 I'm Greg Garrick. So what do you do? I work in the research group at ARM. I work specifically on family process technology and more on the future process technologies, the road maps. So what is the research group at ARM? Is that like a special place? No, it's distributed. We have researchers in Cambridge. I happen to be in Austin, Texas. We've got researchers in here in the Bay Area and it's across the spectrum of ARM. So you have researchers at the architecture level, at the physical design level. There's specifically people working on IoT, on memory. So any of these technology issues going forward, I happen to be the one that works on a lot of the wafer process technology, future road maps. Is it R&D or is it like more futuristic research? In my case, it's actually pretty future research. It's more understanding the technology trends coming forward and how they might affect ARM products. So you're talking about, you had a presentation, like you were talking about a lot of things. What are you talking about? Well we, as I mentioned, we want to really understand the future technology nodes, what we think we can expect out of them. We have expectations as designers on Moore's law nodes that they're going to be cheaper, faster and lower power. And that's beginning to be very difficult to do. So the questions are exactly how much cheaper, how much lower power. So we spend a lot of time talking with the researchers in these fields to understand what can happen to or three Moore's laws nodes from now and see how we think that will fit with our product expectations. So down to ten or seven? No, yes, ten and seven are actually in the industry to find nodes. That's going to work, right? They should work. They're difficult. I mean, you've seen there's been semi-yield hiccups in the foundries and stuff's getting harder, so you expect those to happen. But in terms of actually making things that work and provide additional value in future nodes, we can see a path ten to seven to five, certainly. So we got ten, we got seven, and now we're talking about five nanometers and three nanometers. So that's pretty small, right? Yes, yes. There's a little bit difference in the naming of the nodes have gotten a little divorced from the physical features. When I say five nanometers, I really mean that the gate length of the transistors, which is a defining number for the nodes, that's still around 18 nanometers. So we're not at the atom level yet. We're still at things we can manufacture with known manufacturing processes. So you were talking about new materials. Is that the important thing? Without them, we're not going to be able to do it? Well, the traditional methods of scaling, which are lithography to make smaller features and then shrinking the transistors to get the power performance, those are running out of their capabilities. And the way to get those capabilities back is to add materials engineering to the picture. And we've had that industry many times. We've had the high-key gate dialectics in transistors, which we use now, but you'll see much more of that in the future in terms of wires, in terms of the materials we actually make the transistors out of. So all this whole time in Moore's Law, 51 years, it's all been silicon. And the day will come when we need to talk about things beside silicon to make the transistors. And humans usually when they have big problems, they find solutions, right? So we have to be what's called optimistic that it's going to be solved. It's easy to look at the basic math and be pessimistic about Moore's Law, but when you consider the proven ingenuity of our industry over this 51 years, it's really hard to bet against the interests in this ecosystem. And they keep coming up with interesting ways to get around problems. There are many cases of people in the past years that said, for instance, one nanometer will never scale past one nanometer. It cannot happen. But we continue to find ways. After one nanometer will be zero point something, right? It doesn't go in minus. That'll be interesting because yeah, we've gotten so far away when we talk about a five nanometer technology. We're really talking about things in the tens and twenties of nanometers features. So there'll have to be some kind of interesting naming solution when we get down below the one nanometer because we really don't mean that there are things that are that small on the wafer. And all these new materials and new research and all that, is it arm's responsibility to do that? Or is it just one small part of the whole industry doing this? There's two facets of that. One facet is kind of a forecasting need. We want to know what to expect out of the out of the foundries for our products going to our customers. But also, given the reach of our, we want a healthy ecosystem. We would like to see solutions. And there are many cases where going back to the materials question you asked, people have interesting materials, but they're materials people. They don't know what designers need. And we can provide help in terms of showing them this is how a designer would use a circuit. We can give them benchmarks saying this is how valuable we think your material is. Those are things that the design community can help bring in some of these novel materials more quickly and find the right choices by getting together and having this what I call design technology co-optimization. So there's graphene. You were mentioning graphene. Yes. And there's some other materials? Yes. So graphene is the one we all know, that won the Nobel Prize. It's hard to believe graphene is actually only as old as the iPhone. It's in technology years, very young. Graphene, it turns out, is very difficult to make a transistor out of. But it opened up this Pandora's box of all these other potential 2D materials. I showed in my talk probably 20 that are actively being researched. Some of them look pretty encouraging from the standpoint of how to make a transistor with some of these advanced materials. And so it's very important for ARM. The way ARM designs cores and processors to be very well connected with the actual physical world of chip making and the foundries and all that stuff. Yes. So you can't just design the processor without knowing what's going to happen. Exactly. There's a basic road mapping question. You know, how fast can I expect 5 nanometers to be? I need to make the right architecture for that process. But also, more and more, there are ways you can knowing the details of the process, build better products. And we want to make sure we understand that for our own benefit and for the benefit of customers using our products. So when you talk, speak with all these people, I guess, and the foundries and all that, it's going to be a pretty fascinating world, though. It's pretty cool. Yeah. It's a, I feel privileged to be able to do this job at ARM because it's, it's fun. It's really interesting technology conversations to have. So it's really, it's really neat to be able to talk about these things. So that's my background. I like devices and physics and that kind of stuff. Like, this is the world of like a chip that has billion things inside. And the designers who do that are pretty smart people also, right? The smartest in the world. Well, when you talk about a billion things, the designer who's actually dealing with the entire billion, they have to deal with an abstract world. They can't actually know what every transistor is doing. There are different layers that have to talk to each other. Eventually, you do, there is a circuit that has to work with actual transistors on a wafer. And that level, what we call the physical layer, the physical design layer, that is something we spend a lot of time on at ARM as well. And the masks that they use to, it's quite expensive stuff. Is there a way to bring that cost down? Yes. Or is it just going to continue to be more and more expensive? No, that's actually an interesting question. So I had mentioned in my talk that be careful, don't project the future given today's parameters. In the mass making, there are some very interesting technologies coming on board called multi-e-beam mask writers. That, so currently, masks are written off and with an e-beam, one. And it's slow and that's why it costs so much. There are tools in the product road maps that are 2,000 beams in parallel. So you can see that they're, they can bring the mass cost down tremendously with some of the resources they're doing in their tools. So across the board, mask people, resist people, etch people, all of them are making progress towards these goals. That might enable more, cost more crazy chips come out because if it's not so expensive you can experiment with different ideas. Yes. Without having only one chance to make it right and being more conservative than what you do. So that is, you've hit across a key issue. If the mass cost continue to grow, it really does start to box out a lot of the smaller development projects. There are other technologies that actually don't require masks. We don't know if they'll be make it to production, but there are maskless, maskless lithography techniques where you can actually use the e-beam directly on the wafer. And that has some encouraging potential to do exactly what you said. Allow small-scale trials and maybe new chip designs that wouldn't otherwise be able to afford the big mass cost. So there's Austin, Texas and there's around this area here. I mean not far from here, there's some big factories in Taiwan and in China I guess. So you go around and you speak with everybody. And they're all happy to speak with ARM, right? Yes, yes. It's a luxury that we have. ARM doesn't, we don't buy wafers. So we're one step removed from that. So we're able to have sometimes some pretty productive conversation that you wouldn't expect. If there's a business negotiation in there with wafer prices and what have you, that gets sometimes pretty serious. But at ARM, it's more of a let's let's talk about the general ecosystem. How are we going to get to the products we need across the ecosystem? And I guess at the ARM research group there might be a lot of things happening. Yes, there's a lot of top to bottom, top to bottom. So there's new architectures, there's looking at new memories, there's looking at things like neuromorphic computing on the application space, machine vision. There's just a ton of interesting research topics. What we found with Moore's Law is we've gotten to the point where we can enable some really new applications, the vision, the vision capability. That's really enabled by the fact that we can make these very, very fast high-performance processors. And your research is it only in like the high-end ARM chips or it's also helping out the IoT cores and everything? Definitely the IoT. And if you were there for Mike Mueller's presentation he showed the plastic ARM chips. That's an entire division of ARM research where they work on the IoT and specifically very low cost where you're not using wafers anymore, you're actually printing, you can actually print nowadays, you can print transistors with inkjet printers. That technology exists or you can roll them out on rolling machines. Those kinds of technologies are potential for ultimately low cost IoT embodiments and there's research in that area as well. So it'll be like 0.001 cent per ARM chip? Yes. Some of the some of the things you see proposed are giant roller machines that are rolling out sheets at meters per second and they can really, really get to some low cost. They might not be the lowest power in performance but for many IoT applications that's fine. So that's how we get to a trillion chips. Yes. We need a trillion. Those kinds of technologies would make the the number one trillion easier to grasp. Could you print out the also the three nanometer chips on that? No. Those are more on the micron level so they I'd equate that maybe with the 1970s or maybe 1980s type technology but for many IoT applications that type of technology is perfectly fine. It's just being able to do it at very very low cost with these newer printing technologies. So 2016 do you think suddenly could it happen in 2017? Some research comes out with like an amazing new thing and it changes the whole industry. Do you think that happens? Do you think that can happen? I hope not because that's my job so if that happens my boss might not be too happy. So I hope that we understand that there's there's a certain roadmap but there are always interesting new technologies and there's a host of new technologies we don't really fully understand in our ecosystem. One topic for instance is plasmonics where it's very interesting the optical world has come across the ability to to manipulate light information at the nanometer scale. It could have implications for the design of chips. I wouldn't see that fundamentally changing chips but they would it could potentially really improve the situation. That kind of stuff doesn't happen in a year. That just definitely takes five six seven eight years to identify the technology, work between design and system and technology, figure out a way to put it together on a chip and then start to make demonstrators. That's a year's long process. And ARM is always going to be there in the core of that future. Let's say printed chips and all that stuff. Yes. ARM is going to because it's about being an ecosystem of software and everything needs to work. Yes. So there's no like are you checking what computers are doing or basically intel research is is it similar to what you do? Like what they do? Well they run a factory they run a fab so they fundamentally are going to have a lot of researchers doing the actual development of technology. We don't do that. We don't have a fab at ARM. What we try and do is is work one step out from that one step out where we're talking to universities and consortia about the various technologies that are in there they're in the ramp towards the production world. Nice. And I hope in the new structure of the ARM as a company now maybe you get even more resources to get things the future thing happens sooner. Yes. Yes. That's possible. I think that's what's exciting when you when you look at the vision that that MasaSan brings you could see that there there might be more opportunity to invest strategically with some of these newer technologies.