KLAC (2023 - Q1)

Join Jefferies for next up. Very happy to have Oreste Donzella from KLA, who runs their – they call their EPC Group, Electronics, Packaging and Components, very hot area within the capital equipment market. He is going to run through a deck, and then after, if we have some time, I'll do the Q&A, but I'll take it over to you, Oreste. Oreste Donzella: Per

Perfect. Thank you very much. Happy to be back in London after a year. And this is really a dream come true. I mean people talk about semiconductor like there is nothing more important in the world to talk about it. And after 32 years in semiconductor industry, 25 at KLA, seven years working in the fab in TI and Micron for me is like a dream comes true and [a revenge], I guess, my children, they always question me why I'm so excited about semiconductor industry. And I'm still excited, as excited as I was day one in my career. So today, I'm going to talk about a new topic, advanced packaging just because after many, many years, semiconductor, you need to learn something new. But four years ago, we decided to start investing in advanced packaging, packaging overall. It was mostly driven by acquisitions. So we made several acquisitions in the space. We bought a company in Belgium called ICOS. We bought a company in Israel called Orbotech. We bought a company in the financial transaction of Orbotech in the UK that is called SPTS that is doing extremely, extremely well. In a market space that KLA didn't have much knowledge about this real process tools, PVD, etch and CVD, and I'll tell you a little bit the story of the evolution of our journey in packaging since 2020 when we started the journey actually in EPC Group at KLA. So this is, of course, forward-looking statements here. KLA overall, I just have a couple of slides on KLA organization structure. Almost 15,000 people, 2023 revenue was $9.7 billion. We got a little bit of decline in 2023, 8% year decline last year. In a year, it was supposed to be a catastrophe where we got like 8% only down. And this year, of course, we expect the market to stabilize and probably being a little bit up. So these are all the metrics that you care about. And one thing that we are very, very proud because it's the [measurement, our differentiation], our gross margin. So we can command a pretty good gross margin in the company because of our differentiated technology. My CEO, Rick Wallace, all the time, he tells me that the best proof of how differentiated our product portfolio and the solution portfolio is just pretty much to do command high gross margin. This is what we are proud of. So we are in many, many spaces in electronic ecosystem. Of course, in the wafers, logic, memory, specialty legacy, wafer-level packaging and component is particularly available packaging that is going to be the main topic of my speech. But also, we are in printed circuit board and IC substrates, and this is something that we acquired with the Orbotech transaction. And it will become more and more important because some of the advancement in advanced packaging will become a substrate in panel level, and we can leverage at that point our presence in the PCB market as well. Didn't change organization, it has been stable for a few years. So you can feel – you can see KLA is really a combination of three different business groups. One is our core business, the Semiconductor Process Control, Ahmad manages this organization. This is our inspection and metrology for wafer front-end fabs. Then we have a service organization under Brian. And my organization, the spaces across the edition spaces, not really core semiconductor process controlling wafer fab and mostly packaging, components, PCB subset and the specialty semiconductor process with the acquisition of SPTS. Okay. Now let me talk a little bit about market drivers that are going really to lead into packaging and why it's so important, so critical these days. So first of all, one thing that is absolutely a change in the semiconductor demand is diversification. I put this slide together actually four years ago and it becomes more and more true with all these big emphasis on AI and GPU and HBM, everything that is connected. So when you look at the stagnation years that I would say they were like after the smartphone introduction until probably when the first data center cloud and the hyperscalers started to invest in semiconductor. There was like a decade that the semiconductor industry was not really driving fast growth. It was a time that we said the semiconductor was growing as a GDP, probably semiconductor capital keep GDP plus, but no more than that. Now nobody, I believe in this room has any doubt that semiconductor will grow, a semiconductor capital equipment will grow much faster in GDP in the next five, 10 years because the huge infusion and pervasive participation of semiconductor in everyday life. And it's all about diversification. So and on top of that, of course, AI is the big driver. So I tried just to put together this chart to show how different the GPU package is between 2015 and 2024. So of course, the B100 chip, the GPU introduced a few months ago, and this is not enough because Jensen has already introduced the next generation of GPU last week. And you can see the difference. So was just memory, it was 12 gig now we have 192 gig done in HBM. HBM High Bandwidth Memory means memory that are stacking on top of each other. Nothing changed in the wafer fabrication of these DRAM versus a standard DDR DRAM that goes into the PC or smartphone, the secret sauce is in packaging. So the way our HBM differentiates itself versus the other DRAM chips. It's because in the package, you need to stack all this DRAM on top of each other because you need a lot of DRAM together with the GPU in a package, and you need to be on the same package because you want to have a proximity between the DRAM and the processor inside the same package. That's what HBM is all about. We will talk about HBM, HBM, but from wafer fab point of view, it didn't change much. What changes the way how you package the HBM in the wafer level package assembly line. And you see very, very stunning difference in terms of the area. So we move from 2 by 800 millimeters square. Of course, we are now surpassing 200 billion transistor in a single device memory type is different the size, the package type. Of course, this is another big, big change. Everybody is talking about. If I look at the buzz words, AI is the number one buzz word, everybody talk about AI. And then when you go one level down, what are the other buzz words? One is CoWoS, one is HBM. And by the way, HBM is integrating to CoWoS. CoWoS is real package. And this is what is limiting the capacity of GPU right now. It's not a wafer fab. It's really the ability to stack all this HBM, ship to TSMC. TSMC is packaging on CoWoS, CoWoS package. And then the entire packaging gets assembled generally through OSATs like Amkor, ASE, SPIL and so on. And of course, as I said, there are many things happening in the AI entire device. So starting, of course, the GPU needs a very advanced design from wafer fab point of view. We talked about memory that gets start in packaging and eventually, the CoWoS that put all this together. And when I look at the momentum of AI, and by the way, we are really at the tip of iceberg right now. Think about AI getting all this press, all this CapEx spend, and we are talking about cloud AI. We are talking about really AI in the hyperscale in the cloud. But AI will be implemented the edge or hybrid devices in the future, you will get a ton of demand for AI devices and eventually requiring a lot of semiconductors and a lot of advanced packaging. So this is what we are betting right now in KLA. So semiconductor is going to grow. We are in the right industry. And we have some other segment of these industry that were very, very small, like packaging that is growing even faster than the entire semiconductor market. Okay. Let me go through a little bit of technology slides here. So now you know the driver. I'm talking mostly about AI. I'm talking mostly about advanced packaging. But before advanced packaging because I've been in this industry for a long, long time, I've seen so many technology node scaling. My first job was 1 micron product engineer on a DRAM fab 32 years ago, 1 micron. Now we move from micron to nanometer. Now we are moving angstrom very, very soon. So we see really how pretty much the technology was able to invent itself and make huge progress just scaling the density of the devices and the number of transistors. And this was what, what more lows. So how we can pack double number of transistor in the same piece of silicon every 18 or 24 months. And lithography drove all this. But at a certain point, it always started to slow down, becoming more expensive. EUV scanner is a state-of-the-art tool and the high-NA EUV definitely is going to be the most incredible product that human being has ever produced. But is not enough anymore and it's too expensive. So people are thinking more and more ways how to deliver the performance and required power because it's not only performance. It's not only computing as Lisa Su said many, many times, the NBC or other important players in this industry the computing going to come together with power consumption because power consumption may become the biggest enemy of computing road map in the future. So you see on top of lithography scaling that is happening with the EUV or high-NA EUV, we see transistor architecture changing. So we have seen the transition between planar transistor to finFET. We will see it all around coming, it's coming already. We will see other transistor architecture later. The way how we distribute the power the so-called backside power delivery network right now is pretty much in the road map of the most advanced customers. The way it works is instead of delivering the power on top of your transistors, you deliver from the backside of the wafer, saving space, efficiency and eventually consuming less power at the device level. And then, of course, the topic of my speech is packaging, how you integrate multiple functions coming from different wafers on the same package in a cost-effective way. And this is what heterogenous integration or chiplet is all about. Okay. I'll skip this. Just I want to make sure that on top of 5G, of course, the semiconductor industry is progressing well on front end as well. Package is complementing whatever the front end wafer fabs is doing a lithography level, a transistor level and the power distribution level. So let's talk about packaging and all these vibe around heterogenous integration. So many years ago, there was a concept called SoC, system on a chip. So the big CPU producers at that point, they figure it out and say, I can have multiple functions on the same wafer, on the same chip and everything is getting connected each other. The problem with this approach is you were building a 3-nanometer wafer, for example, with some functions like control logic chips that they don't need to be a 3-nanometer. So why you are wasting 3-nanometer wafer space on functions that don't need to be a 3-nanometer that can be cheaper. So the idea about the heterogenous integration is two ideas really in one. One is I can have a mass cost-effective way to integrate multiple functions in packaging instead only wafer, number one. Number two is because of the computing and power requirement, I need all these functions to be very, very close to each other. And this is what generated the need for the so-called heterogenous integration. So putting more functions of a chip inside a pretty much a package. So building on a different wafers and then cut the wafers and then building this lego block that people call chiplet in packaging. And we have seen many of examples already, all the CPU and GPU right now are made with heterogenous integration packaging, the most advanced one. And there is – there are many, many ways to package multiple chips at the packaging level together. And from low bandwidth to ultra bandwidth depending on the applications, depending if you are stacking for example, DRAM on DRAM or SRAM on logic or logic on logic, there are multiple integration schemes of these heterogenous functions in a package in a wafer-level packaging architecture. So let me say – let me give you a couple of examples. One is what we call 2.5D, that is, of course, a CoWos package. This is the NVIDIA chip goes into CoWos packaging architecture and then I'll talk a little bit about 3D hybrid bonding later. And eventually, the way the packaging technology road map is evolving is to get these two together. So the next generation of packaging architecture will have a 2.5D configuration with 3D hybrid bonding on top. So even there are so many flavors of CoWos today that are going to that direction. And eventually, they may find application not only in the very, very expensive cloud high-performance computing ships, but also mobile in the future with different flavors in PC and so on. Whenever the AI is going to implement it on the edge. So that's the reason why the beauty of this packaging technology road map is, we are just at the start. When you talk about CoWos is just a start. When you talk 3D bonding on HBM memory is a start. There are infinite numbers of a potential combination of heterogenous integration architecture in packaging that can be done in the next five, 10 years. So this is the 2.5D here. Of course, you see the HBM on top of each other, connected with the bumps. And then you see the GPU landed on a silicon interposer and the silicon interposer is landing on a substrate. I'm not talking about substrate today much. But as I said, the substrate will become an important component of packaging and substrate is nothing else than very advanced PCB. And now you go back and say, why KLA invested in Orbotech? Well, at least we felt that PCB may evolve in something more technologically advanced. In packaging, we were right. It didn't come yet, but we believe the substrate will become more and more important next – in the near future and it's going to be much better growth engine for KLA as well. For now, we'll talk about what is happening on top, okay, the substrate? And then of course, you have the big ball connected to PCB. And when you look at 3D packaging, now you talk about two other names, one is Foveros Direct. This is what Intel is trying to build in their foundry market and foundry supply. So this is a big better Intel, of course, is playing to become a foundry, and they become a foundry if they are successful in delivery, this particular package is called Foveros Direct, it is a 3D integration with hybrid bonding. No bumps, but copper-to-copper direct bonding. And then SoIC, of course, is a packaging architecture that TSMC has already built for some premium chips for AMD. So this stuff is already in the market, a very limited, limited, limited volume and is going to grow exponentially if the market will demand more computing at power in the future for AI. So this is the case where you can see the 3D, you don't see bump anymore in the upper part of the graph, and you see chiplet connected to silicon and eventually in this case, copper-to-copper bonding in [die-on-die and die-on-wafer]. Many challenges, and this is where KLA plays a role. Let's start from inspection. I remember four years ago, I was talking to customers and they say, KLA wanted to enter the packaging business in inspection. And this guy said, I don't need you. I mean, I need one micron defect sensitivity. KLA is so advanced and is a front end now. And now we get pressure by the most important customers, they want 100-nanometer defect inspection sensitive packaging. They want 15-nanometer. Just to give an idea, how pretty much adding tools from front end and of course, customized for back-end because remember, the tools that you have in front end not necessary are good for packaging. You need to do a lot of work in customizing the handling, in customizing for the noise source, in customizing for warpage thickness. So everything in package is different. It's not standardized. In front end semiconductor, everything is standardized, in packaging it's customized. So we need to bring this capability from front end customizing for packaging, and we did it. And now people are looking for KLA because on this sophisticated type of packages, they need a front end like inspection capability or metrology capability. That's one big requirement coming up in packaging. The second requirement is that it's important for KLA in the sense that we not only have a process control, as you know, we bought SPTS here in the U.K. that I'm very, very proud to manage in my group. And SPTS is not an inspection metrology organization. It's a PVD, CVD and etch organization. And the good news about SPTS is we don't bring solution from 3-nanometer that are too expensive for packaging in that particular field, we customize and we build a solution for packaging. So our plasma etching tool or our plasma dicing tool is number one in the world because it was developed exactly for the need of packaging technology road map. And that's the reason why you see a lot of challenges here and KLA is working all these challenges inspection, metrology, but also process challenges like etching, or silicon, or plasma singulation of wafer and so on. And as I said, substrate is going to be the next engine for growth. Right now, this is what we did with Orbotech acquisition. We see a little bit of growth momentum there. I believe the real momentum will come in the next few years on the substrate and panel type of inspection and metrology. Okay. And I have a couple of slides to wrap this up again, as I said, a lot of challenges with the new architecture of packaging, heterogenous integration base. And we have a full suite of products, a process tool from SPTS on the wafer side, the plasma etch, dicing CVD and PVD. On the panel side, we have photolithography tool. Basically, in Germany, actually, this is another company we acquired under the Orbotech financial transaction. In Germany, Jena, of course, we have all the inspection and metrology tools coming from semiconductor front end customized for packaging or something that we developed in the last four years probably for packaging. And not only. Another thing that is important about KLA is when I want to – and I have been exposed to this company for a long, long time as an employee and before then as a customer. I believe, KLA, if I want to define a KLA, I would say KLA is a data company. So we produce an insane amount of data out of our inspection and metrology tools. Now the question is what do you do with the data? Many, many years ago, KLA invented a new way in semiconductor wafer fab to analyze – collect, analyze and manipulate the data in the wafer fabs. And 95% of the fabs in the world use our data analytics hub what we call Klarity. And the beauty of this is all the data in the fab regardless if – this is an open architecture software. I don't care if the data come from this software, doesn't care if they come from KLA tool or competitive tool, they get integrated and connecting our data analytic hub. And then eventually, you can produce feedback to correction loop that can improve the process based on the data you collect from the inspection metrology tool. It's a feedback, feedforward loop. So we are doing this for packaging right now. And I got to tell you there is a huge enthusiasm to have the same capability that KLA developed in the front end wafer fabs now driving data analytics and correction loop in packaging. It's a massive opportunity because, as I said, packaging is not standardized. And there is no concept of really monetize and materialize and capitalize on data and we are building this infrastructure right now. So not only tools but also developing data analytical software and algorithms and so on. And that's my final slide, and I think speak by itself. This is what we are forecasting this year revenue packaging, huge leap. So you can see we are almost doubling the revenue in 2024. We just gave – actually four weeks ago, it was $400 million. And then in the last four weeks, we got a lot of orders. And then Bren, Kevin and I were talking about or maybe we should say pretty much keep updating because in reality, especially for silicon interposer and CoWos, heterogenous integration architecture with HBM memory there is incredible appetite right now for demand and of course, for demand of capital equipment to develop these kind of packages. And we expect pretty much to land in $450 million, $500 million of revenue this year as a packaging revenue for entire KLA. And this is my last line.

I mean amazing timing. It just clicked zero. So you're an expert.

Yes. I didn't check.

I appreciate your time. Thank you very much.

By the way, thank you very much, and I'm – again, I'm very, very glad to be here last year with a different topic. Actually, last year, I talked a little bit about silicone carbide in automotive. That was a huge wave last year. Now it's kind of flattening a little bit. Packaging is the new one. Probably next year, I'm going to talk about something else. Thank you.

Appreciate that.