Intel Foundry: Connecting Customers and Chiplets

How is the foundation of computing being reinvented to meet the demands of High-Performance Computing? 🖥️

‍

At the Six Five Summit: AI Unleashed, we're ecstatic to feature Kevin O’Buckley, SVP and GM, Foundry Services, Intel, as one of our Semiconductors track speakers. He joins host Patrick Moorhead for a deep dive into how Intel Foundry is revolutionizing semiconductor design and manufacturing through advanced chiplet-based designs and packaging.

‍

Key takeaways include:

‍

🔹Next-Gen Process Roadmap & Performance Optimization: Explore the cutting-edge advancements in Intel's process roadmap, including the introduction of Gate-All-Around (GAA) transistors and backside power delivery with Intel 18A, and how these technologies are optimizing performance and power management.

‍

🔹Strategic Alignment with AI & HPC Demands: Understand Intel Foundry's strategic roadmap, specifically how advanced packaging technologies like EMIB and Foveros are crucial for meeting the requirements of exploding AI computational demands in large die products and 3D stacking.

‍

🔹Success in Advanced Packaging & Chiplet Integration: Learn about recent breakthroughs and successes in advanced packaging and chiplet integration at Intel Foundry, showcasing the versatility and innovative capabilities of Intel's manufacturing processes.

‍

🔹Fostering an Open Ecosystem for Heterogeneous Integration: Discover the vital role of strategic partnerships and the creation of the Chiplet Alliance in fostering an ecosystem that supports the development and widespread adoption of heterogeneous integration for future AI and HPC.

‍

Learn more at Intel.

‍

Watch the full video at Six Five Media, and be sure to subscribe to our YouTube channel, so you never miss an episode.

‍

Or listen to the audio here:

‍

Patrick Moorhead:.Welcome back to The Six Five Summit. We are talking about AI and making it real and putting together the infrastructure and the building blocks to get that payoff for consumers and for businesses. We love AI here at the Six Five. And you know what we love even more? That's semiconductors. If you want to hear about semiconductors, you are in the right place, everybody. We're talking about chip designs, we're talking about architecture, we're talking from CPUs to GPUs to accelerators, memory networking and foundry services and pretty much everything in between. Joining me today is Kevin O’Buckley from Intel, a friend of the Six Five. We are going to chat about chiplet based designs, packaging, heterogeneous integration and I don't know, whatever Kevin wants to talk about.

‍

Kevin O'Buckley: I love it. I love it. Pat, thanks. Thrilled to be here.

‍

Patrick Moorhead: Yeah, it's great to see you. It's been a long time since we met in the desert of Las Vegas at your big event and you've subsequently had your big tent event to give updates and you know, a lot of people might have missed the event. I unfortunately wasn't able to be there in person, Kevin. I was there in spirit though. I just, just want to let you know.

‍

Kevin O'Buckley: I never doubted it for a second.

‍

Patrick Moorhead: I appreciate that. So hey, in four years you have put together a process roadmap that's back on track with Intel 18A. It's easy to just do an 18A process, but inside of that are a lot of innovations. Things like gates all around and BSP, otherwise known as backside power delivery. Can you talk about the significance of these two technologies and why does it matter to customers and maybe even talk about what they're saying about it?

‍

Kevin O'Buckley: Yeah, you bet. I think as I reflect on our career in, in the semiconductor industry, Pat, if you, if you go back to when you and I were young pups here, semiconductor long, it wasn't that long ago.

‍

Patrick Moorhead: Long, long. 

‍

Kevin O'Buckley: It wasn't that long ago. But if you reflect on early in our career, most of the advancement in semiconductors was about shrinking things. It was about making things smaller which reduced capacitance and you know, allowed the devices to move faster and allowed us to pack more in. And I think everyone has a high level understanding that at some point we ran out of gas on a lot of those, just making them smaller elements that we could count on. So innovation is now coming in different areas. And the two innovations that you described that we've introduced in our 18A technology, what we call ribbon fed or gate, all around technology. The first of those is about changing what a transistor physically looks like. And in my career, we've really only done that one other time. And you know, in 20, 25 plus years, when we went from a planar transistor, you know, sort of a flat transistor to the FinTech devices, and that was 10 to 15 years ago, depending upon, where you were in the industry. So, you know, going from a gate that sort of wraps around the top of the transistor to now a gate that goes all the way around top and bottom of the transistor is a huge innovation that allows us to make transistors faster without compromising performance. And then our backside power technology addresses the other element of silicon scaling, which is, okay, I got a really cool fast transistor. How do I get electrons in and out of that transistor? And backside power for the first time anywhere in the industry allows us to, you know, use metal levels on both sides of the transistor, both from above and below, which offers huge advantages in power performance. It's a big deal for us to be first in the industry to combine these two technology elements.

‍

Patrick Moorhead: Well, it is. And particularly given where you came from. You know, I've been tracking intel for 35 years and you know, a lot more ups than downs on, on what's happening. So it's been pretty astounding to me that, you know, it's funny, people say four years, Gosh, that's a long time. I'm like, four years, wow, that's not a long time in this industry. You know, I used to work for a company that had a foundry and then it didn't. And given how hard it was and how capital intensive it was. So yeah. And you know, the last four years since I've been even meeting with pretty much every executive at the company on the ELT, my first question was, hey, how's five nodes in four years going? How's this going? And people are like, well, you don't care about the design? I'm like, no, of course I care about the design. And Intel needs really good designs and really good technology to manufacture that in. And it does seem to be coming together for you, but I want to auger in into AI. I mean, what an opportunity for silicon. Listen, I always thought silicon was cool and sexy and important, but it took a little bit of time for the rest of the other markets. You know, the capital markets, the software markets. Right? Software guys are like, hey, it's all about software. Software is eating the world. And my response was, you're not going to run that software on air. Okay. And now really, semiconductors are leading the charge  in AI. So a pretty broad question, how does your roadmap align with the needs of AI and HPC workloads? And again, you've got to make a bet now for something that might have, might pay off in four years or four years ago. You would have made the bet then for it to pay off now.

Kevin O'Buckley: That's exactly right. I think I'll hit it from two angles, Pat. The first from the fundamentals of the semiconductor. The good news is from an Intel perspective, the good news is our, our legacy, our history has been as a semiconductor technology manufacturer focused on high performance computing. You know, we've been developing technologies that, you know, for what had been our only customer for many, many decades, Intel products to enable high performance compute, which is very, very similar and very well aligned with the needs of AI accelerators and the compute host nodes that connect to them. So whether it's at the advanced nodes, the two technology elements that you and I just chatted about, you know, our 18Atechnology with backside power and ribbon fat, those are exactly the drivers that the industry needs to enable the highest level of AI compute. The complement to that is, I think we all understand AI compute, especially in the training space, is not about what's happening in our pocket, it's about what's happening in mega scale data centers that are just extraordinarily interconnected. So for that you need networking capability, which is core to our offering. Everything from copper connectivity to optical connectivity. But just as relevantly, the packaging technologies that allow instead of a computer being one chip, maybe with a couple sockets on the board or four sockets in the most advanced server configurations to. No, no, no, dude, this is 100,000 boards all working coherently as a single compute element, which is literally what's happening in our data centers right now. So for us as Intel Foundry pad, it's the semiconductor technologies that we talked about and now the interconnect technologies, the advanced packaging technologies that allow that scale and those massive compute systems to operate coherently. And I'm thrilled that we've got both as core to our Intel Foundry offering.

‍

Patrick Moorhead: Yeah, the industry has come so far. I mean it used to be just a discussion about CPUs and then it, you know, CPUs, GPUs, and then once we ran into bottlenecks there, we started talking a lot about HBM and then we also talked a lot, you know, Packaging used to be a second class citizen. Now it's as important as the wafer itself. And I'm sure, you know, we could debate this, but you're not going to have advanced SOCs without advanced packaging. So it's all you and I together.

‍

Kevin O'Buckley: You and I don't need to debate this one, Pat. You, we're definitely on the same side of this. You know my perspective there, and it sounds like it's aligned with yours. Not that long ago, like five to 10 years ago, packaging was really about, hey, I've got these little tiny solder bumps on a chip and I need to connect them to big solder bumps on a board. And that's what packaging was. It was a glorified space transformer to let us connect solder connections. Now it's just the fundamental driver of 3D interconnection, compute density. And you know, it's just such a change. There just in the last couple of weeks was the annual ECTC conference. It's one of the major packaging conferences and it was filled with rock stars. I mean, you know that those packaging conferences as a leading indicator of where the industry's putting its attention, attention and its investments, was just filled with some of the most exciting advancements in the industry.

‍

Patrick Moorhead: Yeah. So one of the biggest architectural changes to chips is moving from these monolithic designs into chiplets. I mean, listen, if you can, if you can have good yield on that, you know, 800 millimeter wafer, go for it. Okay. But the reality is that that's not the case. And why put something on a leading edge node when it doesn't have to, when you can put it together in a chiplet configuration and put the right silicon on the right node, but you can't sacrifice performance and you can't sacrifice power. So we had a good conversation about packaging, but talk about some case studies, recent successes. Right. You know, is this, is this working? Are you actually doing this right now?

‍

Kevin O'Buckley: We are. There's a couple of awesome use cases that I think that are right down the middle of the examples that you used. One of the areas where scaling in semiconductors broke most fundamentally was in memory scaling, though the industry's gotten really effective and creative about continuing to scale what we call logic. You know, the fundamentals of making flip flops and storage gates and things like that. But the sram, the fundamental storage unit, is actually really not scaling well. So an example of use case that we're seeing a lot of, AMD was a great pioneer of driving this and some of the Solutions they're doing intel products is as well and a number of other companies is disaggregating their compute, the fundamental core compute blocks from a lot of the memory. And I'm not even talking about DRAM and HBM, I'm actually talking about SRAM. So there are some incredible advanced packaging applications today where the most expensive, aggressive logic logic technologies are used to get those fast transistors to compute. But the memory can be sandwiched right on top, you know, using three silicon vias so that you can do those SRAMs in a more cost effective technology without substantial compromise. And by the way, the advantage is those 800 millimeter monsters that you talked about, I don't need to compromise that with my sram. I can do the SRAM up here and I can use that full 800 square millimeters or less if I want to be more cost effective for the logic devices and the logic devices only. And that's allowing a lot more compute capacity to be deployed more cost effectively. The other big application there, Pat, that I think of is connectivity. You know, it used to be when you did what you and I would call an SOC, you would have, you know, the compute and then you'd have PCI, you'd have your DRAM interfaces all integrated monolithically into a die. And we can now push a lot of those expensive, you know, esoteric, I mean that is a compliment analog circuits onto a separate chip and you can again focus your compute investment in the center die and get all of your N minus 1 technology utilization of hey, perfectly good enough. 64 gig or 128 gig per second PCI PHI is for the, for the latest configuration or the latest DDR or HBM PHIs, you can put on a separate die. Huge, huge leverage in AI in particular.

‍

Patrick Moorhead: Yeah, I appreciate you correcting me. You can do it. You can have an 800 square millimeter die and you can stack stuff on top of it.

‍

Kevin O'Buckley: It ain't cheap, dude. Though you're right, it ain't cheap. That is an extremely expensive implementation.

‍

Patrick Moorhead: Yeah, we definitely see that with, with memory for sure. So hey, it does take a village to be successful as a foundry company. I mean from tool partners to IP partners, right. To really pull it all together so you can have a potential customer that can come in and have the confidence that their design is supported every step of the way and it's going to be. Nothing is easy, don't get me wrong, but as easy as it can get, so they have a decent time to market. Can you talk about on the chiplet side, the ecosystems, the collaboration, the partnerships you're pulling together to make this a reality.

‍

Kevin O'Buckley: Yeah, you got it. This Pat, maybe as I reflect on the transformation of Intel Foundry over the last few years, I tend to jump a lot to hey, look at these new technologies we've developed. We've got these new features. We're driving state of the art, maybe perhaps the biggest transformation to the company and the business has been in a necessary embrace of an ecosystem of partners so that we have what our customers need. Right. And just speaking openly, that's not historically what many would consider Intel's core strength. Right. Intel had a, you know, hey, this is how we do it and embraced companies like Synopsys and Cadence and Siemens on even just the EDA tools. So we're using what all of our customers expect to use. And in the chiplet space, embracing IP partners that can put in place standard chip to chip connectivity that you know will work whether we, Intel Foundry manufacture the die. But we may be talking to a chip manufactured at another foundry that has to work. Customers cannot have to worry about that interoperability. So having common partners for IT and standards is really critical. We've announced a chiplet alliance that has so many of the familiar names that you would expect on the IP side and the EDA tool side, but also design services partners, even packaging partners. One of the announcements we made at our Foundry event a month or so ago, Pat, for the first time us as Intel Foundry, we actually brought the Amcor team out on the stage and said, hey, if you want to use our most advanced packaging technology, you know what we call EMIB. Historically you've only been able to buy that from Intel, but you've told us customers that you want an ecosystem of partners to be able to manufacture these advanced packaging technologies. And we're enabling Amkor to do so because that's your voice and that's a big change for us.

‍

Patrick Moorhead: It is. And I have to tell you, people talk about the new Intel. I think your side of the business embodies the new Intel a lot. And you know, part of it is you have to because other design companies like to really fulfill your vision, you need to bring in new design companies other than Intel's design. And a relationship like any relationship is, give and a take, right? You're giving things back to the industry and the industry is giving you back things as well. And that's just a, this is a, the mature way to look at technology and listen, you know, you're going to fight like heck, you know, to win as much market share as you can. But finding those smart places to work with others and make it a, make it a good bi directional relationship, you can make money, they can make money and you can, you know, walk up into the sunset. Gosh, I'm about to cry here.

‍

Kevin O'Buckley: Let's go, let's go find a beach together, man.

‍

Patrick Moorhead: Exactly. Hey, this has been a great conversation. We have time for one more question here. So how do you see? It's funny things that sometimes come up on people's radar screen. I'm thinking that's not new and I kind of ignore it. But this whole idea of heterogeneous integration, right, which is essentially, I mean, a couple ways to look at this. Heterogeneous can be in, in design, in the way that you do this. It can be just a distributed architecture. There's a lot of different ways to look at this. But how will this market evolve here? You know, it's funny, on the, on the mobile side, they're all like monolithic designs, baby, right? Anything that seems to be below a certain threshold is, you know, it's not a heterogeneous design. And I'm just curious, like not necessarily even on mobile, but just overall, how does this market play out and what is your role in this?

‍

Kevin O'Buckley: I think heterogeneous means different things to different people. I love the question. The way I see it, Pat, is I really do think heterogeneous integration is the inevitable endpoint for us as an industry across all markets. Right now. Of course there's an intense focus on AI, but we've already seen heterogeneous integration in general purpose compute, even in PCs. Lower end PCs even are using heterogeneous compute to enable customers to develop products that can address a broad spectrum of applications. I have no doubt that we will see this in the mobile space. I have no doubt that we will see this in the automotive space in a variety of different markets. We can debate the when, but my personal perspective is it's going to happen. Economics will demand it. And the compute demand and the connectivity demand that we're putting on our products will also demand it. So I think it's inevitable. There's one other aspect of heterogeneity that I want to acknowledge. I think that's really important and it's one I'm actually, I truly think we're standing tall as an intel team in delivering and that is, look, my goal is to add value to my Customers. Right now, as big a company as intel sounds like I'm a startup in the foundry space. I am a startup. So I have to do things to ensure that I'm adding value to customers. And one of those that is heterogeneous to me is heterogeneous with a capital H. I'll put an Intel foundry die next to a piece of TSMC silicon or Samsung silicon. Heck, I'll take other foundry silicon and put them using my EMIB technology to connect them with and we're doing that today. So heterogeneous for us is about figuring out what the customers want, what the market needs and enabling and we are investing to be a premier heterogeneous supplier because I really think that's where the market's going.

‍

Patrick Moorhead: Yeah, I think it is too. I mean, to me it's a foregone conclusion on certain types of silicons and certain types of market. I'm waiting for that first smartphone chip to pop up and have it. But having them in PCs I think is a good first step here. So, Kevin, man, this is great. I really appreciate your time here. You know, you have a lot more work to do, but you've achieved a lot. I'm looking forward to seeing Panther Lake, I'm looking forward to seeing Clearwater Forest. I'm excited about those. But to be honest with you, I'm more excited about, you know, your foundry customers and when you start cranking out chips on 18A for them. Even though I know you're doing, you know, other processes before that. But I'm, I'm especially excited about 18A because 18A I believe, is really the, you know, I'll call it the grand opening for Intel Foundry. The doors have been open, but you know, with customers walking out with the goods. I'm going to put it on my calendar.

‍

Kevin O'Buckley: We see it the same way, Pat. We see it the same way. Thank you, man. I always get energy talking to you, man. I really do. I always appreciate it.

‍

Patrick Moorhead: I hope so, and the same. I hope you feel the same. I mean it. I love this stuff. So, yeah, thanks for coming on and hopefully we can have you on again in a few months.

‍

Kevin O'Buckley: Anytime. Always fun, Pat. Take care.

‍

Patrick Moorhead: That was a great conversation with Kevin. Hopefully you loved it. I just want to thank you for attending this semiconductor spotlight for the Summit in its sixth year. We're all talking about AI and semiconductors, from GPUs and AI accelerators to foundries, chiplet ecosystems. This track is looking at the entire full stack moving forward. You can view the entire agenda at sixfivemedia.com/summit. Stick around. We've got more insights and executives coming up.

‍

‍

Disclaimer: Six Five Summit is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

‍