From Atoms to Reactors: Multi-Scale Modeling for Semiconductor Fabrication

Abstract

Accurately predicting surface topography evolution is essential for manufacturing complex three-dimensional (3D) semiconductor devices. To overcome the limitations of slow and costly experimental development cycles, a predictive, physicsbased simulation approach that bridges multiple scales is required. This paper presents comprehensive multi-scale modeling approaches that integrate simulations from the atomistic to the reactor level. At the lowest scale, we use Density Functional Theory (DFT) and Molecular Dynamics (MD), including machine learning-enhanced methods, to derive fundamental parameters like sticking coefficients and sputtering yields. These inform robust feature-scale models implemented in the open-source ViennaPS simulator, which captures complex surface kinetics by tracking local properties such as species coverages, passivation layer thickness, and physical damage. At the highest scale, reactor simulations provide physically-based boundary conditions, while surrogate models are developed to efficiently link equipment settings to process outcomes. The predictive power of the methods is demonstrated through a calibrated model for selective SiGe etching that accurately predicts etch profiles for different device geometries and process conditions. This holistic approach provides an end-to-end simulation capability, enabling predictive process optimization and establishing a foundation for future applications in automated process control and inverse design. Index Terms—Multi-scale modeling, Semiconductor fabrication, Process TCAD, Topography simulation, Molecular Dynamics, Plasma etching, PECVD, DTCO.