Process TCAD (Technology Computer-Aided Design) refers to the simulation of fabrication steps in semiconductor manufacturing. As device dimensions continue to shrink and new materials are introduced, experimental process development becomes increasingly costly and time-consuming. TCAD tools address this challenge by enabling predictive modeling of key fabrication steps, such as lithography, etching, deposition, implantation, and annealing.
By capturing the complex interplay of physical and chemical processes at the feature scale, Process TCAD helps engineers design, analyze, and optimize semiconductor processes, reducing cleanroom iterations, improving yield, and accelerating innovation. 
These fabrication steps are not performed just once. Instead, they are repeated across dozens of cycles to construct a complete integrated circuit (IC), with photolithography playing a central role in each iteration (Fig. 1). Our research focuses on simulating this multi-step process in detail, using advanced frameworks and high-performance algorithms. Some of our recent developments and contributions are highlighted below. 

ViennaPS is our open-source, high-performance simulation platform for 2D and 3D feature-scale process modeling. At its core, ViennaPS employs advanced level-set methods to accurately capture evolving material interfaces, while leveraging Monte Carlo ray tracing for simulating the transport and interaction of particles involved in deposition, etching, and other surface reactions.
The framework is highly modular and extensible, providing C++ and Python interfaces that enable customized simulation pipelines and tight integration with data-driven or AI-based workflows. ViennaPS supports a wide range of physical models, including:
    • Directional and isotropic etching and deposition
    • Sticking, scattering, and re-emission of particles
    • Boolean operations and morphological filters
    • Complex topographies via mesh-based surface descriptions
A schematic overview of the ViennaPS architecture, showing its integration with level-set solvers, ray tracing backends, and chemical modeling modules, is shown in Fig. 2.
One of the key strengths of ViennaPS lies in its high-performance computing support. Simulations can be accelerated using multi-threaded CPU backends as well as GPU hardware via NVIDIA OptiX™, enabling large-scale or high-resolution models to be executed efficiently. This makes it ideal for applications such as high-aspect ratio etching, gap fill studies, or topography-aware process optimization.

Accurate and predictive process simulation requires the integration of multiple physical models spanning a wide range of spatial and temporal scales, from atomistic reactions at surfaces to plasma behavior in the entire chamber. Our multi-scale approach, illustrated in Fig. 3, combines atomistic, feature-scale, and equipment-scale models to achieve physically consistent and computationally efficient simulations for advanced semiconductor processes.
At the atomic scale, techniques such as Density Functional Theory (DFT) and Molecular Dynamics (MD) are employed to compute fundamental material interactions. These models provide critical data, including sputtering yields, adsorption energies, sticking coefficients, and surface passivation dynamics. Such data serve as input parameters for feature-scale simulations, particularly in precision processes like atomic layer etching (ALE) or selective deposition.
At the feature scale, the core of our simulation infrastructure is ViennaPS, a high-performance, open-source simulator based on level-set methods and Monte Carlo ray tracing. It captures the evolution of 2D and 3D geometries under directional fluxes, incorporating physical models for re-emission, shadowing, etching/deposition dynamics, and Boolean operations. ViennaPS supports both CPU and GPU backends, leveraging NVIDIA OptiX™ for ray tracing acceleration, enabling fast, scalable modeling of complex nanoscale topographies such as trenches, vias, and fins.

Accurate and predictive process simulation relies on well-calibrated models that reflect experimental reality across a range of operating conditions. To streamline this calibration, we present ViennaFit, an open-source optimization framework tightly integrated with ViennaPS. ViennaFit automates the parameter fitting process by comparing simulated surface profiles with experimental data, such as 2D SEM or TEM micrographs, using advanced discrepancy metrics including area difference (AD) and narrow-band distance (NBD). These metrics quantify geometric mismatches and drive gradient-free optimizers (e.g., dlib, Nevergrad) to iteratively minimize them. Fig.5 illustrates the automated calibration loop in ViennaFit, where signed-distance fields (SDFs) of experimental and simulated structures are compared, and the discrepancy is minimized through optimizer feedback.
Beyond calibration, ViennaFit supports global sensitivity analysis (GSA) using Sobol indices, allowing users to systematically identify which input parameters most strongly affect simulation outputs. This dual capability enables not only efficient calibration but also robust model reduction, design-space exploration, and inverse process design.
Building on this foundation, our research increasingly incorporates artificial intelligence (AI) to further enhance simulation workflows. In particular, machine learning (ML) models trained on hybrid datasets, combining physics-based simulation data with sparse experimental measurements, enable both rapid prediction and greater interpretability. For example, in processes like plasma-enhanced chemical vapor deposition (PECVD), explainable AI techniques such as SHAP have been used to analyze the relative importance and interdependence of key process parameters.
AI methods are also being leveraged for surrogate modeling of complex equipment behavior at the feature scale, accelerating multi-scale simulations and enabling real-time optimization of process conditions. These developments reflect a broader trend toward data-driven TCAD, and are discussed further in the section AI for Microelectronics, where we explore how ML and physical modeling are merged for physics-based and accelerated semiconductor process simulation.