Judy Warner (00:01.24) Hi, Richard. Hi, Roberto. So good to have you both today. I've been working on what I'm calling an AI design engineering revolution, and I've had the pleasure of working with you both at Keysight EDA. So thank you for coming on board and lending your voice and your expertise in this topic that is taking shape all around us. Why don't you start by introducing yourselves for our audience? Richard, how about you tee off? Richard (00:25.773) Hi, yeah, I'm Richard Duvall. I'm an EDA marketing manager for RF products. And I've been at Keysight five years. And before that, I spent over 20 years in test and measurement industries and video networking. Judy Warner (00:41.464) Hmm. Roberto, how about you? Roberto Piacentini Filho (00:44.881) Awesome, thank you, thank you Judy. I've been in the test and measurement space for over 25 years and for the last two years I've joined Keysight. I've been leading the design engineering software marketing team which includes ED, high speed, as well as RF. I also thought about saying a little bit of my fun stuff about me. So then I wrote a book called The Illustrated, Trippin's book and I'm a huge fan of Star Trek. So super huge fan. Judy Warner (01:12.696) That's why we're friends. We're fellow Trekkies. I love it. Okay, well, gentlemen, I don't know that you've been paying attention so much to my podcast lately, but I'm probably five or six podcasts in talking to a variety of different people that are talking about the way AI, LLMs, and machine learning are going to... Roberto Piacentini Filho (01:15.621) you know that's cool, fun. Judy Warner (01:42.22) really transform design engineering workflow across many segments. And so I know both from your high speed digital side of EDA at Keysight and the RF, there's a lot to talk about. So it looks like we're about to have a huge tipping point in this industry. That's what I'm observing. And so these tools are leveraging this technology, but it's very early days. So what do you think's coming next as we lean into 2025? Richard (02:19.897) So Judy, that's a really good point of discussion because AI assisted tools automate complex design tasks and they can reduce the time and the effort required for things like circuit design, layout optimization, verification, all of the tasks that we're doing at the moment. But we can't incorporate AI into a workflow unless we can automate the workflow. So I'll give you an example of some of the things that we are up to in. In ADS 2025, we're introducing a whole set of Python APIs. And you know, the idea is to unlock the platform. A lot of things that we do are preset or have limits. So by the idea about unlocking the platform makes ADS easier to orchestrate and makes the tools customizable and easy to integrate. And so that's really just step one, which is open connected workflows, which are really enabled with Python APIs and Python scripting language. And really we're calling, following our customers lead. They've told us they don't want to be locked into one EDA platform. They really want to be able to choose best in class workflow elements and integrate them together. And so for us, this is just the beginning. Judy Warner (03:36.418) You know, we talk about heterogeneous integration when it comes to chips and packaging and all of that, but it is absolutely tool and it comes to the technology stack that design engineers want to be able to use. And you're right, they want that flexibility. So I'm happy to see that the industry is large is leaning into that trend. So on the RF subject, I had a little stint of my life, mostly around our PCBs. And know, IoT is everywhere and along with AI and LLMs and machine learning, you know, we're always pushing the frequency and now we're moving towards 5G and 6G. So can you give maybe a more tangible example of how that workflow gets stitched together? Richard (04:32.857) Well, I've got two really good examples of how we see AI being used as an engineering assistant. So in this advanced 5G on the path to 6G trend that we're on, one example is 6G will use AI and network operations to overcome path loss and improve throughput. So for the physical layer research that's going on, our system design product is being used to do two things at the moment, generate training data sets that are reliable real world 6G faded waveforms for AI ML algorithms. And the second is to validate AI ML algorithms under real world environments to quantify design margin and performance. that's, know, that there's a small group of people doing that. Another good example, this is something we're introducing in the launch event in December that I'll talk more about in ADS. So we've come out with an AI enhanced load pool toolkit and load pool is a big issue because of all the beam forming and phased arrays, the effect they have on power amplifiers. So this load pool toolkit creates advanced simulation models using artificial neural networks and it integrates measured data from lots of different sources that are in different formats. And then it extracts the simulation model and validates it in just a few minutes. we've got a great video and an app note on this topic that we'll link in the podcast. Judy Warner (06:18.08) Okay, good. I'll be interested in seeing that because the other thing I'm hearing is about how to make sure that with these LLMs, like we're not pulling the data from the internet. And I'm sure that I know Keysight and I'm sure that it'll be really interesting to talk about how exactly that's done. you know, seeing things go from maybe weeks to minutes is truly stunning. And I'm seeing it all around me right now. Roberto, how about you? You're more on the high-speed digital side of the house. so what's happening in regards to AA? So Roberto, what's happening with the high-speed digital side of technology that you're overseeing? Roberto Piacentini Filho (07:01.745) Yeah, no, this is a really good question, Judy. One of the things that we talk often with AI, and I don't see people talk a lot about this, it's really AI infrastructure. If you think about the... For AI to work, it requires a lot of computing power to be executed, right? And where does it happen? Data centers, right? This is all about data centers. Now, if you think about how data centers, they operate and they work, this is, it's a computer. It's a computer. So you have a CPU, you have GPU, have an AI, I'll hub on it to talk. like USB, UDDR, PCIe. But think about that, you're have hundreds of these compute units inside the data center, right? And it doesn't get really scalable if anytime when you need to run a new AI model, a new... AI workload on the data centers, just, let's go buy it. Let's go in one of those. Back in my days, it was like a store. I don't know if it exists in California. Well, it was kind of foreign. used to be a large technology store, right? I would go to Fry's, buy it by the board, a CPU, just put stuff together. You cannot do that kind of stuff with the data centers. And if you're going to put a computer, a server together for every new model you've got to put in there, there's no physical space. Judy Warner (08:06.478) Mm-hmm. Roberto Piacentini Filho (08:22.045) Computers are kind of large. So what people started to do with AI is that they started to build this dedicated AI chips. And if you think about a GPU, a GPU is kind of a dedicated chip on itself. It's a chip that does something really well, like for example, on the Matlab. So think about, you start, in the past you had the motherboard, CPU, GPU, put everything together, large computer. Judy Warner (08:23.787) Mm-hmm. Judy Warner (08:34.296) Mm-hmm. Roberto Piacentini Filho (08:46.043) Fast forward to today. You have like a what people call a system on the chip You put all of these pieces together in a very tiny space because system on a chip now What is about system on a chip? Well, it you design it's it's a large monolithic thing You have the CPU block. You have the CPU block the GPU GPU I hub And and then you you make this device you put on your data center If you need to, let's say you need more GPUs, what do have to do? Well, you have to start a whole process again, build a new system, a new system on a package, this whole thing takes time. So do you think this is scalable? What are your thoughts for you to be redoing this thing all the time? No, right? It's not. Judy Warner (09:31.533) Mm hmm. Roberto Piacentini Filho (09:32.899) So there's something that happened in the last couple of years that is not really a new technology, it's chiplets. So imagine you have these common blocks, the CPU, the GPU, IOU, you make this monolithic device. But I imagine that you break this into smaller components. So then you have a chiplet for the CPU, a chiplet for the GPU, a chiplet for the IOU, and then you put some very high speed interconnect in between them so then you can have the data flowing back and forth. So instead of this monolithic design, you have now what people call this heterogeneous design. You have all these different packs together, these different triplets together. interconnect like UCIe or a bunch of wires, BLW, to send it back and forth. And then when you need to, let's say you need more GPUs, all you do, well, let me put another block of GPUs in here. So it's very scalable. The other interesting thing about, and you might know this, the other interesting part about Hitcher Genius Design is that you don't need to have all of these blocks using the same technology. Let's say maybe the USB controller you're using five nanometers, seven nanometers, which is very well known technology. It's a cheap technology. do that. Perhaps for the GPU, you use three nanometer. You need to pack a bunch of transistors there. So you have the ability to really cost optimize your design with this various, this different technologies for the different blocks that you need. Now this makes it really hard for a designer, for example, to design all these different things together if they don't have the right tools in place. And this is one of the things that Keysight is really keen on, is trying to figure out ways to help the designers to do their jobs in a faster, easier way. So I think you and I talked a couple of months ago, we released a really cool technology that Judy Warner (11:16.599) Right. Roberto Piacentini Filho (11:35.354) was introducing the Keysight EVA, Chip.Fi designer. So this is part of the ADS software and it's an amazing tool that really helps the designers incorporate Chiplets into their new designs. Judy Warner (11:39.16) Mm-hmm. Judy Warner (11:50.38) You know, around that time we were talking, Roberto, I had the chance to make a chiplet podcast with Steven Slater, who is a product manager on your team. And it was fascinating, but I also noticed that he said, which is sort of what we're talking about now, like we're definitely making En-Rose, but the days are still early. So I kind of want to throw a question here we didn't talk about ahead of time, which is how do we manage what's here now? and be prepared to move into these, this future of technology. I mean, if you're a design engineer, what things should you be doing or learning or adopting right now that you could go ahead. Roberto Piacentini Filho (12:32.371) Well, I think. Yeah. In my head, I think you should not be afraid of automation. You should really embrace automation in this new age of AI assisted tools. I mentioned the Keysight Triplet 5 designer, but I figured to say that one of the things about you really embracing these automation tools is that you have a company like Keysight that has decades of management science, many, many years of design experience. Judy Warner (12:48.824) Mm-hmm. Roberto Piacentini Filho (13:03.845) expertise. We participate earlier on into several industry, industry bodies like the cheaply the organization for for DBR. And as we participate in those things, we have not just the opportunity to influence given our background in the management side, but we also learn firsthand what is changing. So then we can anticipate how the market is changing and then we can implement this functionality in our in our solutions. So in this next generation of our software, Judy Warner (13:11.233) Mm-hmm. Judy Warner (13:23.95) Mm-hmm. Roberto Piacentini Filho (13:33.701) 2025 version, we actually introduced the BunchOW support for the Chiplier Center Connect. We are also introducing additional analysis for CrossDoc that also includes support for Part 3 Clock QDR. which if you have all these tools in core, there's all this functionality, it's been functionalities in the tool. This means that the designers can just click a couple of buttons and do the analysis. They don't have to be doing all of this stuff by themselves manually, you know? So this is one of the things I think people need to be really open to is embrace automation. It's here to help, you know? Judy Warner (14:08.46) I've, I, since I've been doing this series on AI, I've had some funny conversations on LinkedIn and there's always the old stodgy engineers that are outstanding engineers, but you know, they'll push back and say, yeah, I can never do what an engineer, but the message I'm hearing over and over again, and I hear from actually you both now is that it is AI assisted. And I like the word that a lot of people using in their marketing materials is AI co-pilot. But everyone is saying right now, it helps take some of the tedious stuff out of your design workflow and leaves you in the driver's seat to do the really critical engineering judgment and use your intuition so it actually frees your hands up a little bit. Is that your sense of what Keysight is leaning into in this regard? Roberto Piacentini Filho (15:01.427) Yeah, and you know what, can I use a, perhaps a, I can't help, but I'm gonna use an example from Star Trek. I guess I have to say the easy thing. When I was younger, I used to watch those movies and I said, no, this is not real. You see, like, I could, Judy Warner (15:08.695) You Roberto Piacentini Filho (15:16.507) teenager, know, somebody in the teens years, years designing robots and designing very sophisticated machines. And then I would say, no, this is not possible. This is never going to happen. This is what really is AI assisted. You know, these people, they're not more intelligent than us today. The people from Star Trek. It's just that they have the right tools that are allowing. Judy Warner (15:36.36) Absolutely. Roberto Piacentini Filho (15:37.627) somebody with very young age to do things. is not doing the job for that person. The person has to say, computer, can you please create a circuit design that does ABCD and does this other part? It gives you a really fast quick start. It abstracts the beginning. So then you don't have to start from a blank page. where do go from here? So I think that this is now actually, when I look back those movies and seriously, you know what? I was completely wrong. failed to see the future and see that this is actually possible. With the right tools, those things are possible. And this is I think that we are going with the AI assisted tools. And Keysight is in the forefront of these applications. Richard (16:09.273) in the future. See you then. Judy Warner (16:12.995) Yeah. Judy Warner (16:23.128) I want to go back to your thoughts about the infrastructure because we're in this funny place. And I'd like to hear from one or both of you on this is that we're using AI to help us do the jobs, but then we need to make data centers and the chips that support us to, you know, all the compute power. And then there's power integrity problems and thermal problems and all of that. And we're all, we're doing it all at once. And I saw a really interesting. short of someone walking through a data center with all like oil-cooled, know, 8 NVIDIA GPUs on in one system. It was completely quiet. It was so space-age like you're saying a Star Trek. looked like science fiction. What are the implications of that data center and where are you guys positioned specifically around the data center? I mean, you've talked about chiplets and those kind of things. So we can do it more in a modular or systems way, but... Roberto Piacentini Filho (17:23.911) Yeah. Roberto Piacentini Filho (17:29.895) Yeah, this is a really good question. And we have a colleague at Keysight, Heidi Barnes, that she always says, don't forget power. She has this funny thing that says, forget power. And this is true because we don't always think about the power delivery and reliability. This is, well, I take it back. We all have to think about it. But the thing is, it's on the most important parts of circuit design. And just to keep it simple, it's not just about power delivery and reliability, but it's really power integrity. And then if you think about it in training devices, you this is super low power. Judy Warner (17:38.903) Yeah. Judy Warner (17:48.12) huge. Roberto Piacentini Filho (18:05.683) And then you think we started with five volts and then if you look at the components that I mentioned before about the components of the compute units, the data center, the CPU, GPU and so forth. A CPU today, a modern CPU, uses anywhere from 1.3 to 1.8 volts. A DDR, 1.2, and then GPU is less than a volt. Well, super low voltage, right? But then think about that this is a single competing unit. In the data center you have Judy Warner (18:34.848) Exactly. Roberto Piacentini Filho (18:35.669) hundreds, thousands, right? So and then I was kind of just back of the napkin type of calculation. And then if you think about, if you multiply by, by, Judy Warner (18:38.071) Right. Roberto Piacentini Filho (18:45.043) thousand by hundreds of this kind of thing, we're talking about for the CPUs alone, over 300 amps. And then for the DDR, another 100 amps, then GPUs, which you say, less than a volt, but those are the ones that I was mentioning earlier on the AI dedicated chips. They are the ones that really crunch the numbers for this large language models, you know. Judy Warner (19:07.544) Mm-hmm. Roberto Piacentini Filho (19:07.955) So unless it's less than a volt, but by the thousands, it's over 600 amps. So we are talking thousands, like over 2,000, 3,000 amps in a data center. You have to deliver this in a very reliable manner, consistent manner. And then when you get to these levels of power, then think about the electromagnetic and thermal issues that can occur. Yeah, exactly. Yeah. Judy Warner (19:29.8) And signal issues and, and it just EMI, all of it, just all stacks up. really, in my mind, not being an engineer and not being Heidi or Steve Sandler, you know, they did that. He created that thing, the 2000 amp, you know, refrigerated probes. And I'm like, what? But did it, you know, and used your tools to, to do it. So it's pretty incredible. Roberto Piacentini Filho (19:37.209) It is, it is. Thank Roberto Piacentini Filho (19:44.976) Exactly. Roberto Piacentini Filho (19:57.031) Yeah. And I'll you one thing, if the listeners, if our audience are interested, we have actually a really nice paper on, Heidi did a really cool paper recently. We can put a link for people to download later. But back to your question there. then, yeah, so then what we're doing here. So we are continuing to help the engineers to have this single. Judy Warner (20:09.357) Yeah. Roberto Piacentini Filho (20:21.459) solution or platform if you will that can really tap into high-speed digital, RF, power integrity, signal integrity, triplets, all in the single group of products if you will. And for power integrity in 2025 we're bundling together, we are calling it the Power Integrity Designer Bundle. So then basically we're putting together our super well-known and really really powerful product, PI Pro, Power Integrity Pro with conducted Judy Warner (20:48.526) Mm-hmm. Roberto Piacentini Filho (20:50.009) EMI, people also call it Semi. So designers can have this complete end-to-end digital twin solution that covers the entire power delivery, network design workflow, all the way from pre-layout, post-layout, and debug. And one thing I have to say is that one of the key differentiators of our solution is that because of our measurement expertise we've been doing this for quite some time. Our solutions they are super high quality, the high fidelity models that you can get for power analysis it's amazing. It's a huge advantage for designers to have such a tool for end to end, the entire workflow. It's a time saver, it's really a time saver. Judy Warner (21:13.986) Mm-hmm. Judy Warner (21:33.966) Yeah, well, every before I really had gotten to know Keysight EDA, every, you know, all the experts, Eric Bogatyn, Steve Sandler, Heidi, you know, all the signal integrity, the power integrity, EMI, everyone would say, Keysight is only one that has end to end. And many of them have taken the time to become a Keysight expert. That spoke volumes to me because that wasn't you talking about you. was people. using it on really challenging ways in the real world. So that's really what compelled me to want to learn more about the solutions you're providing is because people outside of Keysight just can't say enough good about it. But then also you're leaning into using this other parts of the technology stack like Ansys or Altium or whoever it is you're partnering with. So people can regulate. their own workflow. Richard, Roberto and I have been chatting way too much, which we could do all day long because we're both chatters. So I know that Roberto mentioned and I'd heard from you that ADS, you're about to launch or you have launched, I don't know, the new, annual release of ADS. But what is Keysight? EDA direction overall when it comes to AI and ML, you know, going ahead from now into 2025. Richard (23:07.321) Well, that's a great question. especially with this product launch that we're in the middle of, we're integrating AI ML across device modeling, RF circuit design. All of the groups that we work with are leveraging Python integration. But one of the things that's, I think, a measure of what we're able to offer at the moment is, Keysight EDA recently went a bid with the US Department of Commerce called AID RFIC Design Enablement Program, which is a mouthful. But the AID part stands for Artificial Intelligence Driven RFICs. the program group is called NATCAST. And the aim of the program is to automate complex design tasks using AI-driven tools. Judy Warner (23:46.275) Yeah. Richard (24:04.297) And what Nat Kess talked about in the webinar, which was the invitation for the bidders, was a real concern about workforce and nail madness. The technology becomes more and more complex. We're losing senior people as they start to retire who have 20 years of doing design experience. And we've got new people coming out of university that can't possibly ramp up that fast. this project has an aim to help towards that. and there are specific project deliverables. The Keysight team in this project, which includes other partners, aims to make the AI enhanced workflow accessible to designers of all experience levels. So the definition from the program is by driving automation optimization and generation of advanced and unique RFIC designs, which is RFIC and MIMIC, by the way. Judy Warner (24:57.57) Okay. Richard (24:57.629) The project will improve the productivity and competitiveness of participating members in the US semiconductor industry and also become a framework for innovation. really our, what we're looking forward to is by participating in this project. Keysight DDA expects to help advance AI driven technology, assisted engineering as we've been talking about, which will directly benefit the state of. Judy Warner (25:20.8) Mm-hmm. Richard (25:24.712) RF circuit and system design in the medium term. Judy Warner (25:29.216) Yeah, it's funny because I've been doing this series on AI and design automation. I had the chance to go down to an advanced packaging materials company in San Diego and sit in on a semiconductor executive round table. But there is a presentation by a data analytics company that will help create that cohesion in the workflow too. So I'm just seeing this, you know, from the government, from, you know, it's coming at me at all sides and it's really quite a view. And what you're saying just sort of ties everything up that I've been seeing with the bow on top, right, is it leads to greater productivity. It helps engineers on board faster. They still need to know how to engineer, but they get those core skills in college. So it's just so interesting. to sort of see it half baked and it's really exciting to see the things that the Keysight's doing and really leading the charge, which you guys kind of always have. So this has been a fascinating conversation. So gentlemen, any last words before we wrap up? I know that you're very busy and I need to let you go, but any last thoughts before I let you go to wrap up? Richard (26:52.653) Yeah, one thing I just want to make sure everyone knows about is we've got a launch event webinar on December 3rd that we would love everybody to register and come find out a whole lot more from our experts about what we're doing in these two areas. They can register for our live webinar on December 3rd to learn about the latest release. And if they can't watch it live on December 3rd, can... They should still register and watch it on demand. And we'll put links in the podcast here for you, Judy, that they can access the demo videos, the app notes, and the white papers. We've got a page on our website that we call the What's New page, where we post all of the new stuff all in one place. And they can go check out all the new stuff on those What's New pages. Judy Warner (27:45.634) Great, well, I'm going to get all those links from you for sure. And I will also share it not only with our podcast audience, but also on LinkedIn, because this thing's moving so fast. And I think the work you're doing at Keysight EDA is fascinating. Well, anything from you, Roberto? I have one really quick thing to mention. There's one white paper that I wrote that is very interesting and it's going to be on the page that Richard mentioned. It talks about how concurrent engineering, it's helping fuel or accelerate... AI adoption. It's a really fun ebook that we put out there. It's going to be on the page that Richard mentioned. And I want to invite everybody, come see this special event we put together, EDA, Keysight EDA 2025. It's a must-see event. Judy Warner (29:01.983) Well again, thank you both for giving us an insider's view of everything you're working on and I'll make sure I include all those links and resources for our audience. We will see you next time. Gentlemen, I hope you come back in the new year and give us more updates on what you're doing in this exciting area. Richard (29:20.729) Okay, Judy, talk to you soon. Roberto Piacentini Filho (29:22.779) Bye. Thank you, Judy. Bye. Judy Warner (29:24.127) See you soon. For our audience, thanks so much for joining us today. I trust you enjoyed this conversation with Richard Duvall and Roberto Piacentini. We'll see you next week. Until then, remember to always stay connected to the ecosystem. Babam.