From hype to engineering: How AI actually works in 3D IC design

Tova Levy (00:08.322): Welcome back to the 3D IC podcast where we explore the cutting edge technologies, challenges and innovations shaping the future of semiconductor design. I’m your host Tova Levy. And this season, we’re taking a deep dive into every corner of the 3D IC ecosystem from design and manufacturing to packaging, testing, and everything in between. Today, we’re tackling one of the most talked about topics in our industry and everywhere else, artificial intelligence. You’ve probably heard the hype, AI is going to revolutionize everything, automate all our workflows, maybe even replace engineers. But what’s the reality? What can AI actually do in 3D IC design today? And more importantly, what are the challenges that nobody’s talking about? Joining me today is Sudarshan Deo, software engineering manager for 3D IC at Siemens, who’s been working hands on with AI implementation in 3D IC workflows. We’re going to get practical, get real, and maybe bust a few myths along the way. So whether you’re a seasoned design engineer, a newcomer to the 3D IC space, or just curious about where AI and semiconductors intersect, this conversation is for you. Let’s dive in. Sudarshan, welcome to the 3D IC podcast.

Sudarshan Deo (Siemens) (01:25.320): Thank you, Tova. How are you?

Tova (01:28.322): I’m good. So can you start with introducing yourself and telling us about your role at Siemens?

Sudarshan Deo (Siemens) (01:35.432): Sure, so I work, as you said, I’m working as a software engineering manager with the Central Engineering Solutions Division within Siemens. And I manage the software R&D team across different geos, Cairo, Yerevan, Armenia, and then we have a team in the US as well. And basically the team is driving all of the strategic advancements in advanced packaging, right from packaging to IC to DFT, and then all of the sign-off analysis, multi-physics flows that we have, stress, thermal, as well analysis flows. My team is building the software infrastructure and the workflows and then integrating this into the Siemens 3D IC ecosystem right now. And on top of that, we are heavily involved in working on agentic AI or design space exploration optimization, or even the machine learning terminologies or stuff within the 3D IC workflows right now. So I think that’s more or less my role at Siemens right now.

Tova (02:42.094): All right, so before we go into the AI aspect of 3D IC design, what are the practical challenges that companies face when starting with 3D IC design that people don’t typically think about?

Sudarshan Deo (Siemens) (02:58.718): I think the biggest misconception that we have in 3D IC right now is people assume that 3D IC is just like a packaging problem. So basically we stack chips and then we are done. But I think in reality, it fundamentally changes how you think the entire design process is. So one of the practical challenges I feel the companies face is, I think surprisingly it’s simple, where to begin. So in traditional ICs, in monolithic ICs, teams already have a mature workflow. But with 3D ICs, suddenly there are too many questions that are raised. For example, if we have chiplets, how do we partition the chiplets? Which packaging approach kind of makes more sense? When should thermal happen? At which stage during your design you will introduce thermal? Which tools or workflows should we trust? Which infrastructure makes the best sense for us? So I think these are some typical challenges. And there is no universal playbook for this yet. So this becomes itself like a challenge. So I think that’s one thing that I think companies face a lot of challenges. The second thing is the amount of organizational complexities that every company has. So in 3D IC, you are bringing together multiple engineering worlds together. So you have different teams like silicon teams, package teams, you have mechanical teams who are working on, let’s say, thermal or mechanical stress. You have analysis teams, you have verification teams. And I think historically, these teams are often operating in silos. They are operating in kind of a vacuum. So for me, I’ve spoken this in different forums. This is kind of throwing over the wall problem, right? So you are kind of finishing your job and then throwing the stuff onto the next team. And for you, think you are kind of eliminating the problems, but those problems might be discovered at a later stage in the design as well.

Sudarshan Deo (Siemens) (05:01.848): But in 3D IC what happens is I think this problem is like super expensive, right? Because all of the flows are very much interconnected. Like for example, if you have a packaging or floor plan decision that you have made initially that can improve the performance of your, let’s say power or area, right? But later in your design cycle that might introduce, let’s say, thermal hotspots or let’s say mechanical stress or that can even have, let’s say, power delivery challenges, for example. So I think these are things that you should try to identify at an early stage during your design cycle. And I think that that’s one of the challenges that most of the companies have. So I think in today’s world, I feel the organizations that think about this problem holistically, build a framework around it, and have the different teams kind of collaborate together with each other on an engineering domain. I think those teams or those organizations have a better way forward right now. So I think those are the practical challenges. But if you overcome those, think you have a very good path forward.

Tova (06:11.310): So as you said, the challenge is not just the technical complexity, but the organizational complexity. Right. Okay. So now that we have this clear idea of the difficulties that companies are facing, I mean, obviously they would want to use AI to overcome them, but so let me ask you this. So what are the things about using AI in IC design that people don’t consider or maybe don’t understand fully?

Sudarshan Deo (Siemens) (06:40.882): Definitely. I think during your opening statement, you also mentioned that, you know, is AI kind of replacing humans as well. I think we will speak about that, but that was an interesting statement. So I think the biggest misconception here is, you know, people have different terminologies, right? Either they overestimate AI or they like kind of, you know, completely underestimate it, right? So there is no middle ground to it. So on one side, you know, people imagine AI like, you know, it will automatically design chips like to an edge, it’s going to be like a magic box, right? And on the other side, I’ve seen people who think like, okay, AI is like a hype, it’s like a wave, but it’s kind of a buzzword and whatnot. But I think reality is much more practical, right? And honestly, it’s kind of very interesting as well. So I think first, people often assume all AI works is like a general purpose, like a chatbot, right? But according to me, at least, it’s fundamentally different. So it’s like an engineering AI model. So in semiconductor design, AI has to be very, very domain specific. So it cannot be very much broadly classified. So it has to be very much physics-available. It has to be explainable, what it does. And it has to operate within very strict security and the IP boundaries as well, because there is a lot of data flowing in and out right now. So it’s not just about generating the answers that you have. It’s also about generating trustworthy engineering outcomes. So it should be logically correct as well. So I think I give this analogy like multiple times, right? So let’s say for example, if you ask AI, right? So just in a chatbot, you say that, you know, I have a car wash, you know, which is like, you know, 500 meters away, right? Now from my home. So what’s the best way to go there, right? And the fast response from an AI will be like, you know, take the shortest route from, you know, XYZ directions, and then you can walk from there. So I think logically that, you know, sounds correct at first, right? But I think it’s actually missing the intent, right? That you are actually taking your car to be washed. You know, walking doesn’t make any sense, right? So, so the moment you provide like better context to AI, I think your answers, you know, will improve, you know, dramatically, right? So I think that that’s, that’s one of the key things. And in semiconductor design, especially in 3D IC, I think the context is very, very similar, right? So if we ask AI, let’s say, you know, to optimize only one of the matrices, right? Let’s say, you know, either power performance or let’s say area, right? And then we don’t give, you know, enough engineering context, right? Or like constraints or like know enough objectives right so you will have a mathematically correct answer but a practically wrong answer right so that’s that’s one of the key things that you know people shouldn’t understand.

Tova (09:23.118): Right. So you’re talking about defining limitations. Yeah.

Sudarshan Deo (Siemens) (09:26.686): Exactly, exactly, exactly. And I think, think the other thing is, know, AI models might recommend like, you know, an aggressive, let’s say floor plan or like a, know, TSV strategy approach or like, let’s say, you know, chiplet partitioning approach, right. But if, know, there are thermal constraints or let’s say, you know, there is reliability or, power delivery that are not considered, I think on paper, it’s going to be okay. But, you know, in 3D IC, it’s going to be very, very important to, you know, have all of these analyzed at an early stage because cost is one of the biggest things, right? So the later you identify these problems in the design, it’s going to involve a lot of cost, right? So that’s the key thing. I think the other things would be to understand the engineering data as well, right? So, you know. I think in semiconductor at least you have a lot of data. So if you look at any given chip that a design company manufactures. So none of the design companies manufacture chips right from scratch. So it’s like a derivative of the chips that you have. So every new generation is built on historical design knowledge that you have from the past. So you have different constraints that you identified. You have different failure mechanisms that you have had in the past. Also had the best practices that engineers have followed in the long history. So I think AI has a lot of data from which it can learn and if you start asking the right questions, it should be able to give you the right outcomes as well.

Tova (10:55.630): Yeah. So you’re talking about like, you need to know how to ask and when to ask and about what, and also when to question that. Right. Yeah. And that’s, that’s quite different from the AI will do everything for us, that we, we often hear.

Sudarshan Deo (Siemens) (11:15.932): Yeah, it’s not a magic box. I think I would like to touch base on that comment as well, which started our discussion that is it like a replacement argument or like an augmentation. So I don’t think that’s really the case, right? Because there is always a need for a human in the loop, right? There is always the need, right? Because even if you have AI make a smarter decision for you, which is fine, right? So it’s going to make faster decisions, but not each time it can make a smarter decision. I think definitely we should leverage AI for making faster decisions. Let’s say if you want to use design space exploration, right? So let’s say it’s gonna produce, let’s say 1,000 results in one minute, which 10 engineers might find a problem. It’s like reducing the number of engineers, but it’s not replacing like a human, right? It’s not replacing the human brain.

Tova (12:23.206): Yeah. So when AI does work well within the 3D IC design flow, so how is it currently helping? Like what can it actually do today to help engineers?

Sudarshan Deo (Siemens) (12:53.540): Right. I think that’s where I was kind of diving in. I think one of the biggest areas I see, at least the areas where I’m working on, is design space exploration. That’s where our previous discussion was happening. So in 3D IC, the number of decisions kind of explode. So you have chip partitioning, you have your floor planning, you are stacking the dice on top of each other, you are connecting them through TSVs right now you have you know hybrid bonding approaches where you know you’re connecting different you know chips together now you have thermal constraints you know once that happens you know you have power deliveries you have manufacturability you have cost you have reliability you know there are too many things that you know influence each other right. Now traditionally, engineers explore only a subset of all of these possibilities because there is a crunch in time, there is a cost involved, there is a human labor involved. So that’s where AI can help narrow these search spaces faster. So think of it like almost like having a highly experienced copilot engineer kind of helping your engineers prioritize which options are worth exploring deeply. And at the same time, it can also eliminate which options are like either dead on arrival or they are not useful for a given point. For example, imagine you’re designing, let’s say, an AI accelerator with, let’s say, certain HBM memories, right? So you could optimize, let’s say, aggressively for performance and bandwidth. But again, thermal becomes a problem. So what are the different parameters that you should consider, not only for cost, but also for, let’s say, manufacturing changes, for example. So I think AI can help engineers evaluate all of these trade-offs much faster, instead of relying, let’s say, on human or manual error. So another place I think AI is helping today is, again, as we spoke, like historical data, right? So it’s kind of learning from previous data and then kind of orchestrating the data and then kind of giving you different simulation models. So I think those are the key things that distincts AI from what you are working on today. And I think 3D IC is kind of, it’s a very specialized knowledge field right now. So one other thing is, you can use AI also to train your new engineers, right? So whatever best known practices or best known methods your high skilled engineers have followed in the past. So whenever a new engineer kind of joins your team, you can always use that surrogate model to actually give to that engineer so that they can learn on the fly. So I think that that’s one of the best used models I see.

Tova (15:14.798): I’ve heard from some of the other guests on our podcast that one of the most challenging things is that in 3D IC, you need to get a lot of people to understand a lot of domains so that they can work together. And so this is one of the things where, you know, it really can give that head start in understanding different domains. Yeah.

Sudarshan Deo (Siemens) (15:30.654): Exactly. Yep, absolutely.

Tova (15:36.297): So what you just said is really highlighting that it’s not about reinventing the wheel, but building intelligently on what came before. And so if we’re seeing this kind of thing in the design phase, I’m curious also about the bigger picture, like beyond the design phase, where else do you see the most significant impact of AI in the broader 3D IC ecosystem?

Sudarshan Deo (Siemens) (16:11.230): This is actually one of the areas I personally find most exciting because it goes far beyond the chip designing phase that you have. So I think people sometimes kind of underestimate how much complexities exist across the entire semiconductor, I would say, lifecycle management. And I think, especially in 3D IC, one major opportunity is materials and then optimization of the process itself. So again, in advanced packaging, let’s say hybrid bonding or even heterogeneous integrations, very small process change can significantly impact your thermal performance, your stress behavior, your manufacturability, or even the reliability that you have. So historically finding the right process window often involved a lot of expense, a lot of experimentation and too many engineering cycles. So AI can definitely help narrow these possibilities much faster and identify promising materials or good process combinations earlier in your design cycle.

Sudarshan Deo (Siemens) (17:22.150): I think another thing is your yield that you have and the manufacturing optimization as well that you have. So modern fabrications or advanced packaging facilities that you have, again, they generate a lot of data, enormous amount of data that you have. So AI can help, again, identify anomalies that you have. It can detect process drifts that you have, improve, let’s say, defect predictions, or it can also help to reduce the yield cost or the yield losses that you have. So that is one of the key impacts. So in this case, let’s say even the numbers are pretty small. Let’s say the numbers are like, let’s say 2 % or let’s say 5 % improvement in the scale. But in 3D IC that can result in let’s say millions of dollars. So even if the numbers look pretty small, they have a massive business impact. So that’s one of the huge opportunities that you see in 3D IC. So I think again, the other thing is, how do you predict the reliability of your entire chip design? So I think in 3D IC reliability becomes more challenging because you have a lot of thermal cycles. Like you have too many dense interconnects. You have your interconnects that are producing a lot of stress. Because imagine you’re stacking dies on top of each other and then each die has its own stress level. So they are operating at different stress constraints. So it’s producing a massive amount of stress uncertainty. You have power delivery concerns. You have too many heterogeneous integrations. So I think AI can help identify these patterns that humans might miss and then predict long-term failures that you have much earlier in your design cycle. So that’s the other thing. And I think the last thing I want to touch upon is I think that’s the direction in which the entire semiconductor world is kind of leading is like physical AI. So today we think that AI is kind of helping us design systems. But tomorrow it may help us continuously optimize themselves even after we deploy, let’s say the systems. For example, imagine hyperscale design centers dynamically, let’s say they’re optimizing, let’s say thermal and then power behavior. And then there are autonomous systems adapting to these workflow conditions and then continuously improving the efficiency of these systems. AI won’t just stop at the chip creation at that point in time. I think over the time it will extend into how semiconductor systems are manufactured, how they are deployed, how they are monitored and then continuously optimized in a real world. I think that’s where the world is moving on right now.

Tova (20:15.566): Right. And I have to say, I love that some things within the IC world, they just touch on science fiction for me. Like this is what you just said sounds like science fiction. And I love that.

Sudarshan Deo (Siemens) (20:27.806): True. It’s kind of, well, yeah, I think five years ago, if you would have imagined, you know, Cursor or ChatGPT, or, you know, even all of these, you know, skill frameworks that have come in, you know, that was really a science fiction, but right now, you know, it’s actually real. So in the next five years, yeah, exactly. Yeah. So I think next five years, physically, AI is going to be the boom in the market.

Tova (20:42.314): It’s crazy. Yeah, it’s like science fiction is meeting our reality right now. Yeah, so. All right, so let’s talk about the future. What do you see on the horizon that has you most excited, let’s say, within the next three to five years?

Sudarshan Deo (Siemens) (21:05.096): So, personally what excites me the most is, you know, I think we are reaching like an inflection point right now. So for years, semiconductor complexity has been growing very steadily. But with 3D IC, chiplets, all of these advanced packaging, heterogeneous integrations, I think the complexity curve is kind of exponentially growing right now. And honestly, I think we have reached a stage where AI is becoming less of a, I would say, nice to have, to more of a necessity right now. So the number of engineering trade-offs that we have is simply becoming too large right now for traditional workflows just alone. So I think what excites me is the idea of moving towards, I would say, AI native engineering workflows right now. So today, engineers often optimize different domains somehow, let’s say, independently. So you have timing that is optimized differently. You have thermal optimized differently. You manufacturability optimized differently. But tomorrow I think we’ll move towards much more like intelligent multi-physics co-optimization, where AI helps engineers reason across these domains, let’s say simultaneously. For example, imagine an engineer who is exploring a 3D AI, let’s say accelerator design, for example.

Sudarshan Deo (Siemens) (22:34.718): So, instead of manually iterating through dozens or hundreds of combinations, I think there is going to be a proactively suggesting approach. So if you move, let’s say for example, if you move the chiplets to this and this location, or if you adjust your TSV densities to these and these values, or let’s say you slightly modify your package architectures and then change the cooling rate, how it’s going to impact your, let’s say, thermal, or how it’s going to impact your cost. I think that’s going to be incredibly powerful where AI comes into the picture. And I think second thing is, we discussed about this, that it’s becoming like an engineering copilot, right? So you have this agentic AI workflows, which kind of help you not just create the workflow infrastructure, but they also help you collect the data, launch the simulations, move across different tools, let’s say, validate what setup you have. Then once you have the results, it’s also going to help you interpret the results, back-propagate the results and then so on. So it’s almost like a highly specialized engineering assistance that you have that reduces the friction within your flows. And then it also allows engineer to focus more on their day-to-day lives, like innovation, for example, thinking about the next chip that you have, thinking about some other curious aspects to things. So it’s going to reduce a lot of human effort in the future.

Tova (24:03.672): So it’s interesting because you’re saying that like we’re kind of at an inflection point where the technology 3D IC is new enough that AI can really shape how it develops. And instead of, you know, what we’re seeing maybe in other areas where you have set workflows and then you go back and you say, okay, how can AI help me with this?

Sudarshan Deo (Siemens) (24:33.324): Exactly. I think we have been hearing this from several tech execs as well, right? That, you know, not many people are now interested in working or like knowing how exactly, you know, let’s say your tool works or, you know, your infrastructure works, right? But they are more interested or their designers are more interested in, you know, knowing how the tools communicate with each other, how we can build a strategic workflow, right? So it’s more of a, you know, horizontal than a vertical integration right now. So yeah, I think that’s the key thing, especially within 3D IC.

Tova (25:22.613): So there’s always considerations beyond just can we do something. There is also the should we. And so in this context, what are the ethical considerations or potential pitfalls that we should be mindful of when we integrate AI more deeply into IC design?

Sudarshan Deo (Siemens) (25:46.462): Yeah, I think that’s a very, very good question. And it’s an important question as well, right? You know, there is a tendency to talk about, AI can only, you know, do a lot of things. And then, you know, there is a lot of excitement, but, you know, I think in semiconductor engineering or anywhere else, I would say the stakes are very high, right? If you ignore the limitations within AI, right? So the first foremost thing is, you know, trust that you have and the explainability that you have for your AI systems, right? So let’s say if again, you know, going back to that analogy that I made, right? So about, you know, the car. So if an AI system recommends a design optimization, for example, right? So engineers do need confidence in why that recommendation was made in the first place, right? So let’s say in consumer applications, If AI gives, let’s say, a slightly wrong answer on an application level, maybe it’s kind of annoying and then you’re gonna go back and then say, hey, this is not correct. And then give me like a reverse decision. But in semiconductor design especially, right? Small mistakes can translate into a lot of silicon respins right it will involve lot of schedule delays and you know millions of dollars are at stake right now so you know it’s it’s you know it’s pretty high value so explainability here you know matters enormously right now. Another concern or thing that we should consider is data quality and the governance of the data as well. Because here, AI is only as good as the data on which it’s getting fed on. And in semiconductor organizations, they often have, let’s say, fragmented data that is spread across tools, that is spread across, let’s say, business units or teams or product that are generating those data as well. So if the training of this data is, let’s say, incomplete or it’s inconsistent, or if the data has a lot of bias. So in that case, the outcome becomes highly unreliable. And I also think, you know, companies should think very carefully about, you know, the security of your data and then the IP protection as well. Right. Because again, right in semiconductor design, you know, it involves some, you know, intellectual properties. Let’s say, you know, you’re working with, you know, different foundries, for example. Right. So in that case, you know, you are not allowed to share one foundry data or the other. Right. So it involves a lot of, you know, IP protection. Right. So in this case, you know, organizations need confidence, you know, around, you know, how the data is stored, where the data is stored, how the models for the data are getting trained, what information is exposed, what is not exposed. So all of these play a very key thing. And I think another important limitation people underestimate is the optimization bias that you have. For example, AI tends to optimize for the best objectives that you define. But let’s say what, you know, if those objectives are, let’s say incomplete, you know, for example, again, you know, if, you say that, okay, I want to optimize my power performance and area, and then, you know, you’re not focusing on, you know, other factors, let’s say thermal, right? For example, or, you know, even, you know, even the yield that you have, right? So in that case, again, you know, you’ll have a far bigger, you know, logical challenge in the, the future, right? So, so I think, I think those are the key things that, that you should, you know, really be aware of. So it’s again, it’s not like a magic ball. So whatever you throw in, it’s like a garbage in, garbage out approach. So whatever you throw in is what you’ll get. So the better you throw in, the better you’re going to get. So that is as simple as that.

Tova (30:26.190): So what you said about data sharing and security, that’s particularly important because it touches on something bigger and that is the semiconductor industry doesn’t exist in a vacuum. It’s a global ecosystem with complex geopolitical aspects. And how does AI fit into that international picture?

Sudarshan Deo (Siemens) (30:52.201): Yeah, I think this is a fair question to ask because I think here the impact goes like, you know, let’s say well beyond technology, right? So, you know, AI has, you know, potentially, you know, it basically has the potential to, I would say, fundamentally change, you know, how semiconductor expertise is like, you know, distributed, let’s say globally, right? Because historically, you know, I think deep semiconductor knowledge has taken years and years, right? And sometimes even decades to develop, right? So even today, a lot of expertise, they live in highly specialized teams, let’s say, let’s say highly specialized institutions where they have specific knowledge. So here, definitely AI can help democratize or even help easy access to all of these, let’s say, expertized data. Again, I think if you have, let’s say a recent college grad or like a very young engineer who is working on, let’s say, advanced packaging or 3D IC. So in that case, AI is definitely going to help them to ramp up significantly faster. Again, by learning historical data or historical workflows or best engineering practices. If you give that as a manual training, might take like years for that particular engineer to learn or follow the best practices. And again, that doesn’t replace the actual experience that you have, but it always accelerates your learning process. And I also think that AI will also reshape how engineering teams collaborate, let’s say globally. I think I can give that as a personal example since I have several global teams I’m managing right now. So I think semiconductor development is increasingly distributed across different geos, different foundries, different ecosystems for packaging, and even different design teams. So it always has the potential to become like a common layer that helps teams work more consistently with each other, communicate with each other with a better design intent and also reduce the friction that you have across different workflows. So I think it’s becoming more of a strategic differentiator right now. So I think even today, companies that learn how to combine, let’s say, domain expertise, they exactly know how engineering data operates. Companies that know exactly how to effectively use AI. I think those are the companies which will always innovate faster, it will always move at a better speed and it will always have a very strategic advantage with other companies. So I think that that’s one of the key things that is globally changing across the world right now.

Tova (33:56.367): Yeah. So, okay, we’ve talked a lot about the industry landscape. Let’s bring it a little closer to home. What is Siemens specifically doing to address these challenges that we’ve spoken about?

Sudarshan Deo (Siemens) (34:11.518): Sure, yeah, I’m happy to say that. So I think what’s exciting about Siemens in particular is that we are approaching these challenges or the challenges that we spoke about from a very, let’s say workflow-first and a system-level perspective as well. So your 3D IC is inherently, it’s kind of a multi-domain problem. So it’s kind of inherently problems that are coming from several different domains. And then you are dealing with, let’s say, electrical behavior. You’re dealing with reliability problems. All of the trade-offs that we just spoke about, which are interacting with each other again. And one thing we have learned so far is solving all of these problems in an isolated world becomes increasingly difficult. And that’s where Siemens is playing like a differentiating game. So we are not just thinking about AI as just another feature to add on a single tool, let’s say. Let’s say if you add AI on a tool on a vertical scale, so that’s not the intent. So we are thinking about how AI can help orchestrate the entire engineering workflow. So for example, one area that Siemens is actually very heavily investing is in AI driven design space exploration and then optimization of the data as well. So the number of possibilities that you have in, again, in 3D IC is enormous. We are using technologies like HEEDS. We have one of our design space exploration and then optimization platform that can intelligently explore and then optimize different designs, spaces much faster. It can help engineers identify stronger trade-offs between, let’s say, PPA, thermal, then reliability, and then cost as well.

Sudarshan Deo (Siemens) (36:02.406): And then the other thing is, you know, that finds, you know, at least to me, it’s particularly interesting is, you know, the workflow orchestration, right? Through AI and then, you know, the agent AI systems that we have, right? So, you know, again, the historical data that you have, right? How can you use the data, you know, to the best extent, right? So we have our flagship, you know, Fuse AI, you know, platforms that are used for, you know, creating, you know, engineering co-pilots and then, you know, domain specific, you know, assistants and then, you know, chatbots that you have, Where, you know, engineers can you know, interact with all of the workflows, the 3D IC workflows that you have built. You know, they can use, you know, English, Mandarin, or, you know, any other natural language, right? To, you know, query that historical information. And it’s not only about, you know, giving like a prompt, but it’s also creating a workflow, you know, using, you know, let’s say skills or, you know, any other, you know, framework within AI that can, you know, help, you know, define how your infrastructure is, you know, it can help verify how your infrastructure is. But on top of that, it’s also going to launch the simulation, you know, look at the results, you know, analyze the results and then it’s going to be like a phased approach where you have, you know, Fuse coming into one, you know, at one point in time coming into the picture, you know, generating all of these, you know, data and then you have, you know, design space, you know, exploration and optimization as well. Right. So I think it’s going to scale, scale, you know, dramatically. So that that’s where, you know, Siemens is, you know, making a difference. And again, right. If you have seen, you know, Siemens, it’s, it’s, it’s not just, you know, EDA right so it spans you know beyond EDA right so it’s it has its you know footprints in simulation it has you know footprints in you know PLM you know life cycle management in industrial software in manufacturing right so you know I think you know at Siemens you know one of the biggest you know things that that we are focusing right now is you know making sure that you know you have a silicon-to-system perspective right so it’s the packaging and then package out model as well for 3D IC and that that’s where you know most of the fabrication companies or most of the foundries, most of those sites or even the design companies are looking towards Siemens because I think that’s the unique feature that Siemens has that it’s not only engaged within EDA but it’s kind of bridging the gap between EDA and then hyperscale or systems as well. Yeah.

Tova (38:22.638): All right. All right, so we’re going to wrap up with that. Thank you so much for being on our podcast today.

Sudarshan Deo (Siemens) (38:28.798): Sure. Yep. Yeah, it was nice talking to you and then thank you for the opportunity. It was really nice. Thank you so much.

Tova (38:35.650): Thank you for sharing these valuable insights. I am sure whoever listened to this is going to keep all of those things in mind next time they use any sort of AI. The one thing that I’m taking away from this is that AI is not about replacing human expertise, as you said it. It’s about augmenting it and accelerating it and helping us to navigate these complex workflows. But it requires the right frameworks, the right questions, and the right human oversight.

Sudarshan Deo (Siemens) (39:13.810): Of course, exactly. Yep, yep, yep.

Tova (39:19.054): All right, so thank you all for listening. And before we go, a quick reminder, this is part of our ongoing series exploring the 3D IC ecosystem. So make sure you subscribe wherever you get your podcasts so you don’t miss an episode. And until then, stay curious and we’ll see you in the next one.

Sudarshan Deo (Siemens) (39:37.736): Thank you.