SemiWiki.com - Podcast EP283: The evolution of Analog High Frequency Design and the Impact of AI with Matthew Ozalas of Keysight

Episode Date: April 11, 2025

Dan is joined by Matthew Ozalas, a distinguished RF engineer at Keysight Technologies. With extensive experience in RF and microwave engineering, Matthew has made significant contributions to the field, particularly in the design and development of RF power amplifiers. His expertise spans hardware and software applications… Read More

Hello, my name is Daniel Nennie, founder of SemiWiki, the open forum for semiconductor professionals. Welcome to the Semiconductor Insiders podcast series. My guest today is Matthew Ozals, a distinguished RF engineer at Keysight Technologies. With extensive experience in RF and microwave engineering, Matthew has made significant contributions to the field, particularly in the design and development of RF power amplifiers. His expertise spans hardware and software applications as well as design and automation.

Welcome to the podcast, Matthew. Great to be here. Thanks for having me. First question I'd like to ask is, how did you originally get into semiconductors? Do you have an interesting story to tell? Well I don't know if it's interesting. It's probably a common story but you know I started out in college as just kind of a general engineering major and I took a class.

I then went into comp sci so I started doing like basic you know programming stuff in C or whatever and I took a class on the logic gates. I don't know if you remember that class, but it was like, and in or gates and the transistor logic. And something just hit me and I really liked that class and I got into it. And I started to think about like, how do they actually build these, these logic gates, you know, this is really interesting. And so I, you know, I dove down another level and got in, I took the transistor class and transistor and switches.

Eventually, I ended up in this RF microwave wireless engineering field. I did an internship and I started using some of this test equipment and I found that fascinating, like these radio receivers and everything. So, you know, I started, I started researching that more. I took more classes in that and, um, eventually ended up doing this. So ended up in this industry and doing, um, software, software for high frequency circuits. So it's been pretty fun. And I, I don't think I could have predicted the way that wireless

technology was going to take off. You know, at the time cell phones were hardly a thing and there wasn't all that much to do with wireless but nowadays, wireless is everywhere. So everybody is using wireless for all kinds of things. So it's been very interesting and exciting. Good story. So what brought you to Keysight? Yeah, that's kind of an interesting story.. For me, I was a hardware engineer. I didn't dual major. Eventually, I just went into the circuit design, hardware engineering stuff, and I went to

work for a company that was primarily doing research in wireless systems. We were doing phased arrays and things like that for communications applications. I started working on these circuits which were called power amplifiers. And I just found that they were, you know, they seem simple from the surface. You know, a power amplifier, it's like, if you think about, if you're people that are into audio or recording and stuff, you can't take your head, you know, your little headset jack, you can't just plug that into a huge speaker, right? You need a power amplifier, and the power amplifier is essentially giving the signal enough energy to like move your speaker. And so it's the same thing for high-frequency circuits, except instead of a speaker, we've got an antenna. And so you need

enough energy to kind of bubble out these electromagnetic waves from the antenna. But these circuits were, they seem really simple, but when you dug into them, they were incredibly complex because they were actually getting into the operation of the ideal current source in the transistor itself. And like the techniques to make these things work had to do with switching these current, you know, the current on and off in there and creating all these crazy wave forms in the, in the time domain. And so it just, for whatever reason, it kind of fascinated me. And I really got into trying to figure out how they work. And so I did a design there, and then I decided I wanted to just work on these circuits,

all the time. So I went to a company called Skyworks, and I started working on handset amplifiers. And so I was in the lab, just like cutting traces and doing crazy things with diamond scribes on chips to try to make these things work. And got a lot of practical experience doing that. So I was designing like super high volume stuff for some of those flip phones that were coming out and then eventually the smartphones as well. And I learned a ton there. That was just a very empirical thing.

We were in the lab. But at some point, the designs, the technology started getting really complex. And it got harder and harder to solve the problems that we were seeing in the lab. So increasingly, I'd go in and work on simulation modeling. When I first started doing that, the simulators, they didn't work.

You just went, OK, this simulation sort of works, but we got to go in the lab and make this thing work. But by the time I left, I mean around, I guess, 2013, the simulations were getting pretty good. The modeling and the simulations were getting pretty good. Two things drove it really. Compute power, so the computers were getting better, and also the electromagnetic solvers were getting better and also the electromagnetic solvers were getting better.

So we started to be able to predict very accurately what these, these complex products would look like to the point where, you know, when I first started, you know, you got your, your design back and you'd have to go in the lab for days and just, you know, tune it, just get everything working perfectly. If you were lucky, you'd get it working. Um, but by the end of my tenure at Skyworks, you know, I perfectly. If you were lucky, you'd get it working. But by the end of my tenure at Skyworks, I was there for about 10 years,

these things were just coming out of the box and working. And then we would just, we'd have to worry about the, the second and third order problems, which were increasing. So I found that very interesting. And I kind of thought, you know, this simulation thing is interesting and maybe I should go work on that a little bit. And so I came to Keysight to work in the EDA group simulation thing is interesting and maybe I should go work on that a little bit.

So I came to Keysight to work in the EDA group doing simulation technology solvers and trying to put all these things together. And I came in as an application engineer, application development engineer. So primarily my job was to kind of work with the software and provide test cases, examples, educational material, to show how the software could work on these real hardware designs. So it's kind of an interesting blend for me. The thing that was appealing about Keysight, right,

was that, I spent a lot of time in the lab too, so Keysight makes all the hardware and test equipment. And so I liked the idea that Keysight did both hardware and software. Yeah, that's a good fit. So let's talk about AI. How is the rise of AI technology transforming the RF and microwave engineering landscape? Sure. Yeah. Well, I mean, I'm doing some work in the AI area. And I think I'm sure as you know, AI has got the potential to change just about everything in every industry,

right? So no matter what you're doing, whether it's science or tech or law, right? Any industry you're working in, AI is coming and it's going to change things. So RF and microwave engineering is no different. Maybe we'll start with some of the challenges. What makes AI really difficult to apply for RF and microwave engineering, right? It's a small niche.

And I think the first challenge to me is, you look at these circuits, right? You look at the types of designs people are doing. They're almost like pieces of art, as opposed to something like a pure digital design that's somewhat prescriptive. It's the process is coded, I think, in that case. And so for our microwave circuits, it's this art. And if you look at AI tools, right, and you can

imagine trying to tell an AI, hey, design me this circuit, go, do it, come back with an answer. That's challenging because AI tools are really good at analyzing problems, right, sorting through data. They have this training set of data, but at a fundamental level, they don't really understand the way that we as humans understand. I mean, they do have some level of understanding based on their training set, but it's not the same thing as what humans have, that understanding. And that means, I think, at high frequencies, we've got to apply these tools in the right context to be successful.

It's probably not a good idea to start out just telling this thing to design something. If you think about that power amplifier design problem that I was talking about earlier, I've got a lot of experience doing that. And if you look at someone who has that level of experience, right, they're just, when they make these design trade-offs, they're leveraging years and years of experience.

And some of it is explicit, like they learned, oh, you know, it's not good to put a capacitor here because that's going to increase the feedback and cause an instability. But also there is implicit experience. There's like, you know, you make these judgment calls and you don't even really know why you do it, but you can intuit your way through these circuits

when you've got a lot of experience and you've designed these things over and over again. And I think the problem there is, you know, the data set that is involved with that, it's not really captured. Like there's no explicit, no one's written down everything that an experienced designer has ever learned and to train an AI on.

So I think that's challenging. I mean, if you talk to designers, I don't even think they know why they, I don't even think they can describe exactly what is driving them to make a decision. So that's number one. Number two, AI relies, we all know it relies on lots of data, right? So if you look at like the chat GPTs, they've probably consumed, you know,

there's debates about how they already run out of information, you can, you know, that's interesting, and they've kind of consumed all the data on the internet. But then, you know, we look at RF, micro, you know micro high frequency design, RF microwave engineering, it's a lot harder to come by high quality good data for that little subset, right? And so it's some of the things that limit that, I mean, if you think about some of the data,

especially in high frequency design, when you get up to like millimeter wave terahertz, some of that's like sensitive to national security. I mean, that's completely locked down. That data is not going to be going into a public AI training database, right? And the data that's, even if it's not locked down for national security, you're going to run into the case where a lot of the data is proprietary to the company. And so, you know, a company like Qualcomm, Keysight even, we're not exactly gonna release our state of the art cutting edge design IP

so that AI tools can go train on that. So, just as an example, let's take a 5G chipset, maybe again, you'd like to tell an AI, well, go design me a 5G chipset and come back when you're done and I'll go on vacation. But if you think about all the technology that goes into that, it's like semiconductor processing, how these actual devices, the physics, how they're working. Then there's models and you've got design kits. You've got circuit topologies, you've got

simulations, data, physical design, reliability. I could go on and on, right? But essentially what I mean here is that it's like years of incremental learning that's taking place. These products, we didn't just get to a 5 or now a 6G transmitter. We didn't just intuit that from nothing, right? We've evolved that technology over the last 30, 40 years, really, or more than that, really, 100 years, if you think about that.

And so there's not really an open source version of that incremental learning that you can capture in an AI. And so the companies that are out there aren't really going to open source this design IP. So I guess back to your question, I mean, transforming, if we get to the transforming idea, I think it's just going to happen a little differently. I don't think it's going to be like this prescriptive circuit

design. I think it's going to be more in terms of efficiency of design assistance. You're going to have not a complete engineer there. You're going to have a tool that's going to help make engineers more efficient and maybe change the way design gets done. Yeah, no, I agree completely.

So how effective are companies at leveraging their own information in the design process? Yeah, so the information resides in these companies and that's kind of an interesting scenario here because the company owns the IP. And so what I think is going to happen is I think that these types of AI techniques, they're not going to rise from a centralized place in the industry at first. I think they're actually going to kind of be local, these little local pockets of design

capability, AIs that are rising out of that. And that's totally different from the way things have gotten, you know, things have been done in the past. If you think about how things are done before, right, you have these software tools and they're almost prescriptive in the design process. You know, we build these UIs around the design software and we, you know, we essentially prescribe

a design process through that. I mean, that was a long dance for many years with EDA software tools. You had processes and technologies, EDA tools evolved around those. Then the EDA tools started to dictate the processes. So that's not the state of the industry and the company right now. But I think the biggest challenge really is that it's hard to get information into that type of a design process, right? These EDA tools are now overly U.I.ed and overly automated. And so you have some random piece of data sitting around in your company.

How do you get it into those tools? How do you make that actually a part of the design process? So it's both that the information is in these weird formats, but it's also, it's kind of both sides. The EDA tools also aren't in a state where it's very easy to read all this information. So I think that's really something that's gonna change in the next couple of years to me. The real key to using these AI tools effectively,

if you're gonna get scenarios where these AI tools are rising in these local enterprises, maybe largest enterprises first, they have to be able to access and control their information in the design process, right? So that's that relationship that the enterprise and information has to evolve. And I talked to a lot of customers. I mean, one of the things that I see, especially when I talk to the industry of customers. I mean, one of the things that I see, especially when I talk to the industry leading customers out there, they're kind of thinking a little differently about information.

We're doing the same thing at Keysight too, by the way. You know how like information, we tend to think of information as something like intangible in the business sense, it's what we call an intangible asset. It's not a physical asset. And as such, it's what we call an intangible asset. It's not a physical asset. And as such, it's been something that I don't think companies have paid a lot of attention to.

But now you start to throw this potential for AI into the mix, that's very disruptive. And so I can see a transformation where the information becomes more like a tangible asset, like a wafer fab. And what do I mean by that? Not from an accounting perspective. I don't wanna tell accountants how to do their job. It's, that's kind of their thing. But what I really mean by that is,

when you think about something like a wafer fab, that's a tangible asset, the way that I think companies and management teams think about that stuff is, hey, if we're not using that, we're losing money. You know, this isn't, there's an opportunity cost to not filling your way for that, right? And information is kind of the same thing.

It starts to become kind of the same thing. Look, if you're at a large enterprise, you potentially are sitting on a goldmine of information. There's a lot of information there. You could leverage that now with these tools and technologies that are coming up. And if you're not doing that, you're losing. That's an asset that you're not leveraging, right?

And so I think that kind of becomes a challenge for these companies. We have to get in the scenario where companies are able to leverage their information. And one of the things I wrote a kind of a paper recently on this, and one of the things we looked at is, okay, how many designs, like IP perspective, how many schematics, layouts, designs does a company, you have an enterprise or corporate level? And it's kind of, you know,

it's not as crazy as the Drake equation. You can come up with like a Drake-like equation though, representation of that, right? It's not that complicated, but you basically, okay, we wanna find the unique designs in a company, right? It's like, how many workspaces does each designer have and how many unique schematics and layouts are in each workspace or project

or whatever you wanna call it, right? And so maybe if you have a really small company, like startup, you might be talking about, I don't know, 5, 10,000 designs. When you start getting to this scale of some of these bigger companies in the industry, I mean, you're talking about millions and millions of unique schematics, layouts, IP.

And so that's a tremendous asset. And when we talk to companies, they're not really using it right now. Why aren't they using it? What's preventing them from using it? It's not like they didn't think about this. It's more that the, I think the asset is hard to get into the design tool. How do you, how do you leverage that into the tools? And, and again, it goes back to that, that overarching platform, that UI that we have, that prescriptive design process that was built around circuits from maybe the last generation of designs. That doesn't work as well in this scenario. What do you really need? You need flexibility. You have to be able to automate your processes so that you can bring in, you can access that

information. So one of the things we demonstrated, like we ship a software product, ADS, Advanced Design System, and we have something like, I don't know, we have a couple thousand examples in our database that ship with the product, which is kind of, hey, here's how to design a low noise amplifier, just kind of these examples of how to use the feature. So what we were able to do using these automation APIs, we were actually able to go through

all the designs, extract them, centralize them into a database. And this was all done with scripting and not even very complicated script like anyone could write this script. We extracted all of the net lists from our circuit schematics and layouts that we had. And then we started to go like, what can we figure out from this information? So one of the things we did, we just wanted to find, okay, let's see if we can write a Python script

to find all of these, all circuits that have a particular topology, like maybe a ladder, we use the ladder topology. So that's like series L shunt C, series L shunt C. If we've got a couple of those, can we identify it in our database, the designs that have this? And we were able to. We were able to go through the database

and we went through a couple thousand examples. We identified maybe 12 designs that had this topology. And if we can do that with a simple topology, then anyone can do that with a more complex topology. And when you start thinking about the implications, I think it's pretty profound. Because now you can imagine that you have a circuit that has a problem.

Maybe this letter topology is causing some manufacturing problem, like down the line. You can go through all of your designs and quickly identify everyone's design in the whole company who has that topology and maybe fix it, you know, or address it, or, you know, on a layout, you have a physical structure. Maybe we think about reliability there, you know,

there's a trace that's too thin and there's too much current going through the trace and it's blowing up parts, you know, when we put them in the field. Okay, well, now if you have a way to access the information, find out, give me all the layouts that have this particular structure in it, in the whole company, you know, out of our 100 million designs or whatever we have, find all of them that have this, you can do that instantaneously.

And that's very, very powerful, even without AI, you know, forget the implication of AI without doing that is even powerful. I think the first step is really getting your information into your design process and that's where automation comes in. What are the key challenges and opportunities presented by the transition to 6G technologies? I mean, that's coming. Absolutely. Well, if you look at the requirements, at least from the early 6G requirements and you look at that from a technical perspective, I can

assure you there'll be no shortage of challenges with 6G, so there's plenty of challenges. Given that I look at circuits, RF microwave circuits, right, those RF and microwave, those both refer to frequency, so I guess I'm obligated to say that the biggest challenge will be frequency, at least for the customer set that we work with. And what do I mean by that? So right now we're sitting at millimeter wave frequencies for 5G, for some of the cutting edge 5G designs, right?

And if you look at the wavelengths there, that's kind of in the centimeter range. So it's a little bigger than the chip, for the most part, in the wavelength. When we go up to what we're looking at, hundreds of gigahertz higher, maybe 250 gigahertz, that's like sub-terrahertz, even maybe into the terahertz range of frequency, what happens is the wavelength of the signal starts to go to millimeters. And that's significant because millimeters start to become like chips are on that millimeter scale, right? So now you think about like bond pads,

bond pads start to become patch antennas there because for antennas, you have to have antennas that are on the order of wavelengths. So, you know, a half a wavelength, a quarter wavelength, these structures will resonate and transmit signals, essentially, into the atmosphere. And so that becomes very challenging because now on a chip, instead of just,

we kind of think about, okay, on a chip design, yeah, we have some electromagnetic effects. But now think about the localized transmission and transmit and receive paths happening, like, on the chip. effects. But now think about the localized transmission and transmitted received paths happening like on the chip. So if we have a filter in between two pieces, the signal can just radiate right out, jump over the filter, and go to the output. So those are the kinds of challenges that I think are going to make it difficult to design these types of circuits. I think that what does

that mean? Well, multi-physics analysis, you know, electromagnetic simulation, that's, there's no way to do a design at those frequencies without having that, you know, an electromagnetic simulation. You could have thermal effects, other simulations. I think the other implication of going up to the high frequencies, I mean, right now, if you look at millimeter wave, 5G, a lot of it's silicon based. It makes sense. But if you look at the packaging technologies,

kind of comes from processors and processing world. So like the heterogeneous integration packaging you're seeing at 5G types of designs, a lot of that comes from processors. But when we get, you know, that stuff, that's not gonna get us over a hundred gigahertz here. I mean, we start getting over a hundred gigahertz. Now we're looking at scenarios

where there's probably more three, five technologies in there and there's harder integrations that take place. And so, you know, I think, I do think, cause we were talking about AI before, you have this interesting case here, right, setting up in the next couple years. On one side, you've got 6G. And so that's that's just really like engineering challenges galore over here. And then the other side, you have potential solutions with AI. And so I think when those, these are like,

you can kind of think of them like two big stars and they're about to collide. When those two things collide, I think you're gonna have like an explosion in hardware functionality is gonna come out of that. So what I mean, the types of hardware that we're gonna be looking at, maybe 10 years, five, 10 years from now,

it's gonna make the most complicated stuff today look simple. I mean, it's going to be, there may be circuits that we don't even understand what, you know, from just from looking at them, what they do, you know. So that's going to be such an interesting space. You have a problem, you have a potential solution. I think it's just, there's going to be so much innovation that happens in that overlap between 6G and AI. Yeah, I agree. So let's talk about packaging. What role do chiplets and advanced packaging technologies play

in next generation designs? Yeah. Well, I mean, in the context of 6G and from the customers that I talked to and even internally, there is some consensus, I think, shaping up that, hey, chiplets may be the only way to get the swap C requirements you need for 6G or whatever, some of these higher frequencies, more advanced standards that we're running into.

So swap C is, let me see if I can remember this, size, weight, power, and cost. And then people throw in performance, too. So I think it's like there might be a second P in the swap C acronym there. So I think that, so we already have chiplets and we're probably, you know, that the listeners are probably pretty familiar

with the idea of chiplets. But when we're talking about going up to these like 6G frequencies, we have to think about chiplets a little differently. And I think the thing that's gonna be different there is potentially the entry of these more exotic technologies. So things like 3.5s and the like, having those technologies now start to come into play, I don't think silicon is going to totally go away there, but I do think these technologies

will start to come into play. And now we're going to have this dynamic between the challenges of 3.5 design and the silicon pieces there. So, I think what's going to happen, at least from an RF microwave high-frequency design space, so we're moving to this world of interchangeable technology. But think about the software tools. If you look at the software tools that are out there, those things really are technology specific.

What I mean by that is like, if you're designing a silicon chip, you use one piece of software to like make that thing, you know, if you're designing a gallium arsenide chip, maybe you make use a different piece of software to say, you know, if you're designing a package, there's a third piece of software, you know, et cetera. So each technology, we've evolved to the point

where each technology kind of has its own software tool. And it's not even just the tool itself. The philosophy of how to design those things is completely different. Like I used to, one of the things I did when I was doing hardware development, I used to try to teach, because we did packaging and chips. So I would try to teach people who were trained

like out of college how to use a packaging tool. Packaging tools like routing, you know, you're doing like avoidance routing and all these fancy things. And it was amazing. It was so hard to do. If you're coming out of that chip world, everything's a polygon. Now you go to the package world and you're drawing traces and stuff. It's just a completely different philosophy. So it takes people a long time to pick up on those sorts of things. So what I'm getting at here is I think that the whole process of software tools is going to have to change as we're evolving through here because technology becomes interchangeable with chiplets.

So now before it's silicon chiplets, but now imagine we've got three fives. Well now the technology, now you do what-if analysis. Okay, what if I use, you know, this particular gallium arsenide? What if I use p-hemp? What if I use silicon? And you've just got to be able to swap these pieces in and do, you know, kind of what-if analysis there. That means that to me, that there has to be a reworking of the tools around the design process, which is a pretty radical change from where we've been here instead of technology, used to be in technology.

And so there's kind of two things that come out of that. The first is we're potentially gonna have a lot of tools. If we just try to blend what we've got today and make that work, if we diagram out the design process and then say, okay, let's throw tools in there. The first thing is might end up with like 20 or 30 tools. Yikes, that's very, very difficult.

And the second challenge is just the learning curve and the inefficiency that happens there. So if we're in the design process, if we've turned interchange a tool that does this one thing with a tool that does it a different way, now you need designers to understand all that stuff. So those things are difficult. Again, I think going back to the automation, I think the automation is really key here because we want to have tools which are automatable because I think companies, back to the localized idea, companies are going to rise with these design processes and AI technologies, at least initially

in the RF microwave space. So we need to have automation to enable that. We also now need to have tools that go across these technology silos. And actually, like you, we need a tool that you can use for an IC layout and a package at the same time. Because just from an efficiency perspective, we need less tools. And the other thing I think that starts to come, so there's going to be a shift from technology-specific tools to tools which work over the entire design process.

And I think in that case then also too, it seems likely that there's probably a measurement angle here too. You know, this starts, When we do that, this starts to look like test and measurement a little bit, which Keysight, we're very familiar with. Test and measurement, obviously, but it's these. Think about it. Now you start to get into these. If you want to do a test, you've got your test equipment, your test equipment that has specific functionality, and you're putting it together and you're automating and orchestrating this test equipment to give you a result. It starts to become very similar, I think, as the software evolves into this paradigm. And so I think it's just more efficient then to blend design and test.

I think that's going to happen as we move to 6G and higher frequency. So I think that is really, there's going to be some interesting years ahead here as we evolve into these new next generation designs. Yeah, that's for sure. So final question and kind of a summary, what do you think the next generation software tools will need to do to enhance productivity and innovation in high frequency hardware development? Yeah, well the first thing and I kind of got at this in the last question as well, but flexibility is going to be really, really important for these tools. You know, if we're going to build tools around design flow, you can't just have, you know,

a tool that does one thing. You need tools that can be flexible and that are obviously automatable. What do we see coming down the pipe for the next generation, you know, hardware development? Well, one interesting thing, especially when you get into this blending of AI, is optimization becomes very, very interesting, both in terms of like, you can parallelize this, and also there's some really interesting AI tools

that are evolving around optimization. So Keysight, I mean, we're working on some interesting optimization technologies internally, but again, if we're going to, you know, if we have a customer base here, maybe it's our customer, some of our customers who develop optimization technologies that are very interesting with AI. So we want our tools to be able to allow those algorithms to come in, wrap around our simulator, wrap around our data analysis, right?

And to be able to use the simulator and drive it instead of the old way, which is you have an optimizer in the simulator, you click the optimization button in your little UI, and the simulator does something, the tool does something in the background. So it's going to be a more active, I think, role in optimization. That's what you're going to need to do AI work, right? AI training and also plugging in the optimizers. Also, I think spanning technology boundaries, I think that just becomes necessary going forward.

If you wanna have productivity, or you need to be able to use the same tool to do different things. And that's one example at Keysight. I mean, there's a lot of different electromagnetic solvers, but one of the things we looked at when we were developing our RFPro EM tool was, hey, could we use the same electromagnetic solver on a chip? And then we also – so we make chips, right, in packages.

Could we use it on a chip? Could we use the same solver on a package? Could we use the same solver on a huge brick board that we put in instrumentation? And so a lot of our development work there went into making something flexible that could actually give you good results for all three of those things. Because if you can use one solver across all of these paradigms, you don't have

to worry about the inconsistencies that happen. And that's the really big thing. Like if you take a chip and you simulate it in three different EM tools, you can get three different answers. Now, when you're spanning technology, you're moving across technology, that becomes really problematic

because now you have two things that are changing. Like when I was doing PA design, we used to call it Mickey Mouse analysis or whatever, Sesame Street analysis, I'm sorry. So it was like one of these things is not like the other. Basically what that means is if you're doing an experiment and you change more than one thing, you don't know what the cause of the change,

the cause of the change in your results was. And that same problem applies here. If you have multiple engines, that alone will change your results. So then when you're trying to blend everything together, you have two things changing. You have your technology is changing, but also your results based on the engine is changing,

and that's problematic. Same thing in the circuit simulation environment too. We have a new tool that we're rolling out shortly. It's called Nexus. It's a simulation tool, and it spans across platforms. It's all about automation, being AI ready, but also going across platforms. We're making this so it doesn't just work in our, we have a Keysight ADS platform.

One of the things that's great about Nexus, you don't need to be in the ADS platform. If you're over in, for example, the Virtuoso environment and you're doing silicon design, you can use Nexus in that environment. Synopsys custom compiler, same thing, you can use the Nexus solver in that environment. So I think that's really gonna be the important thing. Something, tools that go across those boundaries that you can use, that gives you the consistency and the model to be able to develop

these next generation products. Great conversation, Matthew. Thanks for your time and hope to have you on again later on. Thanks so much. This was fun. It's great to be here. That concludes our podcast. Thank you all for listening and have a great day.