875: How Semiconductors Are Made (And Fuel the AI Boom), with Kai Beckmann
01:10:29 Share

This is episode number 875 with Kai Beckmann, CEO of Electronics at Merck KGAA. Today's episode is brought to you by the Dell AI Factory with NVIDIA.

Welcome to the Super Data Science Podcast, the most listened to podcast in the data science industry. Each week, we bring you fun and inspiring people and ideas exploring the cutting edge of machine learning, AI, and related technologies that are transforming our world for the better. I'm your host, John Krohn. Thanks for joining me today. And now, let's make the complex simple.

Welcome back to the Super Data Science Podcast. Today, I've prepared an important episode for you on the hardware, specifically the semiconductors that underlie all computing and that are fueling the current AI boom. It's hard to imagine a better guest than Kai Beckmann for this essential topic. Kai.

Kai is member of the executive board of Merck KGAA Germany. You may not have heard of that company, but it's an important one. It's a 350-year-old firm. It's the world's oldest chemical and pharmaceutical company. It has more than 62,000 employees across 60 countries. Having worked at that gigantic firm for over 35 years,

Kai's been CEO of their electronics business for the past eight years. Under his leadership, Merck KGAA develops cutting-edge materials-based solutions and equipment for leading chip companies. 99% of electronic devices on our planet contain one of their products. That's crazy. He's a leading speaker within the semiconductor industry. He's an expert in material-based semiconductor solutions, AI, digitization, and change management.

Today's episode will be of interest to anyone looking to understand the hardware that all of computing and data science depend on. In today's episode, Kai details how materials from one company are found in virtually every electronic device on the planet, how AI is being used to develop materials that power more AI, and how AI is being used to develop materials that power more AI.

his vinyl record analogy for understanding computer chip manufacturing, the impact that scaled-up, stable quantum computing will have on society, and how a neuromorphic chip might someday run on the power of a low-wattage light bulb while matching human brain capabilities. All right, you ready for this scintillating episode? Let's go. ♪

Hi, welcome to the Super Data Science Podcast. I'm so excited to have you on. Where are you calling in from today? I'm sitting here in Darmstadt in Germany, so not far from the Frankfurt Main Airport. So that's the location where our headquarters is. Very nice. This is an exciting episode for me because

We spend so many of our episodes on this show talking about software, yet hardware is what drives so much of AI innovation. That's, you know, we have every once in a while a single kind of scientific idea, like a transformer comes along or a neural network comes along and then it's,

It's semiconductors. It's hardware that drives the capabilities from that point on. So really exciting to have you on the show. This is going to be a great episode. Thank you, John. Thank you for having me. So you are the CEO of the electronics business for Merck KGAA Darmstadt, Germany, which is the full legal name that we have to say because we should distinguish against another pharmaceutical company. So

Merck, KGAA, Darmstadt, Germany has different arms. One of those is the electronics business that you had. There's also chemicals, pharmaceuticals. And because of this pharmaceuticals overlap, the name overlap with a pharmaceutical company called Merck out of the USA, we have to do that distinguishing to make sure we say Merck, KGAA, Darmstadt, Germany theoretically every time. So most of this episode, we're just going to refer to your company. Okay.

But so you're the CEO of the electronics business for your company and Merck KGAA Darmstadt, Germany. I'm going to say one last time there because it is the world's oldest and largest still operating chemical and pharmaceutical company, which is really cool. It's more than 350 years old and along with 8,000 colleagues around the world, you push the boundaries of science and technology to develop materials and

and solutions for the world's leading tech companies. This is enormous. You said in a recent interview that the semiconductor industry has entered the age of materials and your products are found in almost every electronic device on the planet.

That is crazy. How is it that your products are so ubiquitous? John, for the team, it's amazing to see that our products are used for making all these electronic devices possible that you can buy around the world at

definitely is an element of pride. And you were referring to the 8,000 colleagues that helped us to make that possible. This is just the electronics team in our company. If you take our company as a whole, it's more than 62,000 people in the

In more than 60 countries globally, working on innovations in healthcare and life sciences, as well as in the electronics area. So that's quite a global and proud team making it happen that we have so many electronic devices supported by our materials. Very cool. That is enormous. It's amazing how there are these kinds of companies.

around the world that plays such an integral part in all the AI systems that we develop, all of the computational systems that we develop. And for some of our listeners, maybe many of our listeners, it will be their first time encountering your company by name. Something that I want to ask you about, given that it's made a huge splash at the time of recording, is Microsoft's Major Anno 1 quantum chip.

So it uses new kinds of materials to stabilize qubits, which are the fundamental where in classical computing we have zeros and ones bits. In quantum computing we have qubits that can take on a range of values from zero to one. And a big problem with these qubits historically is that they are highly unstable. And so you require rerunning experiments many times. But

With this major-ana-1 quantum chip from Microsoft, supposedly these qubits are now more stable. And so we can be scaling up now, apparently, to quantum chips with millions of qubits on them

How do you foresee this kind of innovation impacting the semiconductor industry? And what role does your company plan supporting such groundbreaking developments? If you take the semiconductor industry as the full industry now supporting these many innovations on the software side by very sophisticated hardware, and specifically in the last few years, AI being like the new kid on the block that everyone is talking about. There's so many different dimensions to how these innovations can happen.

And without now lecturing, but you can cluster that in supporting Moore's law. This is like more of Moore. That's maybe the right term. And then you have a second part, which is just additional dimension of Moore's law. This is what is called more than Moore. So this is like packaging innovations just to make sense.

still current architectures more powerful and create better scaling. And then you have an area which is completely outside these considerations in terms of future performance scaling. That is, architectures are based on what is called

post von Neumann architectures, architectures that go well beyond the currently used computer architectures based on von Neumann. And in this post von Neumann, you find specifically as maybe the next possible innovation, everything under the umbrella of neuromorphic computing, neuromorphic bringing logic and memory closer together, knowing that the channel between memory and logic is the bottleneck, is a bottleneck from

from a timing as well as from an energy consumption perspective. And of course, the biggest opportunity in this post for Neumann is everything around quantum computing. And quantum computing, there are very different principles around for quantum computing. And

Quite a number of solutions have to do with superconducting approaches to quantum computing. Unfortunately, superconducting requires very, very low temperature. You go very close to zero Kelvin. And this is definitely not an easily scalable way of doing it because you need

kind of cryogenic environment in order to make that possible. And there are different ways to try to make these qubits under those superconducting environments more stable and more powerful. And the one that you just highlighted, the approach from Microsoft is offering new opportunities here.

All of these technologies have to do, of course, with materials innovations as well as on the superconductor side, as well as on the Josephson junction side. They require materials innovations.

The market still is, of course, very, very limited. We're talking about a handful of devices being produced. This is definitely not a mass market for a materials company. However, our company is involved with our intermolecular facility in San Jose, in the Silicon Valley, with exploring different opportunities to improve the qubits with different partners. One we have published is the one with Psi Quantum not long ago. I was

presenting that at the Semicron VEST last year in San Francisco. So there are opportunities as well. This is from a hardware standpoint, if you allow me one more sentence, then from a software standpoint, of course it offers then different approaches, let's say to kind of data center like compute, replacing maybe some of the systems being used today for data center compute. However,

Given the data being generated out of a quantum computer, I believe there will be more compute around it than we had already in these data centers today or in the past. So probably future opportunities for even more scaling on the data center side of semiconductors. Very nice. That was a really interesting answer. The way that you talked about different kinds of emerging technologies,

not just quantum computing, but also neuromorphic computing. Something that I want to dig into a little bit more on neuromorphic computing that I think is really cool is this idea that we talk so much today about how much energy a system like GPT-4, that kind of scale of model, uses in terms of energy, in terms of water, to cool the systems. It's so interesting to me to think that a neuromorphic chip

done correctly, and it would probably take us a very long time, although maybe with AI we'll be able to get there faster than we think, you could theoretically have the power of a human brain running on the power of a light bulb. And we know that because that's how much energy the human brain uses. So it's theoretically possible that you could have all of the capabilities, this kind of

you know, the benchmark that a lot of people say for artificial general intelligence, AGI, is being able to replicate the kind of intelligence that a human has. So if that's true, with a neuromorphic chip designed to replicate the way, you know, much closer to the way our biological brain really works, you could get costs down in terms of energy costs down to something almost negligible.

Absolutely. So I think there's two areas where this could play out. Obviously, most people would think about training these models, these language models, where most of the energy currently is consumed. But I would probably even...

go deeper into inference, the application of these models in daily lives where it's about very cost efficient scalability in order to drive the right data into the right devices. That could be probably even a more beneficial area for driving neuromorphic architectures going forward. And so given the fact that data transfer in these devices is kind of a key area for optimization,

And I don't even think it's taking 10 years. I would say it's a five years plus question rather than a 10 years question. You mean specifically they're like the five? Neuromorphic. Application of neuromorphic architectures in general compute.

Perfect, perfect. But it's probably not going to be down to human brain kind of efficiency. No, not in that case. But given the progress of how AI has found its applications in the past three to four years, if you just apply the same innovation speed going forward, I think we can expect quite a lot from AI in the next couple of years.

For sure, no question. Directly related to what you just said about being able to have less data transfer, get rid of that bottleneck in AI capabilities, where today a lot of the cutting-edge AI that you want to have happen on, say, your phone, your laptop, needs to be sent to a data center where huge servers with lots of NVIDIA chips or something like that are able to process the requests and provide some generated output.

That bottleneck could be removed, of course, as you just said, by having the edge device, the laptop, the car, the phone, be able to do that processing itself. And so, yeah, you've mentioned that the AI boom, you've previously mentioned that the AI boom has so far been driven by data centers, but the next wave will involve AI chips running directly on edge devices.

What are the biggest material science challenges in enabling this transition and how is your company contributing to solving them? If you start with what is currently being used in data centers, it's all driven by further shrink of devices getting better.

more transistors on a die and driving performance based on lower voltages, less energy consumption. That's what we see currently happening with that enormous GPUs or GPU-like systems being used for training the large language models. Of course, with NVIDIA being spearheading that part.

And already here you see all the dimensions, different dimensions of improving performance already under one umbrella. One is faster devices, more integration on a device based on EUV lithography, based on a gate all around or nano sheet architectures on a device that is like scaling with Moore's law.

on a specific device. At the same time, you see that already being integrated memory and the GPU is being integrated in one integrated system called CoVOS based system. This is a chip on a wafer on a substrate that is kind of

bringing data and compute closer together, already another dimension of scaling performance. That's already happening as we speak with these, take the Hopper or take the Blackwell chip that is driving the performance of AI-based data center applications. That is what's happening right today.

as well as on-device with the latest and greatest technologies on smartphones, in laptop computing or desktop computing, or even in devices like automotives and in production systems.

This episode of Super Data Science is brought to you by the Dell AI Factory with NVIDIA, two trusted technology leaders united to deliver a comprehensive and secure AI solution customizable for any business. With a portfolio of products, solutions, and services tailored for AI workloads from desktop to data center to cloud, the Dell AI Factory with NVIDIA paves the way for AI to work seamlessly for you.

Integrated Dell and NVIDIA capabilities accelerate your AI-powered use cases, integrate your data and workflows, and enable you to design your own AI journey for repeatable, scalable outcomes. Visit www.dell.com slash superdatascience to learn more. That's dell.com slash superdatascience.

Very cool. Taking a step back to help me understand the kind of work that you do at your company, something that might help us would be to have some kind of general sense if this is possible. It might be tricky with so many different kinds of industries that you're serving, 8,000 people just in your semiconductor part and the electronics part of the business. I imagine this could be tricky, but would you be able to give us some kind of generalization or maybe a few examples of

a key kind of client? Maybe you don't need to mention them by name, but just a key kind of client that would come to you and make a request. And then how do you fulfill that request? How does, how does your business cycle kind of work? If you allow me, I would maybe start with, so how is the semiconductor, how is the chip being made? How is that happening? So it's in today's technologies of all the, the big chip companies, they use a similar approach. It's a,

a couple of hundred and more than 1,000, 1,400 steps from a blank wafer to a ready-made semiconductor device. Mainly we- - And what's a wafer?

So Wayfair, that silicon disc, it's a 300 millimeter sized disc. For the more seasoned folks listening, it's like a record in the past. So some people may still know a record, exact same size, 300 millimeters. It's a very thin silicon disc. And on that disc is...

you kind of put materials on top in order to build transistors, in order to build kind of the power lines, the data lines on that. And then you cut these disks into small dice

centimeters by centimeters size dies, and these are the chips. That's what we call at the end of the day chips once they are cut. And to make these transistors work, you need materials. You need ways to structure your materials. This is called lithography, like photolithography. And then you build with so-called thin film technologies. You build lines and transistors on these silicon transistors.

wafers on these disks. That's how it's being made. And this is 1,400 steps from a blank wafer to a functioning semiconductor device. That's what it takes. And it's always the same steps, lithography, patterning, deposition, cleaning, planarization. Always these kind of steps happen over and over and over again until that chip is made and it's working. Very

Very cool. And then so now that we have some sense of the process, tell us how the business, the materials business supporting that semiconductor industry works. So you need a number of specialized materials to create that. So, of course, the one which is probably known to still too many people is for the photolithography. You need what is called a photoresist. This is like a photoreceptor.

kind of a chemicals image and a photo in the past and that how it was done with a photosensitive layer that kind of creates a positive or negative image on your on your surface that is only the beginning because then you start building structures on these on these lines by edging away stuff that you don't need by

positioning stuff that you need by planarizing the surface in order to build the next layer of materials on top of another. These are specialty materials. They cover currently about 80% of the non-radioactive periodic table of elements. So if you had

chemistry in school, you know, the periodic table of elements. And if you take that full picture, a couple of them are either radioactive or they have a pretty short half-life. So they're not being used. But 80% of the rest is actually being used in making semiconductors.

And so most of the examples I'm currently using, they are based on what we call a logic chip. This is a CPU, GPU, everything that does any kind of logic switching. The other kind of big area for semiconductors is memory chips, everything used to store data on a permanent way, on a dynamic, kind of a short-term way. So these are different levels of data.

memory being used, but most of the examples I'm using here, whenever I talk about the transistors, it's more on the logic chips, the main area that drive compute right now in the world. Very interesting. And so now that I've asked about the business in general, and we understand a little bit more about semiconductors, you've been working at your company since 1989. Tell us about how you got into that and how you grew into this leadership. The

the thing that's interesting to me about this question, I don't usually ask guests on air how they grew into what they're doing today. But in your case, I think it's fascinating because it's this highly technical field and you're obviously very much on top of those technical aspects. You've been doing it for a long time. How does somebody become expert in semiconductors and then grow into a leadership position like you have? Yeah, it wasn't such a straight line as it probably hindsight. It could be, it,

could be made, it was more, let's say, a path through very different, very different assignments. You know, I left after I studied computer science with a very deep focus on semiconductors already. I left university and so now I was working on semiconductor design back then, improving semiconductor design in the late 80s. It wasn't such a sophisticated area as compared to today, but still it gave me, of course, deep insights into

what semiconductors are. And after being a research assistant at that place, then I was attracted by joining industry in a very different area, more in my, let's say, old home turf in software. And I worked in our corporate IT for quite a couple of years, more in the database space and doing consulting for process improvement. So these kinds of things.

That brought me then, and I skip now probably two decades, it brought me then into the business as well, running a country organization, selling our materials and our solutions.

with another step, then I was heading for a couple of years HR. So more as a board member and a kind of a more general petitioner rather than a specialist in HR, but it was times of massive transformation. And I think the owners wanted to have probably a person that has like practical knowledge of leadership rather than just a specialist. And so we drove the transformation of the company from within HR lens.

And then again, a couple of years later, I was kind of asked to head our electronic materials and solutions business in our company. That was, of course, a great opportunity because it brought me back to where it all started, to semiconductors. And this is why over here it is the leadership part, which is exciting, as well as the deep technology, where I believe I do have still quite a

quite some knowledge from the late 80s, which I refreshed, of course, I polished it a bit over time. But it's exciting with the team, not only to talk about the P&L and the growth plans and the strategy, but to talk about technology. I love to go to the labs and see that. I was just there a week ago and saw a new tool being used there for atomic layer deposition and talking to the R&D folks that

makes me really excited. This is what brings me to work every morning. Yeah, it's interesting how your background blended technical aspects as well as aspects like HR leadership and that blend over

over time, over decades, doing those from both sides, the technical aspects, leadership aspects, it allowed you to get to a point where, yeah, now you're the CEO of this highly technical business. So now that we've talked about your past a little bit, I'm going to talk about something that happened very recently, which is that your company purchased another company called UnitySC, which is a provider of metrology and inspection instrumentation for the semiconductor industry. Now that we know a bit about what the semiconductor industry is,

What is metrology? Why does this acquisition matter?

Yeah, let me just kind of start with the material side and then I like to share what's the logic behind creating kind of a broader footprint for our customers. So the materials, all these differentiated materials, highly complicated materials required to make these amazing structures possible that our customers, all the chip producers in the world need. That requires communication.

chemicals, the real chemistry knowledge in its physics and requires us to understand microelectronics because this is how these things are optimized to electronic properties at the end.

And it requires what we call a vertical integration, an integration of different capabilities. You need to understand how to make these materials. You need to understand how to test the electrical properties. You need to understand the activities of our customers' fabs. So what drives their yield, which is their ultimate target and what drives the performance of these tools. It needs to...

Another capability, which is how do we deliver these chemicals to a tool that at the end deposits it on a wafer? So understanding delivery systems, another important dimension. So this is how we have built the portfolio of our company across these very different domains.

Because our belief is, and our customers confirm this, that's not a stupid idea, is only if you optimize across these very different dimensions, then you are able to solve these very complex problems in the fastest possible speed. Because it's always about speed. How fast can you innovate for our customers? If you drive that in a more integrated way, you save on these costs.

cycles that typically take years in order to drive new technologies, you save massively on time and you get it to our customers much, much faster. That's the logic. Now coming to a metrology. A metrology is what is called the inspection of defects in the end product.

And in this case, the company we acquired, UnitySC, is an expert in visual inspection. This is inspection used on more on 2.5D or 3D, like on layered structures, where they can understand or test whether this whole system works as it's supposed to work. So understanding are these

so-called TSV, trans-silicon via. So these are holes in the silicon wafer. Are they built in the right way? Is the shape exactly as desired? And the inspection tool used is like a video camera in a way, if you want, but of course with a much, much better resolution that the end gives these very small structures or reproduces them in their data stream.

And this is where we learn how materials can be co-optimized in order to drive the performance of the end device. And this is why we are interested in integrating this capability into our materials focus areas. It sounds like that kind of metrology that being able to visually detect defects might itself involve an AI system, but what kind of role does high-precision metrology play in enabling the next wave of AI breakthroughs? Of

Of course, this metrology is specifically used for what I earlier called COVOS, so chip-on-wafer-on-substrates, these integrated structures like being used for Hopper or Blackwell-like chips or systems.

And in these areas specifically, this kind of metrology is being used. And of course, as you rightly said, the data stream generated out of a metrology system by itself, of course, allows you to optimize based on AI algorithms kind of methodology.

once the technology is being used to stack the devices, as well as the materials being used to drive these innovations. So on both sides, I think we call it ourselves, for our company, we call it AI for AI because we use AI to make these amazing materials happen in order to drive AI as the outcome of the chips that our customers produce. I hope that makes sense.

Ready to take your knowledge in machine learning and AI to the next level? Join Super Data Science and access an ever-growing library of over 40 courses and 200 hours of content.

From beginners to advanced professionals, Super Data Science has tailored programs just for you, including content on large language models, gradient boosting, and AI. With 17 unique career paths to help you navigate the courses, you will stay focused on your goal. Whether you aim to become a machine learning engineer, a generative AI expert, or simply add data skills to your career, Super Data Science has you covered. Start your 14-day free trial today at superdatascience.com.

This AI for AI, AI enabling AI, this is a really cool concept because it's a positive feedback loop. The better the chips get, the better the AI systems get. Those better AI systems can be used to make better chips. It's a really fortuitous cycle. You've mentioned a couple of times on the show this idea of Blackwell or Hopper chips and Blackwell.

Many of our listeners would probably know this, but just to make clear, that's a class. It's a very recent class of NVIDIA GPUs. Those are the kinds of chips, technology like that, that's driving the most cutting-edge AI systems that we have today and fueling the generative AI boom, now the agentic AI boom in 2025. I have a couple more technical questions for you before I get into some leadership stuff. So,

When we were doing research about you, we uncovered something called heterogeneous integration in AI chips. So what is heterogeneous integration and how does it impact performance and the packaging density of AI chips? This density thing being critical to building more and more powerful chips because obviously the more transistors you can get in a smaller space.

the more powerful a chip can be. Yeah, that's an important area, and I called it earlier in our conversation. So this is like what is more than Moore. So what dimension drives performance or allows to scale performance beyond just making smaller transistors on a chip? This is the additional dimension driven by heterogeneous integration. And maybe let me just...

quickly with a sentence, come back to the AI for AI. So we call that, we have like branded it the way that we call that materials intelligence. This is the use of artificial intelligence to drive the development of novel materials for applications in electronics. We call that materials intelligence. And this is how our team works as a global R&D team, not just

or in a traditional way, sequentially improving properties of materials by using AI to replace experiments in order to kind of avoid unnecessary experiments and going straight into where it really matters. Where can you really make a difference for the customer technology? How can you anticipate how a material works in a customer setup and how does it drive demand?

the solution of their problems and not just chemicals properties in the first glance. So this is how we drive the development of novel materials. We talk about millions of different options that need to be optimized in order to drive the performance of

So this is just to give you the idea on how that blends into AI for AI. Second is then driving the different aspects of how our customers improve the performance of their devices. And besides that,

shrinking the transistor, building more integrated systems and heterogeneous integration is the important area here. You know, it started traditionally with what is called kind of a front-end process, making a transistor, and back-end was then you wire it somehow that at the end the signal gets to the outside, which is then called in a broader scheme packaging.

Now there's something between these two extremes. It's called heterogeneous integration, when at the end the chip is not just one die, one single chip anymore, when you combine different chips to a system. And I refer to it in this specific example as CoVOS. These structures are being built in the examples I've used. I can use

different customer examples here as well. Just wanted to use one nomenclature, which is pretty common in the current conversation. This is when you glue dies on top of one another in order to build memory stacks, for example. Or you build a memory stack and you kind of almost glue it next to a GPU in order to shorten the transfer of data and to make it more efficient in getting the data to the GPU.

And that is called heterogeneous integration to make that possible. And it requires, of course, technologies well advanced from what was used in packaging historically. So much smaller structure sizes, much more complicated efforts to get your heat out of the system, as one example, or to optimize power consumption. The precision required then needs different technologies, more front-end-like technologies.

technologies, which makes it an area, of course, for materials innovations and for metrology innovation as per what our company is focused on. Interesting. And I'm glad how you tied that innovation, the heterogeneous integration, integration to some other kinds of concepts like metrology and the importance that that makes in getting transistors to be smaller, to be having a higher density of transistors on a chip.

In addition to transistors, another key aspect of effective computing is memory.

And so I have one last technical question here for you on that. And this is related to spin-on dielectric materials, SOD. How can the interplay of electronics and optical technologies and spin-on dielectric materials influence the electronics industry? What are spin-on dielectric materials? You probably have to start there.

You picked a nice domain. Typically, what we differentiate is you've got conducting materials, metals typically, even those metals evolve. When I was in college, a

Aluminum was probably the most commonly used metal on a chip. And then we used copper and we used tungsten. And meanwhile, we used molybdenum and many different conductive materials. And then you have insulating materials, so-called dielectrics, where you want to avoid that you've got kind of unwanted material.

flow of electrons on your device. So metals and dielectrics in a very simplified way. And these dielectrics are specifically being used if you want to stack functions on a chip, but then you need dielectrics in between that you don't get any short circuits.

or if you want to insulate transistors from one another, then dielectrics are being used. The most probably traditionally and commonly used dielectric is silicon oxide. That is kind of the easiest one to understand on a chip. But there's so many more dielectrics being used and very specific dielectrics, chemicals and molecules being used for insulation.

And then comes the way they are applied on a chip and spin on. That is, again, a pretty simple one. Going back to my record player example in the beginning. So if you remember your record player, and maybe some people still do, and if you would kind of

drop some water right in the middle of that record player. And then the water kind of travels over the full record. And this is a spin on technology. You put it in the middle and it spins and you cover the full surface with a thin layer.

film, which is exactly the way how a spin-on dielectric is brought to a wafer. It's the same principle, just of course with a bit more precision and accuracy. This is how a spin-on dielectric builds a quite nice, even film on top of a wafer. If our listeners at home now get their record player out and drop water in the center of it, is that going to cause damage to their vinyl records or is that going to be a safe experiment?

Now talking really about my history, there was a way to kind of play these records with a wet surface in order to avoid any scratching. And that was a pretty tricky way. This worked exactly the way I just explained.

Wow, that's wild. I had no idea. So something to look into there and maybe try safely at home if you can find a good YouTube tutorial. If you find a record player, I don't know how many people still find a record player at home. It's actually pretty common in my kind of set and people I hang out with, uh,

We kind of all have record players and buy records. I don't know if that's normal, but it's pretty common amongst people I know. I got a couple of records in my basement, though. Probably exceeding 2,000, 2,500 records I got in my basement, which I'm not using anymore. I just stack that and wait for a moment when I retire, and then I can resort to my record player and play some records. But that's for another day.

Very nice. Really quickly, to just kind of get a personal sense of you, what are your favorite records? I got lots of 70s and so all kinds of 70s music. And I grew up with Dire Straits and Supertramp and Pink Floyd. And probably I got all of those in my basement, amongst so many others.

I love those artists. We would, we'd be able to, I look forward to maybe someday you'll invite me over. I can check out your 2000 records in your basement and, uh, and we can listen to them. Uh, we'll have to do it in a room with lots of, uh, like shaggy carpets on the walls. Absolutely.

Nice. So going back to my technical question about spin-on dielectrics that led to this tangent on Final Records, my understanding is that spin-on dielectric materials are promising for high bandwidth memory.

in the coming years? It's one of many materials. There are many special materials used, especially in high bandwidth memory, since high bandwidth memory requires, let's say, the best possible DRAM performance. DRAM, dynamic RAM, is kind of the base component of a high bandwidth memory stack. And so you need a number of very specialized materials for high bandwidth.

performance in DRAM. The one which we have in mind as well is, you know, there's a so-called DRAM capacitor that needs to be optimized. And this is why our high-K materials, these are specific materials required to make these capacitors

are absolutely leading in order to drive the performance of a DRAM system. So there's many different things, you know, there's hundreds, if not thousands of different materials being used based on 80% of the periodic table of elements and all being optimized as precursors for very different production steps

So it's very difficult to single one out as being the most important one. But, you know, all of them are required to drive the performance of semiconductor devices. And to bring the idea home, to make it concrete,

The reason why memory, and actually you could speak to this better than me, but the reason why high-speed memory like DRAM is so critical to AI is because with these very large models, like large language models, you have lots of different GPUs communicating with each other and

by having high speed memory, you're able to move information between those different compute nodes more efficiently. - Absolutely, yeah. Another important dimension that we feel we are quite well suited for is once we go beyond electrons for transferring data, we go into photons. Whenever light is used for data transfer, that gives us two advantages, one is speed,

The other one is energy consumption. Just the photons don't create the heat that are created by electrons. And we have just reorganized last year the electronics sector by building an optronics unit based on our display experience and display history, where we know how to manipulate

and how to generate light in a proper way. And using light for data transmission in these systems is an enormous opportunity for further improvement of performance and reduction of energy consumption.

Mathematics forms the core of data science and machine learning. And now with my Mathematical Foundations of Machine Learning course, you can get a firm grasp of that math, particularly the essential linear algebra and calculus. You can get all the lectures for free on my YouTube channel, but if you don't mind paying a typically small amount for the Udemy version, you get everything from YouTube plus fully worked solutions to exercises and an official course completion certificate.

As countless guests on the show have emphasized, to be the best data scientist you can be, you've got to know the underlying math. So check out the links to my Mathematical Foundations and Machine Learning course in the show notes or at jonkrone.com slash udemy. That's jonkrone.com slash u-d-e-m-y.

Very interesting. Yeah, thank you for that. It's been amazing to get your technical insights on semiconductors, which has been the focus entirely of this episode so far. I would also like to now to change gears a little bit and to ask you some questions about the tremendous leadership experience that you have. So we already talked about earlier in this episode how you've been working at Merck KGAA Darmstadt, Germany since 1989. And in that three and a half decades, you've been working at Merck KGAA Darmstadt, Germany

you've grown through the organization. We already talked about that journey a little bit earlier on. The electronics part of the company that you lead that has 8,000 employees, that used to be called performance materials.

Do you want to tell us about that transition, what led to it, and what the challenges were? Maybe there's lessons that we can all learn from that kind of transition that you made at your company. That is a very interesting part of the company's history. In the late 60s and the 70s, the company started and very smart researchers found out about the technology

properties of liquid crystals in modulating light. So this is where all the technology advancements for flat panel displays came from, initially from our work in the late 60s and 70s with the first calculators or watches being based on liquid crystals. And it was such a phenomenal work. It took decades to generate properties

commercial products out of that, which then has led to all the flat panel displays. We had 80% market share for all the liquid crystal materials during its peak seasons in 2005, 2006 to 2010, when all the TVs were replaced by flat panel displays. We developed an enormous experience in

display materials, as well as understanding our electronics customer in this area. The unfortunate part of that is every of these enormous innovations has a lifecycle. They start slow, they ramp nicely, and then they plateau, and then it's getting really difficult. And the question is, how can you switch gears once you were so enormously successful in those days to find something else?

And that something else was how could we apply the same way of improving our customers' product performance in a different domain using as much of what we learned in the display arena. This is where we

entered the semiconductor world in 2014, basically. It's just about 10 years ago. And then we started venturing into adjacencies from our display experience into semiconductor. And in semiconductor, that was an opportunity which got bigger and bigger the more we dealt with it. And this is where we acquired then

Quite a number of companies from AZ Materials, where it all started from, to Sigma Aldrich, the high-tech business, a part of our Sigma Aldrich acquisition that was focused on semiconductors, to Visum Materials, to Intermolecular, to M Chemicals. And, you know, a number of companies acquired to form a market-leading portfolio around semiconductor technologies based on this deep chemistry gap

history, legacy that our company has. We drive chemistry for three and a half centuries and understanding quality challenges since the middle of the 19th century was the first industry's quality promise that was given to customers already in 1850. So there's such a deep legacy and we kind of conquered a new

a new market with semiconductors. And this was like perfect fit going forward since chemistry was needed, physics were needed, microelectronics needed, and all this together has formed a business which is now called the electronics sector. It's pretty wild to think how 350 years ago this company was founded and there's no way, I mean, you could never, they would never ever imagine the kinds of innovations that 350 years later

you would be doing as a company and the capabilities that that would be allowing. It's a while to think

When you think back 350 years ago, what were the leading technologies of the day? A horse and buggy. It's wild. It was even more severe in 1668, the year when our company was founded. Shortly after the Thirty Years' War in Europe ended, a devastating war,

And of course, health care was probably the most important need in order to improve people's lives. And this is where it all started. And then scaling from there into how can we support

other pharmacies, not just our own. How can we support other companies in the 19th century and make other companies successful? This is why we call ourselves in the electronics sector, the company behind the companies, Advancing Digital Living. So we pride ourselves that we help

our customers to drive the latest and greatest innovations, not only in the industry, one could say in the world right now. So this is how it all comes together. And to your point, innovating an existing company is far more complicated

than driving innovations out of a startup of a new company. Typically, you are in your own way whenever you drive innovation within an existing company. Typically, you believe what I did yesterday probably is successful tomorrow as well, which is probably the most fatal mistake you can do in business. Speaking of business and semiconductors, earlier this year at the World Economic Forum in Davos, which, by the way, before starting recording,

I mentioned to Kai how I had a ski injury. And so for our listeners, I currently can't, due to a skiing accident that I had, I can't currently flex my elbow or move my left shoulder. So my left arm is kind of this weird... Luckily, the fingers work, so I can still type and still grab things. And the neurologists say that because my fingers work, everything else should eventually start to work as well. Yeah.

but that actually, the skiing accident happened in Davos last week. And yeah, so anyway, Davos hosts the World Economic Forum every year. It's almost become, that town name has almost become synonymous with WEF. And in an interview at WEF, you mentioned that AI is fueling growth in leading-edge semiconductors, while other parts of the industry remain in a cyclical downturn. I thought that this was really interesting because

We hear so much about things like NVIDIA's share price or TSMC. We've heard about all the innovations that your company has today and things seem to be moving along really well. It seems like at the cutting edge, there is a huge amount of demand. Yet,

You say that other parts of the industry remain in a cyclical downturn. Could you tell us a bit more about that? The semiconductor industry serves very different markets. There is an industrial market. If you take all semiconductors required for automation in the industry, then there is the typical market.

consumer electronics related market TVs and what have you. And then you have mobile phones, desktop and laptop computers. And then you have the big area of data center as well. And if you take this very different markets, still a lot of volume is driven by consumers

All of us replacing our smartphones and replacing our desktop and laptop computers. This is where a lot of volume is generated from. And if you look into the placement rates of smartphones and computers on the consumer as well as on the industry side, it's still low. It's still low. So people try to hold on to what they bought during COVID-19.

Now for year number four, probably year number five already, and that is still dragging a bit, the recovery cycle for the semiconductor industry. The growth is driven predominantly from data center applications right now and data center applications related to AI in the first place. And this doesn't bail everything else out.

This is the situation right now. It's not compensating completely, but of course it drives an important high-end segment of the market. And this is why, of course, for us, it was driving our growth last year and gave us quite some upside from a very nice high-end application.

Very cool. How do you, as the CEO of a big company that has to navigate these different kinds of situations where there's okay, there might be a downturn in this sector, an upturn in this sector. How do you navigate that in the semiconductor industry? Semiconductor industry is cyclical by nature. This is ever since it started to exist because of its huge capacities being built that typically only know two different modes. One is running fast.

full power with full capacity and probably not running as the other option in between doesn't make economic sense. And this is why this industry, of course, tends to have cyclicality because of that kind of demand and supply situation I just explained.

Now, having worked in this industry for quite a couple of years, you get a bit more relaxed as it comes to cycles. You don't freak out whenever things go up and down. In times when we had growth rates for our semiconductor business north of 20% for semiconductor materials, that was not long ago. We tried to be still humble and try to keep our feet on the ground and not get too excited because we know

few years later, then you look into a shrinking market and you have to manage then cost and you have to manage idle capacity. So if you manage kind of both extremes well, then you can be a successful company in a cyclical environment. And our customers know how to deal with that. And I believe our peers, as well as we ourselves, we know how to deal with the cyclicality. And for the team,

I think it's a nice, humbling situation. So nobody freaks out if we have 20% growth rates, nor anybody panics if we have a year of decline. We try to manage our performance across these different cycles and manage our cost, manage our R&D spend well, so we don't get any negative disruption. Nice. Great explanation. You're clearly a pro at this kind of thing.

Something else related to your industry or the AI industry in general in Europe, a lot of people complain, you hear a lot of people complain about regulations in Europe and that potentially slowing things down. But in a panel discussion recently, you highlighted that while regulation can be a constraint, clear regulatory frameworks also provide investment certainty, particularly in things like chemical and pharmaceutical industries.

what kind of AI regulations would strike the right balance in your view between fostering innovation and ensuring compliance? Just take an example. If you provide data in whatever form or shape for being used by the public, then you want to be sure that you get, of course, the returns for making your data available. And if you provide data and everyone can exploit that data,

in large language models or whatever AI-based application without linking it back to the originator. That definitely is not helpful for generating that data. So everything related to protecting the contribution of those who make data available is certainly important in the current AI boom. We need that. And this stability helps.

Of course, on the other hand, there can be regulations that limit the growth of new technologies when things get too complicated. You know, if you have to have a lot of regulatory burden of applying for new technologies, if it takes you, for example, if you make a new material here in Darmstadt in Germany, make a new material and it requires us to get permit, it requires us a year to get the permit to make that new material.

That definitely doesn't help the industry to innovate. So speed of applications for new technologies, novel technologies is an essential area where Europe has to work on. This is where Europe certainly is not leading. So it's both sides. Providing a framework which creates a long-term understanding of how to invest

as well as not harming the industry by slowing down innovation cycles. That's the balance to keep. Nice. Well stated, well stated, like everything else that you've said in this interview. My last big question for you before I get to my usual conclusion questions for you, this is something that is, I

I think really exciting something that our researcher Serge Massis pulled out about some recent statements that you've made. In discussions with Josef Aschbacher at the Munich Security Conference, you emphasized the importance of collaboration between your company and the European Space Agency in advancing AI applications. I thought that was really cool. I don't talk about AI in space very often. So what kind of specific projects or areas of AI do you think hold promise in space?

There's quite a number of different perspectives on that. Maybe the first one, which is the least obvious one, is Darmstadt here in Germany is a bit like Houston. So we have the Space Operations Center right in Darmstadt. It's almost walking distance from here. It's a 20-minute walk and you get there. This is where all the space missions of ESA are being operated from, is here in Darmstadt in Germany.

That is why we have such a kind of proximity to the technology folks from ESA here in Darmstadt. It's one dimension. Both organizations, ESA as well as

as our company, is highly tech and science focused. We are so much focused on driving technology advancement. So there's a lot of similarities in the mindset of people. People click easily from both organizations. And third, and this is when the application comes into play, is, you know, in a...

Under low gravity environment, you can work on biological experiments quite a fair bit. So pharmaceutical research in space makes a lot of sense. Many companies investing in that. Materials research in space, an important area. So how do we drive

our R&D in space could be an important part, as well as new materials needed in order to make space missions more safe, more affordable and more efficient. These are things where materials are being used as well with certain R&D institutes of ESA.

So in general, collaborations are required across the different areas of the value chain. And ISA is a good example for good collaboration. And lastly,

data generated from space missions is an important source for optimization later on. It's in geodata, weather data, but as well as research data, materials data, and other areas where, of course, this huge amount of data being generated can be fed into platforms that we have built for the industry, such as Athenia being a semiconductor industry platform in order to drive

innovation for devices as well as for materials. Super cool. In case people were wondering, this might be obvious, but Kai said this abbreviation ESA, that's European Space Agency, ESA.

Fantastic. This has been an amazing episode, Kai. I've really enjoyed learning so much about semiconductors, the industry, as well as potential future innovations like AI in space that semiconductors will play a key role in. Before I let you go, do you have a book recommendation for us? It's not that you kind of gave me a hint. It's why I have two books in front of me. And I'm taking no tech books. I'm taking very recently...

I wouldn't say acquired because I got them as a present, two books, and I'm kind of holding these into the camera. The one is Pivot or Die from Gary Shapiro. So he's running the Consumer Electronics Show, and he gave it to me in January at the CES in Vegas. And so he's writing about driving changes as consumers.

as a leader. So that's an interesting read. I'm not completely done. I'm reading two now concurrently. The other one is Leadership in Management. This is maybe less known. This is a book written by

by a kind of a leader from NVIDIA, John Chen. And so he wrote that book. And so I was just with my team in the Silicon Valley, was it three weeks ago, two and a half weeks ago? And we were meeting at NVIDIA as well because we drive jointly on materials innovations projects with NVIDIA. And so, and John Chen, a

he wrote that book. So there are two books on leadership. They give you always new perspectives on what drives successes of companies, what makes leaders more effective in today's world. I think leadership is an important dimension of understanding how do you enable a team to do even greater stuff going forward than they did in the past. And then

an important part where it's worth reading at some time. It's an analog. So I'm reading analog while I'm reading a lot of stuff, of course, on the iPad or so, digital as well. But sometimes you need the old fashioned way. For people listening in our audio only format, Kai has the books with him. We can, he can prove that he is reading these analog. It's also my preferred way of reading. I find that

because my phone has so many other things in it, you know, just my mind starts to think, I wonder if I got any, I don't have notifications on my phone.

except a few people like my mom can phone me. That'll come through. But there's almost no way for me to get an active notification. But even then, if I'm trying to read on my phone, I think, I wonder if that email came through. I should just have a look. Or it would be easier than getting through this tough paragraph is seeing if today's puzzle on chess.com is easy. Yeah.

I can fully relate to that. A book doesn't have these notifications. That's a good thing. This is why a book is a book. Although I do do enough reading on my phone now. I read The Economist on my phone every day for probably 20-30 minutes.

and I get so used to being able to hover over a word and get the definition that that is something we've got to get that into books. I mean, you need to make semiconductors small enough and cheap enough that books can do that. We've got kind of

all kinds of reader e-ink devices, and then you can do all that stuff on a book. But then we are back to an iPad, so maybe this kind of closes the circle. I don't know. I'm not sure. So maybe sometimes it's good to have something which is completely undistracted and just using it, no battery, live. It just works.

For sure. 100%. All right. So Kai, for our listeners who want to follow you after today's episode, how should they do that? I'm very active on LinkedIn. This is the way I try to stay connected with sharing some of my thoughts on LinkedIn and maybe amongst a few CEOs who really operate their own LinkedIn account. So I must say I'm a LinkedIn user since 2008 already. And so I'm

I'm the only one who has the credentials. My credentials are my credentials. So what actually is done is physically done by me on the account. So just tell me whether you like what I'm writing there, and I'd like to stay connected. I cannot reply to every service offered on LinkedIn, I must admit. So there's so many...

even personal health and fitness programs being offered to me. I must admit, I do not reply to all of them and maybe to none, but that's a different story. But I like to write and if there's anything more constructive in any of the people's replies, I'm active even in replying to those.

I hope that makes sense. Yeah, it makes perfect sense to me. The key thing for me that I do is I actually have an auto response when people send me a DM that says, if you can write this as a comment, I'll respond to basically all of those. You know, if there's, you'll at least get a reaction. And if appropriate, I'll write something as a follow-on comment response. But yeah, there's just too many

too many service offerings in those DMs to possibly stay on top of it all. Um, but yeah, you know, when I, when I write something, post about an episode, just like you, I'm delighted for people to comment and I will definitely read it and I will reply. Um,

Kai, it's been so awesome having you on the show. Thank you for taking the time out of your no doubt extremely busy schedule to do this episode with us. And yeah, maybe catch up with you again in a couple of years and we can see how the semiconductor industry is coming along. John, thank you. It was a great conversation. I really appreciate it. Thank you.

What an honor to have such a renowned technical leader as Kai Beckmann on the show. In today's episode, he filled us in on the intricate process of semiconductor manufacturing, which involves 1400 steps from a blank silicon wafer to a functioning chip using materials that cover 80% of the stable non-reactive periodic table. It's so wild to me.

He also talked about the concept of materials intelligence, which is using AI to develop innovative materials that power the next generation of AI technologies. He talked about the development of heterogeneous integration, such as a chip on wafer on substrate that allows for more efficient data transfer between memory and compute processors.

And he also talked about how technologies like quantum computing, neuromorphic computing, and photonics could dramatically accelerate society's technological capabilities in the coming years. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Kai's social media profiles, as well as my own at superdatascience.com slash 875.

And if you'd like to engage with me in person as opposed to just through social media, I'd love to meet you in real life at the Open Data Science Conference ODSC East, which is running from May 13th to 15th in Boston. I'll be hosting the keynote sessions and along with my longtime friend and colleague, the extraordinary Ed Donner, I'll be delivering a four-hour hands-on training in Python to demonstrate how you can design, train, and deploy cutting-edge multi-agent AI systems for real-life applications.

Hopefully see you there.

Thanks, of course, to everyone on the Super Data Science podcast team, our podcast manager, Sonia Braevich, media editor, Mario Pombo, partnerships manager, Natalie Zheisky, our researcher, Serge Massis, writer, Dr. Zahra Karchei, and our founder, Kirill Aromenko. Thanks to all of them for producing another scintillating episode for us today. For enabling that super team to create this free podcast for you, we are deeply grateful to our sponsors. You can support the show by checking out our sponsors' links in the show notes.

And if you'd ever be interested in sponsoring a podcast episode yourself, you can find out how to do that at johnkrone.com slash podcast.

Otherwise, help us out by sharing this episode with people who'd like to hear it, review it, review the show on your favorite podcasting app or on YouTube, subscribe to the show, edit the show, you know, our YouTube videos into shorts if you feel like it. But most importantly, just keep on tuning in. I'm so grateful to have you listening and hope I can continue to make episodes you love for years and years to come. Until next time, keep on rocking it out there. And I'm looking forward to enjoying another round of the Super Data Science Podcast with you very soon.