Applied Materials and NVIDIA Accelerate Semiconductor Pipeline from Atomic Modeling to Fab Digital Twins

Applied Materials and NVIDIA Accelerate Semiconductor Pipeline from Atomic Modeling to Fab Digital Twins Applied Materials and NVIDIA have developed an end-to-end digital development model that couples materials engineering and fabrication expertise with NVIDIA CUDA-X libraries. The collaboration accelerates semiconductor innovation across the entire pipeline, spanning atomic-scale discovery, multiphysics process simulation, and factory-level digital twins. Applied Materials integrated NVIDIA cuDSS into its Ginestra platform for up to a 10x speedup in defect simulations, while NVIDIA cuEST delivers up to a 55x speedup on B200 GPUs for density functional theory workflows. The Applied Materials ACE+ platform uses NVIDIA PhysicsNeMo to achieve up to 35x faster topography simulations for chamber and recipe development. NVIDIA Omniverse enables Applied Materials to build photorealistic digital twins of entire fabs to optimize layout, material flow, and operational throughput before physical ramp. Accelerating Atomic-Scale Materials Discovery As geometric scaling slows, semiconductor performance relies heavily on materials innovations in thin films, dielectrics, and conductors. To model complex structures like high-k metal gate stacks in gate-all-around transistors, Applied Materials uses Ginestra, a defect-centric simulation platform. Integrating NVIDIA cuDSS accelerates Ginestra's sparse linear algebra by up to 10x compared to CPU-only execution. For electronic structure modeling using density functional theory (DFT), NVIDIA cuEST reduces simulation runtimes on NVIDIA B200 systems from five days on 64 CPU cores to about two hours on a single GPU, an approximate 55x speedup that makes broad virtual experimentation feasible. Process Modeling and Recipe Optimization with ACE+ Moving verified materials into production requires managing complex chamber physics such as plasma dynamics, fluid flow, surface reactions, and heat transfer. Applied Materials' ACE+ platform models these multiphysics interactions, while NVIDIA PhysicsNeMo transforms the simulation results into real-time digital twins of process behavior. GPU acceleration speeds up ACE+ topography simulations by up to 35x, allowing engineers to transition from multi-day batch iterations to interactive, same-day adjustments before deploying validated processes to manufacturing platforms like Endura. Fab-Scale Digital Twins with Omniverse To address operational dynamics at the factory level, Applied Materials employs NVIDIA Omniverse libraries to generate physically accurate digital twins of complete fab facilities. These virtual environments allow engineering teams to simulate material flows, detect equipment bottlenecks, and test operational strategies prior to live deployment on the manufacturing floor, smoothing the transition to high-volume production.

应用材料与英伟达合作加速半导体材料工程与制造创新 应用材料公司与英伟达展开合作，将半导体制造及材料工程能力与NVIDIA CUDA-X加速库相结合，构建涵盖原子尺度材料发现、工艺开发到晶圆厂优化的端到端数字化开发模型，以应对AI算力需求驱动下的芯片设计与制造挑战。 在Ginestra平台中集成NVIDIA cuDSS稀疏线性代数库，模拟速度比纯CPU方案提升最高达10倍。 采用NVIDIA cuEST加速密度泛函理论（DFT）计算，在B200系统下单GPU将原需64核CPU耗时5天的模拟缩短至约2小时（约55倍加速）。 结合NVIDIA PhysicsNeMo，应用材料ACE+多物理场平台的形貌仿真速度提升最高达35倍，实现实时工艺行为数字孪生。 利用NVIDIA Omniverse构建晶圆厂级物理精度数字孪生，在实际投产前优化布局、物料流并消除瓶颈。 原子尺度材料建模加速 随着几何微缩放缓，先进半导体日益依赖介电层和导电层等薄膜的材料创新。应用材料的Ginestra平台针对全环绕栅极（GAA）晶体管等3D器件中的高k金属栅极堆叠，结合测量数据与第一性原理模拟预测器件可靠性与可变性。集成NVIDIA cuDSS后稀疏线性代数计算提速最高达10倍。同时，针对高计算强度的密度泛函理论（DFT），通过NVIDIA cuEST库在B200系统上实现约55倍提速，将单次复杂模拟从数天缩减至数小时。 工艺与腔体多物理场仿真优化 在材料落地制造环节，应用材料ACE+平台集成流体动力学、热传递、等离子体动力学和表面反应等多物理场模拟。借助GPU加速与NVIDIA PhysicsNeMo，ACE+形貌仿真速度最高提升35倍，使原本需要数天的批处理转变为支持实时探索的交互式工作流，缩短工艺开发周期。 晶圆厂级别数字孪生与量产落地 经过验证的工艺创新可部署于Endura等集成材料解决方案平台。在整厂生产层面，应用材料基于NVIDIA Omniverse构建全厂高精度物理数字孪生，用于模拟物料流动、排查运行瓶颈并验证运营策略，从而加速向高良率、规模化量产的爬坡进程。

As AI workloads increase, explosive compute demand is pushing the semiconductor industry to meet unprecedented performance targets. Even small delays can have outsized financial impact in fast-moving AI hardware cycles. Simultaneously, the shift from chip-level optimization to system-level engineering is compounding thermal and power challenges. Meeting these demands requires breakthroughs in materials deep within the device stack. This calls for advanced modeling and simulation beyond what conventional approaches can deliver. This post explores how Applied Materials and NVIDIA are tackling this challenge with an end-to-end digital development model that combines Applied Materials leadership in materials engineering and semiconductor manufacturing with NVIDIA CUDA-X libraries . This collaboration accelerates the entire innovation pipeline—from atomic-scale discovery and engineering through process development to factory optimization. How are NVIDIA and Applied Materials accelerating semiconductor innovation? At the front end, GPU-accelerated simulations expand the design space, giving materials engineers a powerful advantage in developing next-generation devices. In the middle, physics-based modeling speeds up chamber and recipe development. At the back end, AI-driven digital twins predict fab performance before changes hit the production floor. Putting these together creates a continuous flow of insights that drives faster progress from atoms to fabs. Figure 1. The end-to-end digital model developed by NVIDIA and Applied Materials connects atomic-scale discovery, process engineering, and factory optimization. Image credit: Applied Materials Advancing atomic-scale materials modeling As geometric scaling slows, progress in advanced semiconductors increasingly hinges on materials innovation in dielectrics, conductors, and other thin films. At atomic levels, even minor variations in material structure and interfaces can significantly impact performance, reliability, and yield. Identifying these effects through physical experimentation alone is too slow and costly. Applied Materials addresses this constraint with Ginestra, a physics-based, defect-centric simulation platform that connects material properties and defects to predicted device performance. One key application is the high-k metal gate (HKMG) stacks in gate-all-around (GAA) transistors and other advanced 3D devices, which pack five distinct materials into a space barely 1/10,000th the width of a human hair. By combining metrology data with first-principles simulation, Ginestra projects device variability and reliability directly from atomic-scale material characteristics. Figure 2. Gate-all-around transistor and gate stack layer breakdown . Image credit: Applied Materials Historically, achieving the required simulation fidelity needed significant compute time, restricting how broadly engineers could explore the materials landscape. By integrating NVIDIA cuDSS (CUDA-X Direct Sparse Solver) into Ginestra, Applied Materials accelerates the sparse linear algebra at the core of these simulations, delivering up to a 10x speedup over CPU-only approaches without sacrificing accuracy. Engineers can now run thousands of virtual experiments across material combinations and reaction pathways, dramatically expanding the pool of viable candidates. While this broadens the search space, accurate atomic-level prediction also depends on density functional theory (DFT), the industry-standard method for modeling electronic structure. DFT is highly predictive but also highly compute-intensive. Traditional CPU-based simulations can take days or even weeks, limiting DFT industrial scalability. NVIDIA cuEST (CUDA Electronic Structure Theory) accelerates the most demanding steps of the DFT workflow. cuEST is one of the most recent CUDA-X libraries offered by NVIDIA to accelerate quantum chemistry applications on GPUs. Its modular APIs accelerate the most computationally demanding building blocks of Gaussian-basis density functional theory while integrating with existing quantum chemistry workflows. On NVIDIA B200 systems, simulations that once took five days on 64 CPU cores can now be completed in about two hours on a single GPU—an approximately 55x speedup. Even complex reaction-search tasks that previously required hundreds of CPU cores for over a week can now run in hours. Together, Ginestra workflows and CUDA-X shift more materials engineering into simulation. This helps engineers expand exploration, reduce reliance on costly experiments, and accelerate discovery of optimal materials. Bringing materials innovation to high-volume manufacturing After identifying promising materials, the next challenge is translating those insights into repeatable, high-yield manufacturing processes governed by tightly coupled physics—fluid flow, heat transfer, plasma dynamics, surface reactions, and electromagnetic effects. Figure 3. Applied Materials ACE+ platform simulates coupled chamber physics—fluid flow, heat transfer, plasma dynamics, and surface reactions—running up to 35x faster with NVIDIA PhysicsNeMo. Image credit: Applied Materials The Applied Materials ACE+ platform brings these multiphysics models together to simulate how chamber design, hardware configuration, and process parameters interact. NVIDIA PhysicsNeMo then converts simulation results into real-time digital twins of process behavior. With GPU acceleration, ACE+ topography simulations run up to 35x faster—compressing workflows that once took days into same-day results. Engineers can dynamically explore parameter changes and see instant impact, shifting from batch-based iteration to a continuous, interactive workflow that converges on optimized conditions much earlier. Validated process innovations are then deployed into production using Applied Materials Integrated Materials Solutions platforms. One example is Endura, the semiconductor industry’s most successful metallization system, combining multiple process steps under vacuum to deliver the precision required for advanced chipmaking. From there, the focus expands to the full fab environment, where tool interactions and workflow dynamics ultimately determine throughput, cost, and yield. Figure 4. A photorealistic fab digital twin built with NVIDIA Omniverse, used to simulate material flow, spot bottlenecks, and validate operational strategies before deployment. Image credit: Applied Materials At the factory scale, Applied Materials leverages NVIDIA Omniverse libraries to build physically accurate digital twins of entire fabs. These environments allow teams to optimize layouts, simulate and streamline material flow, identify bottlenecks, and validate operational strategies before deployment—accelerating ramp to high-volume manufacturing. Figure 5. The Applied Materials–NVIDIA digital thread unifies atomic-scale discovery, process engineering, and high-volume manufacturing into one continuous semiconductor innovation pipeline. Image credit: Applied Materials Learn more To meet the growing compute demands of the AI era, the semiconductor industry must embrace a connected model spanning atomic-scale materials engineering, process development, and manufacturing optimization. By combining Applied Materials’ semiconductor fabrication expertise with NVIDIA CUDA-X accelerated simulation, engineers can tackle critical bottlenecks faster and speed up the path to next-generation technologies. More broadly, this reflects the industry’s shift toward cross-ecosystem innovation—exemplified by initiatives like Applied’s EPIC platform—where progress increasingly comes from closer collaboration across the semiconductor value chain. cuEST advances high-accuracy electronic-structure modeling, cuDSS accelerates large-scale engineering simulations, NVIDIA PhysicsNeMo transforms simulation data into real-time digital twins, and NVIDIA Omniverse connects data, models, and workflows through physically based digital twins. Together, these technologies create a continuous digital thread across the semiconductor innovation lifecycle—accelerating the development of the foundational technologies that will power the future of AI. Learn more about accelerating semiconductor design and manufacturing .