BREAKING: MITRE Engenuity Publishes their Blueprint for a Domestic Semiconductor Industry. Read the Paper.
The Future of the Semiverse
Aired: October 19, 2022
Nitin is joined in episode two by Rick Gottscho, Executive Vice President and Chief Technology Officer at Lam Research. Gottscho details the legacy of innovation from Lam Research, noting specifically how their work both touches all semiconductors in everyday devices, as well as how their manufacturing is done primarily in America with an impact felt across the globe.
Gottscho gives an overview and deep dive into his vision of the “semiverse,” a concept that could pave the way for experimentation and coordination among a network of interconnected laboratories and research centers. By utilizing digital twinning and operations through the cloud, this concept could pave the way for interactions and innovations between entities spread throughout the nation providing their unique capabilities to unite for full stack innovation.
Welcome to Circuit Talk. My name is Nitin Shah. As in previous Circuit Talks, we've spoken to experts from the semiconductor industry, such as Todd Holmdahl from Microsoft. Today we have Rick Gottscho from Lam Research. So Rick, welcome, welcome. First, let's start with, please tell us about yourself.
Yeah. So I'm the Chief Technology Officer at Lam Research, which I have been since 2017. Prior to that, I ran all of the product groups at Lam Research, from 2009 through 2017. Before that, I had various jobs, ran the edge division at Lam Research for a number of years joined Lam Research in 1996, after having spent the first 15 years of my career at Bell Laboratories, I'm a PhD in physical chemistry. And my area of specialty was, and in some sense, still is molecular spectroscopy. But so I'm a scientist, but I'm quite proud of the fact that National Academy of Engineering has accepted me amongst the ranks, although I'm not really an engineer.
That's great. Thank you for that background. And, yeah, it's funny you and I share common heritage of being at Bell Laboratories in those very early days. So I wanted to ask you, firstly, a little bit about the context of the Chips Act, and specifically and we'll dig into more detail later on about a concept that I think you've been advocating called samovars. So just a few talk notes about that. And then I also want to learn more about Lam Research itself. So seminars, what does that concept and then Lam Research? What does it do?
Okay, well, maybe I'll start with Lam Research, for sure. So Lam is one of the world's largest providers of semiconductor manufacturing equipment. We provide equipment to everybody making chips all over the world. And I think it's fair to say that any electronic device, this is the prototypical device that people carry around in their pockets. And every chip in those devices, whether it's a phone or a computer, anything electronic that has chips in it, every one of those chips, I can guarantee you has gone through a lab machine at least once. So we're the ones along with our peers and our competitors, who enable all the the silicon technology to exist. We have more than 16,000 employees, we're growing extraordinarily rapidly. We've added 1000s of employees just over the last couple of years during this pandemic, to try and help our customers satisfy unprecedented demand for integrated circuits. were deployed all over the world. Our we have manufacturing facilities in California and Oregon and Ohio, in Europe and in Asia. 67% of our manufacturing is done in the United States today. We have r&d going on everywhere. Every place, our customers have a manufacturing site or development center of their own Lammas, they're also doing r&d collaboratively with them. But the focal point of our r&d, our most advanced r&d is in the United States, primarily on the west coast. So that's Lam. You know, the concept of the semi verse, The behind this, we're still figuring out what we mean by that.
So if I say if I may, you know, a lot of us hear about the metaverse and the applications and so on. So when you introduce the concept of seminars, I'm really curious about how you see it. And it's sort of evolution over time. So please tell us more.
Yeah, happy to do so. As I said, we're, there are a lot of details and a lot of concepts that I think still need to be fleshed out. But we've we I think some things are very, very obvious. So we envision, first of all, the semi verse absolutely is a play off the metaverse firm. And it really corresponds to a virtual representation of the semiconductor ecosystem. But one should not fall into the trap of thinking. It's only about the virtual representation, the virtual representation, and the physical reality that it mirrors think digital twins, everything have to coexist will continue to coexist indefinitely into the future you you can't have a digital representation of something without anchoring it with real data and the real world. So there's a very intimate Interplay that must take place, particularly in the semiconductor arena, advances in semiconductor research, whether its advances in technology, device technology, system architecture, Chip architecture materials, require innovative breakthroughs differences and thinking that you're not going to get from a calibrated model of what exists today. Good, right. So there's this interplay, as we learn more as we do things in the physical world, data will flow seamlessly into the virtual world, make the virtual models more and more capable. And someday, I would never say never, maybe machines will be capable of innovating in ways that humans do. But that's not the near term, or probably even the mid term vision. So we, when it comes to the semiconductor ecosystem, what we're imagining is that you've got a virtual representation, which is much easier for people to access, which is anchored in the real world. And the idea is that you can open this up to a vast number of innovators who can contribute novel ideas and run their experiments in the virtual space, to the extent that it's calibrated against the real space, evaluate new ideas, new approaches, and at for a fraction of the cost in a fraction of the time. That's that's one way of envisioning what the semi verse will be like.
So Rick, if I may, just to maybe recap what you said. I mean, your your industry, your technologies, are deeply grounded in the elements of physics and chemistry and an almost atomic level engineering level engineering. On the other hand, I think what you're saying is that by using these tools, which probably includes Data Automation, maybe artificial intelligence and so on, it becomes much more accessible for designers and innovators to be able to do what I interpret what you said is like, what if experiments to be able to do things in the virtual world initially, and make it much more efficient and faster to actually do things in the physical world as well? Is that the right concept?
You got it? You got it exactly. So the, as I said, this semi verse will have to learn and grow and become more capable in time. So initially, there'll be much more physical experimentation, as there is almost exclusively today, not entirely. And even there, we want to, we envision a semi verse that's set up such that people can access the physical access to assets without necessarily being physically present. So we see a network of interconnected laboratories. Where because if you think about innovation in semiconductors, we've talked about this a lot. You have to innovate, innovate up and down the technology stack. So no one physical entity is likely to contain all the capabilities that do allow you to innovate in a big way. So they have to be interconnected. You want to provide broad access, otherwise, you restrict the number of innovators, and you're going to restrict the pace of innovation. So there there is a remote access, broad access concept associated with the semi verse. Now as you generate data and imagine the laboratories that we have today being augmented with more and more capabilities, particularly centralization, interconnection, data management, so that the data are flowing seamlessly from maybe one facility to another. There, the data are stored with the right context, such that a large number of people can go in access data, mining the data and build a more and more capable semi verse over time, such this becomes increasingly the vehicle by which people start to test their innovative ideas.
So this means sort of being able to do well thought through remote robotic experimentation. Without you're physically having to go there and yet gain access to the data because you can control parameters, temperatures, different aspects of the process technologies that are fundamental issue of resonance, right, and to be able to do that with a multitude of innovations across the nation. Right,
exactly. But again, if we limited ourselves to that physical world, that would be a huge advance forward over where we are today. In my opinion, we're under leveraging the existing capabilities that are out there. But by using those data and continuously updating the digital twin representations of the real world We can start to accelerate the pace of innovation. And that's what's so critical in, in my mind the semi versus all about not just innovating, but increasing the pace of innovation. Yeah.
So for many of us trying to get it was changing subjects a little bit, which is the immensity of the complex problems you solve is quite baffling. You know, people talk about millions and billions of transistors on a chip. And yet you're operating your technologies to create structures, which literally are at the dimensional level of atoms and nanoscale things, and so on. And I was just wondering if you could give a give us some illustrations of that, because it really is, for the layperson, almost impossible to imagine that you've got trillions or billions of these things that are created in multiple dimensions, XY and Z, by by the tools that you build. So perspectives on that, because it's just quite baffling.
Yeah, I mean, to me, this is the essence of why you need a semi verse. Because the complexity and making an integrated circuit is extraordinary and getting more complex all the time. It's causing the ways in our ability to innovate. It's causing increases in research and development costs. It's causing increases in manufacturing costs that are quite profound. And so we have to find a better way to develop solutions to these quite complex integration problems, there's hundreds to 1000s of steps that go into making an integrated circuit. And to your point, the precision that's required is on the atomic scale, we're measuring things in units of atoms, in order to create these most advanced devices. But controlling things on an atomic scale is a little misleading, we can do that with new techniques, such as atomic layer deposition, where you put down one layer of atoms at a time, it's like painting a surface, even with typography. And we've learned how to etch one atomic layer at a time. So we can manipulate things on the atomic scale. But you have to have that atomic scale precision over an entire pizza sized wafer 300 millimeters. And every wafer has to look like every other way for coming out of the same machine. But also out of the chamber next door, or the machine next door, or the fab halfway around the world, all of those devices need to be atomically precise, and you have 1000 steps, where you have to have this kind of precision. So it's an extraordinarily challenging problem, which today is still being solved, if you will, mostly by empirical trial and error. This is where the semi verse comes in. Right? You can reduce the cost of the experiments by orders of magnitude, you can decrease the time to get a result by orders of magnitude. And that's why we think it's inconceivable to me that we have ever optimized the manufacturing of any integrated circuit with that level of complexity and the approach that we've taken today, we the process, the solution space is vast, on just one unit process and etch for example, we calculated some years ago, that there's more than 100 trillion permutations of recipes, we can run on that ECIR, that will make a measurable difference on the wafer 100 trillion. I've been told recently by some of my colleagues, I'm off by nine orders of magnitude. It's more like Avogadro's number of patients today. So there's no way that we can empirically explore that space and determine an optimized number. The only solution is through the semi verse is through a virtual representation, where we can run massively parallel experiments, and put these 1000 different steps together in different ways, and really hone in on the the best solutions, most innovative solutions, the most economic solutions.
So let me if I may, I just wanted to trigger a little bit on data and the value of data, I can sort of see a semi verse helping say a particular technology for a particular machine. But as you said, your company delivers equipment all over the world, to most of them large, and even smaller manufacturers, and therefore, how do you think of the amount of data there is the amount of data collection you can do? And whether you can aggregate that because my assumption is that if you're able to not only do the innovation, but then the production, get data and feedback, you could then learn a lot more about how your technology is actually working in the field and improve your processes as well. Does. Yeah,
that's absolutely true. And I think when it comes to semiconductor manufacturing, it's actually a data rich environment every time you're running away for, there's lots of data coming off of every machine. The wafers go through metrology. Again, they're measurements of precise measurements of device characteristics or structural characteristics across the wafer. Our customers, the semiconductor manufacturers, I think are, are well advanced and moving very quickly, in acquiring the data, conditioning the data, and then mining it, using it to improve their yields to improve their productivity. And they're in they're in a great place because they're in a big data world. Right? What we see though, on the r&d side, which is, and particularly research, the Fuzzy Front End of new product development, whether it's a new chip, or a new piece of processing equipment, we actually live in a little data world, a small data world, right? many instances, there aren't lots of data, the you you run some experiments, you get a little bit of data, and then you move on you try something else. How do you link all that together? Yes. And the we're limited in many cases, particularly with where we have three dimensional geometries, and complex topographies. How do you even measure what is your definition of truth, you have to measure something by destroying the sample, preparing it for transmission electron microscopy, very expensive operation, it might take a day, and 1000s of dollars for one data point, or a few data points. So how can we tackle that problem? Again, we go to the semi verse concept, if you take a virtual representation of your process, and we've done this, yes. And now we can run experiments in the virtual space and think about trade using the physics based model that has some degree of calibration with a little bit of data that you can now use to calibrate, say, a deep neural network, mostly using virtual data in a in a fraction of the time fraction of the cost. And then you can cycle back, inject a little bit more real data and make that model better. So it's a way of, of leveraging a little bit of information and getting a lot more value out of it.
So that that's a big challenge.
I think we're making progress on that Lam Research in particular is invested in that area. But I think the industry as a whole is, we're not the only ones out there trying it. So the data situation is very different depending on which part of the ecosystem you're looking at.
Right. This is a very inspirational discussion today. And again, thank you so much for joining us.
My pleasure. Thank you. Thank you.