the strongest hold for us, followed by computing and automotive, which is developing. Especially the automotive is developing well over the last years. I think that's a phenomenon we have seen throughout the downturn, that this is about the one segment that remained relatively stable and even was growing. Looking to mobile there, the platform that is having the highest representation is the MMA platform, with its camera and sensor solutions, as well as the die bonding platform. This is mainly for 5G applications, but also IoT variables are drivers behind.
If we look then to the computing, here, this is the key positioning of our flip chip platform for CPUs, for GPUs, also for high-performance computing, and also with the evo platform. With the MMA, we are having a good traction in that market. On the one hand side, with these hard disk drive applications and also in the high-performance computing CoWoS segment. On the right-hand side, in the automotive, it's again, the MMA is having a very strong footprint there for sinter applications. Chris was talking about that new driver, sintering, and here also, the evo is positioned pretty well. Naturally, the soft solder is a platform that has specifically designed for that segment. Here we are also having a very strong position.
Now, showing a little bit of detail about our MMA platform. More than 7,000 machines are meanwhile in the field, and that's a huge number compared to, yeah, what's normally out there in the back-end industry. That is increasing rapidly with high shipment rates we are having. One of the most relevant installations at this moment is in the camera space, like you see here, LG Innotek, for instance, is one of the big customers. Foxconn is another one. Yeah, there are other installations, as you can see here in the middle chart, in other application fields, very prominent customers, as you can see. Some of the highlights...
I'd like to mention in the recent time, around MMA, are that we have been able to qualify this platform for a next generation periscope camera as a process of reference, POR. That will not only help us to drive business of this year, but also for the years to come. That was a big step. Other than that, we have increased the cleanliness of this machine a lot, and not only in pure cleanliness numbers, but we have also demonstrated the impact of that improved cleanliness on the yield, the output of one of our key camera suppliers. The numbers were impressive. The numbers were much better, and this will help us to lock in that platform further into the camera supply chain.
The last one I'd like to mention is, last time around, I was talking about a new development of a high force platform for the automotive sinter market. We have meanwhile launched that machine, and I can say that now, ready in time with that platform, since we started delivering that machine, we have seen good traction, and we have meanwhile, a significant installed base of these high force machines in the automotive market. Next in line here is the Epoxy platform. Also, that one has a significant installed base. More than 4,000 units are out there. Basically, all the major OSATs, as well as the IDMs that are in Epoxy and film-based die bonding, are using this machine because of the accuracy of the machine.
We are offering the highest accuracy class because of the speed of this machine and also because of the feature set of this machine. Our claim is really offering the highest end die bonding in the market. New developments or some of the highlights are, on the one hand side, in the automotive mesh segment. Here, we have done a lot of work around this control of the spacing volume and, further on, also on the bond line of these bonded chips. This feature is something that is specifically required in the automotive segment, and because of the superiority of the solution, we have been able to qualify this one for one of our key customers and lock that in for the years to come. Another development area is around inspection.
We introduced newly a six-side inspection system, and not only introduced it, we meanwhile have it also in a HVM environment, running in Taiwan with great success. The last one, also coupled with our developments, around the 2100 platform and all these height measurement systems we have introduced. This can be enabled now in the automotive CIS segment, and there we have also been able to qualify a new customer. Going to the flip chip. As said, flip chip basically is divided into the substrate-based flip chip and the wafer level flip chip. On the right-hand side, you see our QUANTUM and 2100 flip chip platform. That is the one for the substrates. There, it's basically all about speed and cost of ownership. That's a really mainstream and cost of ownership-driven market.
On the right-hand side... left-hand side, sorry, you see our QUANTUM platform. That is the one you have seen in our clean room. That's the one that is tailored to the advanced packaging needs. We have put a lot of effort into that machine in the recent months to drive the accuracy further up. We can now complete an accuracy of 1 micron, which is helping us in the high-density fan-out market and also bridge attach market, as we discussed earlier today. That always comes together with highest accuracies, we have worked on the cleanliness and the automation you have seen, and even a wafer loading system coupled to that QUANTUM, which, yeah, is always a must to have those units together.
On the high speed side, we did improvements, and that helped us to qualify new customers. We are continuing this customer penetration with the flip chip HS machine, and we keep on working on process reliability and process enhancements around our multi eject system. The last product line, the soft solder. I think you're well aware that our current soft solder platform, the 2009 platform, is the tool of reference in the high-end soft solder market. We have started to phase in now the successor platform, the 2100 SSR machine, and we progressed pretty well in the last year. Meanwhile, this 2100 SSI is has been qualified for 2 major auto IDMs we are having.
We have the next one planned in the second half of the year. With end of this year, I do believe that we have the 3 or 3 customers qualified for that platform. We keep on working on process improvements. One important step is diffusion soldering, which is required, which is driven mainly by Infineon, but also others are looking into that process. We are, at this moment, the only one providing a solution on our Esec 2009 platform. Next step is to port that over to the Esec 2100, that will also give us a good lock-in there going forward. Coming to my last slide already, talking a little bit about where are we going from here? What's next? What are driving forces? We basically see four elements. The first one is obvious.
We are now in a cyclical downturn. We do expect the market to grow, hopefully, from 2024 onwards. That will help us in the broad portfolio and the broad field of applications we are in. However, there are 3 additional e-elements where we do expect a growth above average. That is shown here on that slide. The most important one in the mobile space, as we are working, we have qualified a periscope camera line. We are working to further improve that. There is quite a strong growth expected in that market. That's one of the drivers improving camera features further. Another one is on the biometric identification. Those systems are pretty much isolated at this moment. Further integration is expected.
That integration requires assembly processes with higher accuracy, with, yeah, what our machines, what our evo especially can do best. Also this whole new product category, AR/VR devices, where we have seen a first glimpse yesterday. We do expect that 1 to come up, and I think you have seen how many cameras are in there. You have seen how complex that assembly processes will be behind. We do see that as an enormously good growth potential for us. In the automotive space, we all understand the EV adoption is going on. That fundamentally requires a powerful power modules based on silicon carbide, based on gallium nitride. We have at least 2 solutions, 2 die attach specifically for that market. One is the MMA, as we discussed.
The second one is the soft solder platform, especially around the diffusion solder. Last but not least, in the computing space, with the new architectures on the one hand side, and also with photonics being part of the processing unit, we have platforms that can support that, which is our fan-out solution, our bridge-attached solution, the XQ carrier platform. We have solutions also for the transceiver assembly with our Evo that is already being used there, and we do see further potential to grow our business there because of the strong growth that is expected in this market. Well, with this outlook and with these driving forces, that will hopefully help us to grow our business. I'm at the end of my presentation and ready to hand over to you.
Good afternoon, for you, my dear audience. My name is Jeroen Kleijburg . I'm responsible for packaging. Packaging is putting a plastic layer around a chip and then shaping it in an individual chip, microchip. I'm with the company of almost 25 years and 6 days, just like my anniversary. According to us, I now through my probation, you can say. Yeah, as mentioned, it was 1 point. As mentioned, we just had die attach. Die attach is packing of this, placing dies on the carrier and making an interconnection. Without modeling, simulation, which are form, you cannot use it in a computer or in a mobile phone. These product groups are essential for the backend.
These chips are, as Christoph and also as Peter mentioned, they are in the 5G and variables, they are in IoT and computing, and these are typical array type packages. They are on the substrate, so flip chip devices. For that, we have a substrate modeling machine, where we can do 2 sides substrate modeling. We mold first 1 side, we flip it around, we place again as dies, and we mold again. This is specifically used for system and packages. We do the simulation, so we cut the substrate in individual packages, and then you can place them on the STP board. For the automotive and also for the storage and the networking, those are typically single-sided or double-sided individual packages. They are placed on a lead frame or a copper lead frame.
For that one, we also have to place an individual molding package around it, and then with the trimmer form, we punch the products out of the lead frame, and we shape the leads. It can then be ready to place on the STP board. That is what packaging is doing. Although we only do 20% of Besi, there's a huge potential for this market and also for the growth. Let me go into this. First, the molding machine for substrate packages. We have introduced this machine, I think, in 2012, and since then, we have already installed more than 350 systems in the market. This machine is typically focusing on double-sided packages, but also odd-shaped packages. You can see on the top side, there is some five bumps, and these five bumps are the antenna packages.
This is a 5G millimeter wave device or a System in Package as well. That's one side with an odd shape on top, and as well, you see on the bottom side, also a molded package. We can do it two times, so we have a double opportunity here. Next to that, you also have the 5G millimeter wave, is where you have an exposed area. For molding, it's important to keep that area clear of molding compound, only pack the dies. Later on, these devices, or there will be a connector attached, or there will be a sensor attached. For this kind of market, we are currently having close to 90% of market, and these are the typical customers that are using these machines for these mobile market or IoT market, as well for wearables.
This machine is specifically for the SAP packages. They're all becoming smaller and smaller, the SAP, becoming closer to each other. For that, we need to make sure there is no air inside. We are making a program to have further reduction on the voiding, that we can place components closer to each other. That you have a smaller phone or a smaller IoT. Because it's already From 2012, we are moving now to a third generation. We are building up also a new architecture platform in there, that we can expand with additional features, additional interfaces, like, for instance, the AGV handling. We see that in the back end or in the molding, there's not so much experience in these areas anymore, customers are going into automatic handling.
With a robot moving into the machine, drop in the machine, and then it must be handled. Same as they do it in the front end. In order to support that, we need to go move to a new platform. We have for the power devices and also industrial. As Chris just mentioned, there is a whole change ongoing in the automotive for the EV, electrical vehicles, for battery and power, battery management, power management, as well also for body safety. This whole change is going for a silicon carbide or gallium nitride devices.
Used to be these kind of package, these large IGBT packages, power packages, by using of an epoxy or a glob top, then you have a plastic casing, but that is creating problems. They're moving into molded packages, as Chris also mentioned in his presentation. This machine is focusing on those big IGBT, these big power packages, but not only for the car manufacturing, but also for the industrial, like solar panels or wind turbines, and even air conditions. All those electrical devices are moving into this molded IGBT packs. There's a lot of heat in there. The heat needs to get out. They're making a copper layer around it, and that needs to be clear of molding compound, because molding compound is an insulator. For that, we are developing the top floor.
Next to it as well, the price is quite. The amount of money we can earn is quite low, the cost of ownership is very important. We have built in this machine as well, that we can put more frames on more capacity in there to reduce the cost of ownership of course. For the terminal form, although all these power packages are analog devices, and they don't need to be so sophisticated, the reliability is very important. If you have a mobile phone, you drop it, you can simply buy a new one, but in your car, you drop, then you have an accident. Traceability and reliability is key in this, and that's what we have developed in the last couple of years.
We sold almost 1,500 systems already in the field, we're seeing more and more integration in these packages we got, getting. Vision, all kind of inspections that are required, reject function, so looking at what is wrong, that one take out. Do individual marking, where you have QR code. You can scan this QR code, you can see a whole production line, when it was produced, how it was produced. Track and traceability is, for these packages, key. For that, indeed, we have, all kind of integration with the vision, with laser marking, with testing equipment, as well with automatic roller. On the mobile, again, with simulation. We have these, substrate parts, the individual packages, they need to be placed on the SiP board.
As you have mentioned, it is an SiP package, and you need to get it smaller and smaller. In order to get the smallest package as possible, we developed a different sawing technique, where we can get much closer to the edge of the dice. High-speed sorting. We are now able to do 45k units an hour, so we are outperforming at this moment, the competition on the certain capabilities. For the automotive, a step cut, that is for also reliability. Normally, you cut through the whole package. Now you only do a partial cut. You do the plating. My colleague, Bart, will explain a bit more. We make a full cut. We get an increased reliability on these power packages.
Last but not least, Chris already showed that picture where we want to go to the whole line of chiplets. One picture was there, where you also do the chiplet through the molding. For that, we have developed a machine as well, and with that one, we can do different than the competition, is exposed. They can do the die completely exposed. Currently, with the fan-out principles, if you do compression molding, you overmold completely, and then you do a back grinding coming down to the die. In our case, we can do immediately already an exposed area. The second step, what we are now also learning in the market, is that we saw that, I think, in the presentation from Chris as well, that there is an underfill.
With this machine, we have the capability of underfilling the die and also overmolding or make it exposed same time. For packaging, the growth drivers and the key development projects, same setup as what Christoph had just mentioned. For packaging, the 3 major markets at this moment are mobile, car, and industrial. You see the standard revenue, you see the organic growth, and this will be additional growth driver. For the mobile, the 5G millimeter wave, as well as the 3D imaging sensor. For that, the key technology are SiP packages, but as well, also in battery management unit. There is a new thing that's coming up in the market, is where the mobile phone and then the battery must be lasting much longer. That will be a SiP package to control the lifetime of the battery.
What kind of the application the battery is going to use. Also here, an SiP in combination with the battery management. Of course, the wearables, we see more and more people are having technology on their body. For automotive, there for packaging is the power switching, the battery charging, battery management, and then the safety management as well. Especially the top three, that is the driver for silicon carbide and gallium nitride. Right now, your car charging will take about 30 minutes, and you want to lower that to 15 minutes or even shorter, same as when you do petrol. For that, they need to go into this molded package with silicon carbide or third technique. Battery charge of battery management et cetera as well.
Switching from, when you drive, you hit the brake, you charge your battery again. Those kind of battery switching requires new power devices. Then industrial. As mentioned, the energy transition from fossil fuels. By 2040, no more fossil fuels. Everything will be, say, natural power, solar panels, wind power, and so on. All that transition requires new power packages. Factory automation 4.0, also onshoring of production. All these parts will help the molding on the term form and simulation on the packaging and back end to have a further growth. That's my story. Now to Bart.
Thank you. A couple of slides more. I think this is slide 66, 67, 68. I have 2 slides and that's the last. That's the Q&A. Plating is about 10 equipment platforms. I'm not doing you 10 slides, right? I'll try to limit myself to some high-level messages. First of all, let me introduce myself. I joined last year, early last year, but also I have a history in semiconductor, studied physics, a PhD in chemistry, but then I joined ASML International, front end, at 300 millimeter wafer processing. All those cleanroom suits, I've been there. I spent my days and occasional nights in the clean rooms of the Intel and the Infineons. Cleanliness, particles, yields, I've been there.
When I went into managerial roles, robustness, squeezing more out of the equipment, reducing particles, availability, training the people, working with the supply chain, all of that, I did it from them. Now I join that end. Now we can do that with Besi and Besi Plating. Let's see, switching. First of all, this slide is about, you know, the bread and butter, it's called plating. Mostly renowned for its thin solder plating, you've already thought lead frames. Lead frames, in the end, need a layer of tin for the soldering, there's many flavors of that. You know, Meco has been around for a little while.
In 1978, it started more with connector plating, the wrong words for little connectors, those need to be plated as well. We like to focus now on semiconductor plating, right? The semiconductor industry, we like. There's a lot of synergy with the other product groups, but also there is some adjacent markets. Then again, like it says here, plating is now doing 5% of the revenue. The adjacent market is 20% of 5% and 1% of the AC saw. Will not explain too much about that. There's some legacy, but there is also some upside still in the adjacent markets we're looking at.
Richard always says, "You know, it's about the margin." If we can't do high volume, if we can't have a leadership position, we can't have the right margin, well, then let's focus on semiconductor, right? Semiconductor is good at this moment. Semiconductor is also good for plating. We are from Besi Plating. We are leading supplier to the leading IDMs, the leading OSATs, many of these logos. I mean, I think there's a couple here, we could fill the entire slide with logos of companies that have Besi, and there is Meco, plating equipment. According to TechInsights, we do really well. They report 80%. Doubt that number a little, it's likely over 70%. Also, during COVID, I mean, everybody struggled with supply chain disruption. We had customer complaints.
It seems we did better than the rest, right? That's also because Besi Plating is part of a large organization. We have a fantastic operations partner in Malaysia, good buying power, and also a lot of people that help pulling these parts when it's needed. Sales and service organization, a lot of synergy there. Plating equipment. Next slide, I'll show you some more about the growth drivers. They're large, the $1 million-$2 million, right? Let's move to the next slide. The growth drivers. Plating. There's a portfolio of equipment there. There's even more equipment platforms. The key message, it's automotive quality. The requirements are getting more and more stringent. Like Jeroen said, if you drive a car, your ADAS fails or some critical driver assistance unit fails, you can be in trouble.
Automotive and power can be automotive power for your drivetrain, but also automotive has many more chips than just an EV drivetrain. The ADAS infotain ment, I mean, my car is full of electronics. 3,000 chips in an advanced high-end car is not unusual anymore. 3,000, yeah? Needs to work in wintertime, needs to work in the heat of the summer. It needs to work, well, preferably for 20 years, right? Power modules were mentioned, right? Large ones. The larger components, they're molded. Of course, the modeling equipment of viewing is fantastic, but still, when a mold opens, there's some flashes completed, that those need to be removed. These things needs to be attached. There's multiple opportunities for wet chemical applications, so there's more than just in solder plating. Yeah. Power modules is a growth area.
Also, the leading-edge investment is simply more and more chips, then there's these quality constraints, right? Quality means more plating applications. This mold material, it needs to stick, needs to stick to the die, it needs to stick to the substrate. What we also do is mold adhesion promotion. We have dozens of systems already at the leading manufacturers, we expect a lot of more people to dive into the topic of advanced mold adhesion. chemical preplate we just spoke about. flux cleaning, this wettable flank into also the step cut that Jeroen mentioned. I'll spare you the details, but it's a topic that is getting more and more attention. It's quality assurance, enabling optical inspection of the soldering of this square little QFN to the, to the substrate, all quality-related topics.
These topics in the top half, they're about more revenue. There's also some opportunities to create more value out of also the templating that we already do. Automated quality controls, spheres, die trace, traceability. , that Jeroen mentioned. In the future, people want to be able to know each chip, the full history. That means that every process step, all the equipment needs to be connected. There's going to be much more data gathered. That's an opportunity for a lot of the Besi Plating systems that are out there. The robotics, yes, this is high-throughput equipment. Plating system can process about 2,000 lead frames per hour. Anyone that is still running by operators, so it's manually operate people bringing these cassettes. Besides supply chain interruption, we also have labor disruptions, right? It's difficult to get people.
We can't find people. Young people out of school, they don't want to be a night shift operator anymore. I can't blame them. Why, why would you want? Besides the labor cost part, the cost of ownership is just getting the people, and if you have the people, how are you going to keep them, right? These are all good drivers for robotics. I have to say, yes, front end, the school with overhead systems, back end, not there yet. Will likely come. I'm not sure if it's next year, yeah, but it's an important opportunity to prepare for. Lastly, the environmental footprint. Plating equipment, you're all welcome to see it in the Netherlands, but it's large. It's bulky, a lot of electricity, a lot of chemicals involved.
All of us, our customers, but also Besi, we have a responsibility to reduce the ecological footprint. We see even a technical possibility to reduce the footprint and increase the quality. That's a good combination to make a more sophisticated system. It will suit its purpose. Of course, there's also an ability to increase the gross margin of those systems. All in all, it's more than just in plating. There's a portfolio of wet chemical applications, but also tin soldering itself, so the bread and butter. Increase the market share, stay ahead of the competition, and simply sell more systems. Yeah. Any questions? Questions for Richard. I'm sure, yes, honest.
Thank you. Thank you. That brings us to the Q&A. I see one hand raised. Yes, please. I was preparing.
Hi, I'm Madeleine Jenkins from UBS. I was just wondering, you mentioned about high-bandwidth memory, the 36% of high-bandwidth memory is going to be hybrid bonding by 2028. I was just wondering what you're thinking for the rest of the market. Will that remain TCB, or do you think there's upside to that number?
That's a good question. Peter?
Well, the answer to the question is we definitely. That's a forecast given, you know, for a certain year. Typically, how these things are developing, it's going, it's not the end. There is a upside beside that. On the other hand, there is also one thing clear, not the whole HBM will convert to a hybrid. There will be always a mix between hybrid and thermal compression technology in the HBM technology.
Hi, Simon Coles from Barclays. You gave the new hybrid bonding estimates, and you said you were seeing higher ASPs. Could you just give some more color on that there was no revenue numbers on the chart, or maybe just color on how you think generation one, generation two, and so on, will be in the mix going forward? Thank you.
Well, as a rule, and, I would say technology and accuracy, there's an exponential relationship. There was an article in Financial Times this weekend on ASML. I don't know if you read it. There it has two graphs where it clearly tells you there's an exponential relationship between increased accuracy and cost. So you can expect the same for hybrid bonding machines. We indicated that this first generation is between $1.5 million and $2.5 million per machine, depending upon the features. This generation 1 plus will move up a notch, between $2 million and $3 million is a rough estimate. But it all depends on many other factors, die sizes, metrology integrated into the machine. So there are features which also drive up the price.
We also mentioned, I think it was last year already, that once you move below 100 nanometers, it's a whole different story yet again. The cost of that machine could move significantly up because we need different cameras, we need different sensors, we need a different mainframe, which is much more stable. I think Peter already alluded to that. Gradually moving up with step functions, the more accurate, the higher the cost and the higher the margin opportunity. Does that answer your question? Next. I see many things. Nigel.
Thanks. Thank you for the question. I have one on the hybrid bonding roadmap. CPU 24, hybrid bonding 25, 26, if I remember correctly, and then the application processor for the smartphone more towards 27. I think at the last earnings call, you were a little bit more urgent about the opportunity in the mobile space, partly because of the, well, smartphones, also tablets. Maybe that was the piece missing in that presentation. Do you still think that tablets or other mobile devices could be adopting hybrid bonding a little bit earlier?
Well, a little bit earlier, the roadmap Peter explained is similar to what we have seen in the past year. So first focused on CPU, high-end computing, and making inroads into portable devices as of 25, maybe 26. We also mentioned there's ongoing development. The big ones are all testing, is this feasible to adopt a chiplet version? Because the big driver is to eliminate the number of components in a way that you integrate those devices with a chiplet architecture. Instead, that's in full test, and it's hard to tell when that will hit the market. It has huge advantages because you save the cost of interposers. You also need less current to drive the circuitry.
The circuitry is faster, so there are many drivers which should support adoption as soon as it is mainstream feasible. But those roadmaps will still change. Mostly in technology, it is always delays. People are very excited about new technologies. So far, you can say that in the past 12 months, and we've said a lot about that already, it's been a bit unusual how fast the adoption has happened in the past 12 months. And that may have gotten people a little bit more excited. But still, always imagine there's an awful lot of technology involved in the different production steps, which may delay at some point. History tells you it does. So to predict precisely what happens in 2027 or 2028 is not easy. But we're on the right track. It has enormous interest, broadly growing. So that's all good.
Next question. I saw two more hands.
Hello, thank you. I think, can you talk about the top-line synergies between the products? You mentioned a few times in the presentation how hybrid bonding also drives demand for TCB and also for some of the other products. Can you explain a bit how you expect that to work? Is it just going to increase your addressable market, or also, you're going to cross-sell the products to the same client in some of the same lines?
Well, that's a very good question for Chris. Chris, you take that.
I had a little bit of trouble hearing the question.
Say it again?
Yeah.
And so-
Well, I mean, I can hear somebody else, yeah.
The synergies, between the products. You mentioned a few times in the presentation, also to have more demand for TCB and some of the other products. How do you see that working out? Do you expect the clients to buy the products together? Does it simply increase your addressable market? How do you see that happening?
Yeah, we certainly see that opportunity. The kind of the drawing that I showed in my presentation, also Peter showed, demonstrates that hybrid bonding is part of a larger assembly. It's kind of the most advanced interconnect that's kind of used to make a chiplet, but really as part of the front-end fabrication process. In a sense, we're moving in hybrid bonding, from the back-end assembly into the front end. What's delivered out of that is perhaps a sub-assembly, a chiplet, might be comprised of whatever the number is. Maybe that has bumps on it, and then it gets delivered to another part of the manufacturing line that uses the TCB machine to mount it onto interposer, along with other things. There's an attach rate there, right?
That whole thing further has to be assembled onto a substrate, so there's another attach rate with something like an evo. Those configurations are complex, and every customer will have a different one. We're certainly, the reason why we have this strategy of supporting every layer in the interconnect is because we want Besi to be simply chosen for every layer in the interconnect. Generally, I would say hybrid bonding, because its job is to mount these high-density chiplets onto a base tile, we'll have a higher number of hybrid bonded chiplets, and we'll go down to a smaller number of those things that are assembled onto an interposer. Finally, the interposer will be a smaller number yet, that gets assembled onto the substrate. With memory is kind of a wild card in there.
Hi. You have this great new TCB machinery we saw today, and there is this roadmap as well with new machines coming. Is there any sense of what market size you're targeting there? You know, is it a few dozens of machines a year or much less than that, or any sense there? What would you have in mind for putting those new machines?
... we haven't laid out the details perhaps as much as we have for the hybrid as at this point in time. As a rough guess, we clearly see as, once again, coming back to what Chris just said, and it's not one or the other, that could easily be the same, market size as the hybrid in itself. We see the potential for that.
In terms of units or revenues, or how do you?
Well, it's, they have similar price tags, so it's.
Right.
It's the same actually.
Yeah.
Either you measure in units or revenues.
Maybe I can add some flavor to this. This whole trend towards chiplets, you have to think about classical big chip, cutting it in pieces. That's, at the end of the day, it will be a cost question. Is the smaller chips with the higher yield, including the extra cost you have for making the interconnects in between and the input/output that you need extra because you separate these chips now, will it be more effective than one chip? That trend is very positive. We had some discussions with ASML, for example, on that, because it impacts their models also. Overall, we see that in a very good range.
It's extremely difficult to estimate exactly how many hybrid bonds it will be, how many, because you have to take a chiplet, you have to say which parts should go there. So far, our calculations were fairly accurate, I would say, and Peter improved it somewhat. We said the lower side has gone a little bit better. We're trending between the mid case and the high case. The application process, and I think you mentioned, we said a little bit later than originally expected, but not much later, I must say. I must also say, Chris was on the ECTC conference last week, huh?
Yeah, last week.
One of the inputs we also got was that the cost effect of TCB versus hybrid, that it turns out that the hybrid is actually not more expensive than the TCB solution, which actually put a few of our customers saying: We would like, with this knowledge now, to bring a hybrid earlier on than we originally had expected. Because only now people start to get a full picture on the cost here. That was a very, I think, a very positive remark we got from two major customers. Yeah. That I would like to add as a little bit of flavor to this, yeah.
There was a.
Yeah
question on that side.
Yeah.
Yeah.
Hi, thank you. Bastian Heinen , SEB. Question on your HBM4 opportunity, which you mentioned. You have basically two machines targeting the HBM. You have your TCB new nucleus machine and your hybrid bonding machine. Are they both targeted for the HBM4, or is there a difference between them? The TCB might be for the HBM3 competing offering.
Easy to answer.
It will be a split. The current TCB we're developing is mainly for the logic. We're also developing a possibility for the memory stacks, of course, with that, but the main focus at this moment is logic. For the hybrid part, Peter already said, for the bigger stack chip, you could keep it also closed. Also, and that's the part that Chris highlighted, the developers of these devices are really starting to rethink what they can do with this hybrid bonding, because they were classically... Maybe it's a kind of hierarchy. You have hybrid bonding in, say, 5 micron and below pitches. They have 20 as TCB as you go down.
With this 5 micron and 1 micron technology combined, you start to see new possibilities also for the memory itself, because for this moment, the hybrid memory that just pitches it in the 50 micrometer range, you don't need on its own hybrid for it. As you start thinking that you can make different context structures in the device, we can build a complete new memory. That's something that's happening at this moment. The developers are really catching on with it. Does that help you a little bit?
Yeah. Yeah.
Yeah. It's really just.
The also B machine was also targeted for the memory at the end.
Yeah, they both do that. There's still life in the classical HBFs, so don't.
Yeah
take me wrong on that. First step will be logic, and then a few memory. HBM, I think our part of HBM will be probably more in the hybrid section. Yeah.
Thanks.
More questions?
Over there.
I carry the microphone also. I can do that, yeah.
Yeah, thanks. It's Robert from Deutsche Bank. I just have one question, Richard, for you, which is on the Q1 call, you said that the periscope order hadn't landed. You were expecting it by June. Given that we're in June now, I was just wondering if that had resolved itself. I have a follow-up. Thanks. I can come in as well.
Well, number one, this session today is about technology. It's not a market update. We will come with our numbers end of July, as everyone knows. So I can't give any specifics on orders received this quarter, but otherwise, we would have to do a full release. Your next question was?
Should I take the mic back?
No, but just tell me.
It was just about. Let me just for the benefit of people on the line. On the wafer to wafer hybrid bonding, obviously, that's been quite big in the NAND flash market. I was just wondering if with the HBM4 opportunity, if that was under consideration, or is it just not an option to do wafer to wafer versus side to wafer? Thank you.
Technically speaking, it would be an option. Economically speaking, no. I have pointed out, especially on the HBM4, that one aim is, to stack more dies. 16 is definitely the target number for this HBM4. You know, every wafer has its yield, and the more you stack on a wafer to wafer, the lower your yield gets, because on every every wafer, another spot is, you know, is not good. For that reason, that kicks you out, on the economic side, the higher the stacks gets.
Is NAND going to stay out of your circle of opportunity?
No, that's within the circle of opportunity. It just needs to develop.
Just following up on the TCB discussion earlier. In terms of the, you know, when the when you see kind of the growth of that, in the business, do you see it sort of running parallel, or would it sort of have a lag factor in terms of when you see the units of that being shipped out versus hybrid bonding?
Well, as I do not see a lag factor, you know, from a market perspective. Sure, if you know, company compare our numbers, head-to-head, so to speak, you might look at it like a lag factor, just because one machine was earlier in the market to launch and the other one later. That's the lag in between, so to speak. I would not see that from a market perspective, there is a lag in between the technologies. I start here, okay?
Thanks. Simon from Barclays. Just on the margin targets you gave, obviously, gross margin's gone up, which is great. It looks like you took away the net margin target, and now you're going to operating margin target, and it looks like the low end implies slightly lower than before. Is that just giving yourselves more flexibility around R&D? Just a bit more comment on that would be great. Thank you.
No, the answer is very simple. The world, for whatever reason, always looks at operating margins. Companies are not, well, you can correct me if I'm wrong. We only switched to this operating because many of the analysts simply asked for it. Is it at a lower rate? No, I don't think so. I think gross margins consistently in the 60% plus arena, then operating 45%. We reached net 37% and a bit in 2021, then 33% in 2022. This year, we have no forecast for the year, but we are still up and around the 30 sandwiches. Operating between 30%-45% is just a range which we determined, which I determined. Next question.
Hi, just a couple on PCB. Are you going to have new machines every 2 years that are more accurate, or is it just going to be this one machine?
Peter.
We have a second one.
Well, I answer that one first. Actually, we have done the groundwork, you know, for this machine generation, which is a brand new one now, solidly, so that it will not be a completely new machine. We will just add on to this platform and make it more accurate. There's a lot of potential in there to make it more accurate. On one of my slides, the next step was already for the end of the year, you know, this roadmap slide. For sure, we always develop better and better. Just look at the evo platform as another example, where we started with the predecessor at 20 micron, and then we fixed at 10, and now we are at 3. That always goes on.
I'm just trying to understand, when is it more economical to use TCB, and when is it more economical to use cyber bonding?
I did a lot of cost analysis, also together with AMAT in this case, because they have the full portfolio of all the processes and all the costs associated. We came to the conclusion that these cost levels are actually, if you compare all in all, quite comparable. Initially, a lot of industries said, "Oh, no, it's much more expensive." Now we start to see that actually these calculations were quite. For hybrid, you have to add this planarization, so that you have a very flat surface. You have to add a little bit of plasma, a little bit of clean, which are not so expensive. That's. The hybrid process by nature is much faster because you don't have a temperature cycle.
TCB is easy, but you have other things, like you have to do flux cleaning, and you have to have other things. Your process by nature is slow because you go up in temperature, you stay, you go down, so the throughput.
possibly small. There's also another technical aspect is, as long as you work with solder material, and I'm not sure if any one of you have ever soldered water pipes. You have to put some flux on there, and then you can solder it, but you're always solder bubbling off there. Now, on these contacts, you have the same effect that at a certain moment, these contacts become too close, the solder will start contacting the next contact. There's also a physical moment where you have to go to hybrid bonding, where CCD is completely not an option anymore. I hope that I gave you a little bit of that, but we did a lot of these concepts because it is a complex question. Any more questions? Good. Yeah.
Hi, it's Nicolas from Softbank. I just have a question on the manufacturing use of the acceleration hybrid bonding tools. My question is, at what level do you expect to start shipping the new tools? When do you expect that you raise about 99% manufacturing yields?
I had in my presentation, one slide where I said, we will be ready to ship manufacturing tools of the 100 nanometer, so that's the next generation, then in volume, in at the end of Q1, next year. That's the data I think you're looking for. With respect to the yield, we are showing actually a yield of more than 99.9% already today in the production. We do not have an expectation that this will be different for the next generation, as it is based on the existing one.
Yeah.
It's, you know, it's not an all-new development. I would agree, you know, then you would need to start doing a lot of things once again, but, you know, it's just an involvement of the existing machine, and for that reason, that should be there.
That's a very good sign that we're on the right track. Customers will never tell you a thing goes well because they will get increased price, but they only will tell you things that go wrong. That's how life is, correct? On the two things, we never had discussions at this moment on the hybrid bond is yield and also correlated with that partners. We are very well there. There's no issue. We have a lot of things always. It's technology, but those points are never raised, so they must be good. There was another question there, I think. Yeah.
Thanks. I have a couple of questions on bonding steps, and I think it was on Chris's slide, so I guess addressing to you. You mentioned 4 bonding steps for that Ryzen CCD chiplet, and then 13, more than 13 chiplets on the MI300. Is that 13 times, more than 13 times 4, or is it a slightly different configuration?
Yeah, yeah. Let me address that. The short answer is we don't know yet because we haven't seen the real part in a while. What we know is that there's actually more than 13 active chip, active die. In the example I showed on the Ryzen CCD, in addition to the active die, there are typically spacer chips. We call that structural silicon. We simply don't know how many of those are needed yet, or we can say. There are also the sometimes temporary bonding steps. It's well above, probably well above 20, but we don't know the actual number yet. We'll know more probably in a couple of months when we can look at the parts, and everybody can look at the parts, and then we can talk about it.
Then just on mobile, when that comes to the market, what would you expect for the number of bonding steps in a typical product there, if there's any kind of indication that you can give or you can share? Then more conceptually, once these markets start adopting, is this a kind of a one and done, so you get your product, it's got hybrid bonding in it, and it has X number of steps, and then it's all about adoption? Or is there an evolution that you think over the next 10 years where you keep on adding more and more steps for a given product?
I think in the mobile phone, the number of bonding steps will be less simply because the, you know, the available area, the die area, the point of doing it for AP is to maintain the same die area and not grow it, because the phone is very limited in form factor. I don't know what the number will be. It could be, like I said earlier, it could be 3, 4, 6, maybe, but we don't know until we really understand how the designers choose to architect using hybrid bonding, because like I said, it's very new to them. But that's in the range that I would expect, something like that. Of course, there could be some, these, spacer die, there could be these, temporary steps as well.
As far as the future, if you know, start with where we are today with the high-performance computing, I mean, we're already at a very high number with this Genoa-X device from AMD. That will, I think, continue to grow because we're not, again, we're not shrinking the transistor size. You know, we're still in vertical and things like that now on the front end. You're asking about a 10-year time frame. I think the area of silicon we have to settle through, and we're not gonna want to increase the chiplet size too much because the whole point is to maintain a reasonable yield on the path. Therefore, we try to leave more chiplets. Yeah.
Okay, thanks.
Thank you. I just had a, like, quick follow-up on the hybrid bonding yield. You said, you've got very impressive numbers, like 99.5%. I assume that's that you're telling these customers. I was just wondering, how easy is that to replicate at your other customers when you're out there?
Peter.
Well, the numbers that we are talking about are the numbers, you know, of our machine, you know, in the whole process chain that a customer then needs to do. As at an other customer, our machine is the same, the replication should not be an issue. Typically, when you see the yield is different than our customer side, you need to look for sure if the bonder is okay, that's clear, also then specifically if the preprocessing is okay because that has a big influence. Yes, there might be always, once in a while, a challenge with another customer is coming up, because he might have a different machine configuration before. It's not at our machine.
We typically, because once again, that's the importance of the process development as we are doing that with AMAT together, we also can then analyze and say, "Well, change this, change that," and then we get to the, to the group.
We definitely learned that this hybrid bonding process, it has to be far more accurate than with, let's say, a 5 or 10 micron process. Every single step you were preparing the die is more important, and customers have to learn that. Also, initially, they came with even top customers came here with material that was simply not bondable, not because of the bond was simply because the preparation was. We then came together, especially with AMAT, on that front. Of course, we've learned a lot, so we can now already judge much earlier, this material will not work. The learning is tremendous. Okay, I think we're done, I say here.
Either our presentation has not been clear or are they still more passionate about this topic? Anyone? Yes, please. Go.
Just talking about the number of bonding steps, it sounds like high-bandwidth memory might have more hybrid bonding steps with the pitch than mobile. Looking through to the kind of end of the 2020s, in terms of number of machines, should we expect higher volume from the high-band memory side or from the mobile side, do you think?
I guess the memory side, please.
The unit volume.
Yeah.
The unit volume is mobile or extremely high, so, it could be similar, actually.
Any other questions? We've come to the end of this session. May we thank you all very much for coming here to Radfeld, and sincerely hope that you have learned a lot, and in case you have any further questions, don't hesitate to reach out. Safe travels. Again, thanks.