Semiconductor Insiders - Podcast EP354: How Siemens EDA is Conquering New Lithography Challenges with Sagar Saxena

Episode Date: July 10, 2026

Daniel is joined by Sagar Saxena, Senior Product Engineer at Siemens EDA specializing in computational lithography, optical proximity correction (OPC), and advanced patterning solutions for leading-edge semiconductor manufacturing. Sagar has led the development and deployment of advanced OPC technologies, including… Read More

Hello, my name is Daniel Nenny, founder of Semaywiki, the Open Forum for Semiconductor professionals. Welcome to the Semiconductor Insiders podcast series. My guest today is Sagar Faxina, senior product engineer at Siemens EDA, specializing in computational lithography, optical proximity correction, and advanced patterning solutions for leading-edge semiconductor manufacturing. Cigar has led the development and deployment of advanced OPC technologies, including curvilinear OPC methodologies, computational lithography workflows, and next generation mass synthesis

solutions. His work has contributed to high-volume manufacturing enablement at advanced nodes for some of the world's most advanced semiconductor processes. Welcome to the podcast, Sigar. Thank you, Daniel. It's good to be here. So first, I want to ask you, what brought you to your current position at Siemens-EDA? Right. So, you know, I did a master's in computer engineering from Rochester Institute of Technology. This was back in 2017. And my first job, actually, you know, I started at a company called D2S. It's a GPU exploration company. Great people.

I learned a lot. So that was my first role where I did pretty much everything from model calibration to OPC, a bit of ILD recipe tuning, and then a bit of mass process correction. So that was my introduction to the field. and, you know, the reality of manufacturing. So I spent about four years at the company, and then I decided to switch. So when I switched to Siemens EDA, I decided to focus more on the OPC and the ILB aspects of things, collectively called as, you know, mask synthesis solution.

So that brought me to Siemens EDA. Oh, great. So first off, for this conversation, can you quickly give us some background like, What is a photo mask? What happens during mask making? And why is that important now? So a photo mask is essentially the semiconductor industry's version of a stencil. It contains incredibly detailed patterns that defines where the transistors, wires, and all other circuit features will be created onto the silicon wafer. Now, during the manufacturing process, light is projected through the photo mask onto the wafer, thus transferring those patterns layer by layer.

So in a very real sense, every chip that starts with the information that's encoded on the mask. On the other hand, mask making is the process of turning a chip design, typically described in formats like GDS2 or OSS, into a physical photo mask with extreme precision. And that involves sophisticated data preparation, pattern optimization, mask writing, then inspection, and followed by verification steps to ensure that the final mask accurately represents the intended design. And that's where the challenge comes in. As semiconductor technology continues to scale to smaller and smaller nodes,

The patterns on these masks become extraordinarily complex. Hence, the margin for error is extremely small. So today, photo masks are no longer just manufacturing artifacts. They are a critical enabler of the advanced semiconductor technology. Even a small improvement in the mask accuracy, the quality, or the manufacturing process can translate into meaningful game. gain in the chip performance, the overall yield, and the production efficiency. Oh, great description.

So from what I've read, the industry is moving from Manhattan to curvilinear masks. What do these terms mean? And what are the challenges and why does this shift matter? Great question. So first of all, a quick definition. When people talk about Manhattan masks, they are referring to mask designs. mask designs made up almost entirely of straight lines and 90 degree angles. Now that has been the standard approach for decades because it was easier designing software and manufacturing tools to handle. On the other hand, curvilinear masks are different. So instead of forcing everything into rectangles and right angles, they allow smooth curves and more natural shapes. So why does that

matter. The physics of chip manufacturing does not really operate in perfect squares and rectangles. As chips become smaller and more advanced, curved shapes often match what the mask writer can actually print much more accurately. So as a result, you get better pattern fidelity with fewer distortion, improved yield, and ultimately better performing chips. Now this transition is important because it is happening right now. As industry moves to increasingly advanced process nodes, manufacturers are reaching the limits of what traditional Manhattan-based designs can deliver. Curvilinear masks are becoming a more practical way to squeeze more performance and manufacturability out of the

leading-edge technologies. So the real challenge is that this isn't simply a design upgrade. Curvilinear mass creates a huge increase in the complexity, the data volumes are larger, computations are more demanding, and companies need new algorithms with more powerful computing infrastructure and more sophisticated verification as well as the quality control processes. So in many ways, the entire semiconductor workflow has to evolve. But that is why the shift matters. The industry is moving from shapes that were convenient for software and design tools to the shapes that are better aligned with these realities of manufacturing as well as the physics. And that alignment can directly translate to better chips, higher yields, and a more efficient

path to future semiconductor innovation. Great. So can you bring us up to speed on OPC and fragmentation for curvilinear masks? All right, so OPC or optical proximity correction is a technique used to compensate for the distortions that occur when a chip pattern is transferred from the photo mask onto the silicon bifers. So in simple terms, engineers intentionally modify the design, the mask design, so that after the imperfections of the lithography process, the printed feature ends up looking exactly or close to the mask design so that after the imperfections of the lithography process, the printed feature ends up looking exactly or close. to the intended design. And a key part of that process is called fragmentation. Traditionally, the mask edges are broken into many small segments or you could call them fragments. So the OPC software can analyze and adjust each of these fragments independently. For decades, this approach worked very well and in conventional OPC, each fragment serves to purpose. First, it is the location being optimized,

and second it is also the point where correction is being applied and this model fits naturally with the Manhattan style geometries which are made up entirely of straight lines and right angles however curvilinear mass changes the equation here now curve shapes require far more detail to describe them accurately to capture those curves you often need many many more fragments and every other additional fragment increases the computational workload. As a result, there can be an explosion in data volume, memory requirements, and the overall optimization time. So what has worked efficiently for traditional geometries for the Manhattan OPC can now become a bottleneck when applied to these highly

complex curvilineate design. And which is why this topic is now getting so much attention. As the industry adopts curvilinear mask to improve the pattern fidelity as well as the manufacturability, it also needs at the same time new OPC approaches that can handle these sophisticated shapes without overwhelming the computing resources. So the true challenge is preserving the accuracy benefits of curvilinear design while keeping the runtime data size and overall complexity under control. So to sum it up, in short, curvilinear mask promises better printing accuracy, but they are also forcing the industry to rethink some of the core OPC and fragmentation

techniques that have been used for decades. Oh, interesting. So what are your recent innovations in mask making and OPC for curvilinear masks? Right. So we started by recognizing a fundamental challenge. With With Curvlinear Mask, a single shape can have thousands of potential locations where OPC optimization could be performed. But in reality, not every point along that shape is equally important. So if you try to optimize everything, you create this enormous computational overhead with larger data volumes and increasingly complex OPC recipe. So our question was simple. How do we focus our efforts where it actually?

actually matters. So the answer to this is what we call a vector-based decoupled site and anchor point framework. Let me break it down. So first off, it is a vector-based approach, meaning we are representing the shapes using their mathematical descriptions for points, lines, curves, and the direction, rather than treating them as a collection of tiny fixed elements or fragments. That gives us much greater flexibility when working with the complex curvilineal geometries. And the real innovation here is the separation of sites and the anchor points, which are kind of linked in the traditional approaches. So what is a site? A site is simply a location along the edge of a shape where you want to evaluate or optimize for accuracy. On the other hand, an anchor point is a control point that guides how those corrections are going to be applied. In traditional OPC, these two concepts are essentially tied together.

So every fragment is both the place you're optimizing and the place where these corrections are being made. But we have decoupled them. That means we no longer have to optimize every single fragment along an edge. Instead, we can place these optimization sites only where the accuracy is most critical, such as tight pitches, complex curves or the known hotspot while using the anchor points at the same time to efficiently control and guide the correction process now why is this important because it allows us to dramatically reduce the unnecessary computation while maintaining or in most cases even improving the optimization accuracy this results in more efficient OPC you have a simpler recipe with faster

with faster turnaround times and finer control over the most critical feature on a mask. And as curvilinear mask become more widely adopted, that efficiency becomes increasingly important. So in reality, we are moving from a world where every fragment was treated equally to a world where computational resources are directed exactly where they create the most value. And that is a key enabler for making advanced curvilinear mass technology practical at the production scale. So last question. Do you have customers using this now? Can you share some proof points? Yes.

So what I can say is that this is no longer a research project or a future concept. It is happening today in the production environment. So we are seeing leading edge foundries on both the logic and memory sites. adopting curvilinear OPC across a wide range of their applications. And these are some of the most advanced manufacturing processes in the world where every nanometer matters and where the improvements in pattern fidelity can translate directly into the yield and performance gain. What is particularly significant is that the conversation has now shifted from can curvilinear mask work to how do we deploy them efficiently,

at scale. That's where the innovations in OPC, the fragmentation, data representation, and computational efficiency becomes critical. Another strong indicator is the growing number of industry collaborations and technical publications. We are seeing more joint papers between the EDA providers, the foundries, memory manufacturers, the research institutes and the ecosystem partners that highlight the real world deployments and their production use cases. Overall, the broader industry is investing heavily in making curvilinear manufacturing practical at scale. And for anyone who wants to dive deeper, I would encourage you to visit the Siemens DDA Resources and Technical Paper webpage. You will find a growing collection of

conference paper, technical publications, and the joint presentations with our customers and partners that have showcased many of these advances and the real world application. So Covalinear OPC, mask data appropriation, mask process correction, and the verification technologies, these are all the active areas of collaboration across the semiconductor ecosystem today. So the proof point is really the momentum. Leading edge manufacturers are investing in this technology, deploying it on the advanced nodes and expanding its use across more and more layer as well as the application. And that tells us that the curvilinear mass and curve linear OPC are becoming an important

part of the industry's roadmap, not just for the next generation of the chips, but for the generations that follows as well. Great conversation. Thank you very much. Hopefully we can have you back on another topic. Oh, yeah. Thank you for listening. I mean, I truly appreciate your time and attention.

The semiconductor industry continues to evolve at a rapid pace, and it is an exciting time to be a part of it. So I appreciate your time, and I look forward to our future conversation. Thank you. That concludes our podcast. Thank you all for listening, and have a great day.