No AI Without Power: Inside Infineon's AI Infrastructure Revolution — COMPUTEX Keynote — Infineon

[00:00:00] Ni hao. Ni hao. This is amazing to be back at Computex. Honestly, you can feel it. You can feel [00:00:12] the electricity. You can feel the love with the collaboration. And of course, this is a huge [00:00:19] window of opportunity for us and our partners as we move forward. I'm feeling the love. Are you [00:00:27] feeling the love? Good. That's what I like to hear. Okay, then. So what I'd like to do here [00:00:35] is play the video, please. Look, we're in an AI revolution. Whether that is autonomous vehicles, [00:00:44] whether that is humanoids, whether that is the smart grid infrastructure. One thing is for clear. [00:00:53] The more that we use AI, the more compute power that is required. Hence, then energy. And we know [00:01:04] that data centers will be built. The AI factories. And we within Infineon, we wish to partner up [00:01:14] with the world leading customers to make sure that we really do have a system understanding, not just [00:01:24] at a product level, but a system level to show that we've got the innovation and to make sure [00:01:30] ultimately we are bringing value and scale. Scale is absolutely essential. And then ultimately, [00:01:40] we wish to bring grid to core to physical AI. And what we'd like to do today is to be walking through [00:01:50] a little bit of this topic from grid to core, and then physical AI. I've only got here 25 minutes, [00:02:00] and it's very difficult with me, with a wonderful audience, a big stage, a massive TV to only show you [00:02:08] a glimpse of what we will be showing today at this amazing event with the new announcements and the new [00:02:17] launches, as well as the collaborations with our partners. So let me start with the grid. Two amazing [00:02:26] inventions that we're working with the ecosystem partners on. I'm sure you recognize as we move up to the higher [00:02:34] voltage AI data centers to ultimately, ultimately to the micro grids. Today, this is a solid state [00:02:42] transformer. I'm sure that you are aware of them 20 tons. They've been in the industry for years, and of course, [00:02:49] this is incredibly robust. Now, moving forward, we recognize by using semiconductors, we can take these [00:02:58] conventional solid state transformers and make them to solid state or the conventional transformers to [00:03:05] make them to the solid state transformers moving forward. And here you can see, they could be 30% [00:03:12] light lighter from a size perspective. And of course, the efficiency will increase and further increase [00:03:20] increase as we develop this technology. So solid state transformers, we believe over the next years, [00:03:27] will become mainstream. And we're gearing up now to make sure that we can further scale with those customers [00:03:36] customers of ours and the ecosystem partners. The other area that we'd like to highlight is the solid state [00:03:44] circuit breaker. And again, these solid state circuit breakers are something that is now being deployed, [00:03:52] not in the future. Today, these are being deployed. And here, of course, the benefits are very, very evident. [00:04:02] We talk about the ultra fast fault isolation protection. And let me tell you a funny story. [00:04:10] We work with one of the lead solid state circuit breaker companies. And the CTO and the test engineers, [00:04:16] we were with when they first tried out our new technology, which I'll talk about in a moment, [00:04:22] they thought there was an error. They couldn't understand why the solid state circuit breaker was [00:04:27] not coming out with the traditional isolation schematic or the simulation. Then we realized that they [00:04:36] actually had to zoom in from milliseconds to get to the microseconds. And as they zoomed in, [00:04:43] this was game changing to this particular CTO that we're working with. And now the technology to do that [00:04:51] is a new technology here, silicon carbide cool sick called JFET. Very few companies on the planet Earth [00:05:00] can do this technology. And we have also today announced a further portfolio of JFETs to enable solid [00:05:09] state circuit breakers. So there's no, there's no one, you know, people are coming to Infineon and saying, [00:05:17] please ensure that we further scale this technology. So we have on voltage levels up to 1200 volts, [00:05:24] new package technology as well to take advantage of this amazing circuit breaker topologies that is today [00:05:34] shipping and further shipping at scale. So now let's move in. Let's go now to the evolution of the data [00:05:43] center racks. So if you remember, and last year we showed the same graph, but we wanted to walk through [00:05:52] a little bit more context with you today legacy. We know these data centers, the racks are below [00:06:00] whatever, 200, 300 kilowatts, the standard single phase. Then moving forward, we can of course see [00:06:11] the generations for gen two going at the free phase for the power sidecar. And we see these in abundance now at this show. [00:06:20] So, whether that is 500 kilowatts, 660 kilowatts, these are now reality. [00:06:27] And then on top of that, of course, what we call inside Infineon technologies, the North Star. [00:06:34] How do we enable the one megawatt racks? And it's a challenge. But those challenges bring in opportunities. [00:06:46] Because again, there's no AI without power. Now, I'll go through this in a little bit more detail. [00:06:54] However, we also wanted to make sure that this audience understands there is another miniature [00:07:00] revolution taking place on a Gentic AI. This area is, of course, utilizing the CPUs. [00:07:09] The ratios of the CPUs will change versus the traditional GPU architectures that we've seen. [00:07:17] And of course, the benefits of this are very, very clear that we can do certain compute capability [00:07:24] in the CPU agentic AI domain. The efficiencies, the latencies. That is very now clear to us as the [00:07:33] industry. And we at Infineon are very proud to be working with majority, if not all of the vendors [00:07:41] in this space. Because again, the technology matters. The performance matters. The reliability matters. [00:07:50] And we need to make sure that we have the capacity, the manufacturing, the sheer manufacturing capacity [00:07:58] to support this incredible growth that is in front of us. Now, let's just take a look. [00:08:06] Now, Infineon some time ago, even three, four, five, six years ago, we recognized that you can address [00:08:15] the AI data center infrastructure with a technology. But we anticipated with our system architects and [00:08:23] with our customers, the one technology is not good enough. We need three technologies from SICK, [00:08:34] from silicon, as well as gallium nitride. And this gives us the opportunity to really be able to walk [00:08:44] into the customers and give them the best solution with a combination or hybrid of all three. [00:08:52] Now, let me just explain here now how we see some of the build outs of the functional building blocks [00:08:59] of these AI data centers. Silicon is the fundamental backbone of our industry in many, many applications. [00:09:08] Now, Infineon, of course, one of the clear leaders in power management in the areas, of course, of silicon. [00:09:16] We're famous out there for ramping up the leading edge technologies. And this will stay. And in fact, [00:09:25] today, we also have announced a further portfolio of our Optimus 8 technology. The industry is using an [00:09:33] abundance of Optimus 5. The Optimus 8 is 44% better at the RDS on. So this technology out there is, of course, [00:09:46] of high interest to anybody trying to make sure that data centers are running efficiently as possible. [00:09:54] And of course, making sure that they've got the better performance and the manufacturing and robustness [00:10:02] scale. Now, of course, we haven't bought it today, but we don't stop there in our silicon roadmap. [00:10:10] Later this year, we will be bringing out a new technology in silicon. I promised our teams I would [00:10:18] not give too much away. But again, the performance of this, wow. Silicon still absolutely has a disruptive [00:10:29] behavior in these functional building blocks. And in the PSUs, of course, we just announced an 18 kilowatt, [00:10:37] as well as a 30 kilowatt PFC. Again, utilizing here, some of this great silicon technology. [00:10:46] The next area, silicon carbide. As you move up [00:10:51] from a frequency perspective, and you need then, of course, the capabilities of higher power, silicon [00:10:59] carbide is the chosen choice of a solution. And we see this now more and more in a number of [00:11:07] applications. And I'll talk about that a little bit later. But silicon carbide is something, again, [00:11:14] we, Infineon, have recognized. And it's not just making sure we throw in silicon carbide that goes, [00:11:21] for example, in other technologies like wind or solar. We're optimizing our silicon carbide for these [00:11:30] functional building blocks. And at the same time, we've got one of the biggest and one of the most [00:11:37] efficient, cost effective manufacturing locations in Malaysia to make sure that we're building this at [00:11:45] scale. Because we will need this in the industry. And then finally, the tipping point here of gallium [00:11:54] nitride. I'm smiling because it's been some journey of working in the industry to show the value and the [00:12:02] value of gallium nitride in simple language, the higher the switching frequency, then gallium nitride [00:12:10] comes to life. And we see this now for the first time coming now in these building blocks, that of [00:12:18] course, you can go from the 48 volt to six volts. You can, of course, go to the intermediate bus [00:12:26] conversion on mid-voltage, high-voltage, which I'll show you. But this is the sweet spot on gallium nitride. [00:12:32] And to the audience here, again, we're thinking ahead. [00:12:39] GAN has been in the industry on the eight-inch, of course, started on lower. [00:12:46] But now what we are moving forward to is further looking at scaling up our 12-inch gallium nitride [00:12:54] capability in Vellac in Europe. This is going to, again, give us an incredible opportunity [00:13:03] to bring GAN to the market with scale. And of course, from a performance point of view and being [00:13:10] cost competitive. Now let's go through a little bit more talking about the first, the industry first, [00:13:19] battery backup units. And again, there is here from the mid-voltage, as well as going into the high [00:13:28] voltage. I wanted to bring this up because it's important for you to understand that we've been [00:13:34] developing again with the ecosystem, with a lot of customers here in Taiwan, in many other regions [00:13:41] around the world. And we have got these solutions right the way through from the mid-voltage, [00:13:47] going there from the 1.5 kilowatts all the way up to the 12 kilowatt. And then for the higher voltage, [00:13:53] the 24 kilowatt. And we've got today also another announcement, a pure play 24 kilowatt [00:14:03] solution using silicon carbide. So again, this is something we've been designing, optimizing our [00:14:11] technology to be supporting. Now let's also move into another area, which is the intermediate bus [00:14:19] conversion. And here again, very simply put, we have got the mid-voltage as well as the high voltage. [00:14:26] And these are the down solutions. On the mid-voltage, we recognize there's topologies out there that we [00:14:34] have created and we support within the industry on the interleave solution. And then of course, [00:14:40] there is performance benefits. And on the high voltage, as you can see here at the 10 kilowatt, [00:14:46] 1.6 watts per millimeter squared. I mean, honestly, phenomenal. And again, that is a pure play GAN solution. [00:14:56] And Infineon, if I be bold enough to say in this room, we were invited in to the immediate bus [00:15:04] conversion because our customers said we wanted reliability and performance. And this was a key [00:15:11] asset, a key value proposition that we bought into the industry. And we are ramping up these [00:15:18] intermediate bus conversion. Then let's move into something pretty amazing. [00:15:25] Breaking the speed and density barrier on the V-core for the GPUs, for the ASICs, for the TPUs, [00:15:34] even for the CPUs moving forward. Now, this time last year, we presented this. This was our roadmap. [00:15:43] These were some of our modules, our vertical power delivery modules, down solution modules. [00:15:48] And if you remember, we actually went from one amp per millimeter squared to two amp per millimeter squared. [00:15:55] Honestly, as we stood off the stage here, many people coming up to us saying, wow, that's amazing. [00:16:01] That's incredible. We're in production, significant production with these technologies. [00:16:09] Now we wanted to bring something else to make sure you see the roadmap, the commitment that we have [00:16:16] within Infineon to the industry. [00:16:18] Let me just walk this through to you. [00:16:20] I'm going to talk about this part in a minute. [00:16:24] This part here is what we call here, the 2.5 amp per millimeter squared, the four-phase solution. [00:16:32] The slim vertical power module. Let me hold, because I'll show another slide on this very special technology. [00:16:43] Then, moving forward, we've got a dual-phase solution. [00:16:47] This, again, will go up to three amp per millimeter squared. [00:16:51] And then, moving forward, we will have a greater than three amp, and this one is a dual-phase high-density. [00:16:59] And, again, this will be four very specific applications that some of our customers have said, [00:17:05] can you bring that to the market? [00:17:08] And then, of course, that is substrate attached. [00:17:12] And then we move, of course, into the embedded substrate strategy. [00:17:18] Greater than four amp per millimeter. [00:17:21] This is quite phenomenal. [00:17:26] And we're working, again, with a lot of the lead, the who's who, win with the winners in the industry [00:17:33] to make sure our technology is ready and performing. [00:17:38] Let's have a look at this slim vertical power module. [00:17:42] Now, I was actually going to bring this with me. [00:17:46] But, unfortunately, when I held it up, even with this incredible screen behind, [00:17:51] you couldn't see it. [00:17:52] It was so thin. [00:17:55] And here, this makes a huge difference when you actually put this into the application. [00:18:02] From a pure size, from a z-height, from a weight perspective. [00:18:09] I'm not going to give all of our trade secrets away about how we're doing this, [00:18:13] and the filing of IP that we've got. [00:18:16] But what I will say is, it is definitely less than three millimeter. [00:18:25] That is incredible. [00:18:28] So, we're looking forward to, again, working with the ecosystem and making sure we support their journey [00:18:35] as the GPU goes up, of course, the powers of the ASICs and the TPUs, as well as, of course, [00:18:44] some of the CPUs now, we want to be part of that journey. [00:18:49] Now, here, what we'd like to show you is moving now from the AI, going now into the physical AI. [00:18:59] Now, let's just switch gears here, because Infineon, we've got a plethora of solutions that are going [00:19:06] into humanoids, as well as robotics. [00:19:10] Some of the latest announcements, whether it was at CES, or whether it's in these types of [00:19:16] conventions, there is a lot of Infineon technology and content in these humanoids. [00:19:23] We have got, of course, the computing, the control, we've got the connectivity, we've got the charger, [00:19:29] we've got the battery management systems, and, of course, we've got the sensing, the motor control, [00:19:34] as well as security. [00:19:37] And I would like to now just do a little bit more of a deeper dive into you, and let's talk about [00:19:42] the motor control. Of course, Infineon, we're pretty famous of understanding motors over the years. [00:19:49] A humanoid has, here, potentially 70 joints. And again, we're bringing in our latest technologies [00:19:58] to support the areas of motor control. And some of the areas, of course, need specialist technology. [00:20:08] So, let's now just go through a little bit of what we're doing. For example, here with the microcontrollers, [00:20:15] we've got our standard industrial consumer microcontrollers at scale with our PSOC family. [00:20:22] But at the same time, we have the world famous, the infamous Oryx microcontroller, number one, [00:20:29] undisputed, in there in the industry with FUSA capabilities. And on top of that, we have developed [00:20:38] a special GAN technology with driver to be supporting these motor drive applications. [00:20:46] And here, it's very interesting. One of the visits that I had with one of the lead companies that have [00:20:52] got very, very high ambitions for motor control. We finished the technical discussions. One of the [00:21:00] very senior procurement people came in, looked at me and asked me one question. Do we have enough [00:21:06] gallium nitride capacity to support their demand moving forward? Of course, bringing out the 12-inch [00:21:15] story was a no-brainer. And then in the sensors here, again, Infineon, we are a power management company. Yes, [00:21:23] we are. But we've also got incredible sensor technology. And this area, whether it's the position [00:21:31] sensors or the current sensors, these are finding their way now into such applications. [00:21:38] Zooming into the dexterous hand again, without going into too much detail, but we have a dedicated [00:21:45] controller capability, the Motex family, and again, the sensors. And again, there is value propositions [00:21:54] for us to bring this type of technology, because we help the battery life, help with the miniaturization, [00:22:01] and of course, ultimately helping with the performance. [00:22:07] So, and the other area here is for us recognizing connectivity. We're recognizing that the data loads [00:22:17] and the high volume data is significantly increasing in humanoids. As the humanoids become more and more [00:22:26] sophisticated, also again, thanks to some of the AI activities that are out there in the industry, [00:22:33] we recognize for the value propositions of latency, for example, for the high performance, that you need [00:22:42] Ethernet. And this here is something that we're bringing out. And we did a recent acquisition that is [00:22:49] going incredibly well from our post-merger integration. And we're finding, of course, significant [00:22:55] home for Ethernet. It's not just in vehicles, in cars, but also here into robotics, as well as humanoids, [00:23:04] for exactly what is behind me on this screen. And now, the other area, closing out the chapter for [00:23:12] today, is the area of security. And I'm sure all of you are recognizing the importance of security. [00:23:22] I don't know about you, but these humanoids and robotics, whether they're in factories, whether [00:23:27] they're in the streets, or of course, moving into our homes, we want to make sure that there is [00:23:33] trustworthy security. And of course, this is to protect the models, the, of course, integrity, [00:23:42] ultimately out there. And today, we'd also like to really talk about a special announcement with NVIDIA. [00:23:53] This has just been announced. We're working with NVIDIA to enable the quantum resilient [00:23:58] hardware route of trust for physical AI. And again, they're the value propositions here, [00:24:07] from the, from, of course, the keys, the models, the protections. And this is something that we are [00:24:13] recognizing as a company is an absolutely essential backbone for our partners. [00:24:20] So, as of course, NVIDIA announce and scale their Jensen Thor platform, the Optima technology here from [00:24:29] Infineon will find its home because the ecosystems trust Infineon's security. [00:24:38] So, this is actually a glimpse of what I wanted to show to you. And honestly, here, this is something [00:24:48] that we are recognizing and we are very, very grateful to all of you. And there is, of course, [00:24:56] challenges but opportunities of bringing AI. But we know exactly here, no AI without partnerships. [00:25:07] And here, we would like to thank all of you for your dedication, for your support. [00:25:13] We love partnering up with you. And we want to make sure that we further scale. And remember, [00:25:20] there is no AI without power. We're proud to be powering AI from grid to core to physical AI. [00:25:29] Xie xie.