Advanced packaging allows chiplet integration and maximizes device performance with faster product development cycle, lower cost, and higher yield. As the package size becomes bigger and the device is getting more complicated, there is growing motivation to employ manufacturing process simulation, Artificial Intelligence (AI) assisted process optimization, yield and reliability prediction, rather than conventional methods, to ramp the yield and to ensure the reliability of a new product. The key for an accurate process simulation model is to input precise material properties, such as modulus, Coefficient of Thermal Expansion (CTE), dielectric constant, glass transition temperature, etc., which could change non-linearly with temperature, moisture, as well as other environmental factors and process conditions. Molecular modeling and molecular dynamics can provide insights into post chemical reactions or physical transformations via atomic and molecular simulations Lithography Techniques for Redistribution Layer (RDL) fabrication are the foundation of Advanced Packaging techniques, such as Fan Out Wafer Level Packaging (FOWLP), Fan Out Panel Level Packaging (FOPLP), 2.5D, 3D, and 3.5D packaging with RDL interposers. The continuous scaling-down of critical dimensions (CDs) in advanced packages, including via diameters, routing line and space (L/S), to a few microns, or submicron level, as well as the increasing number of RDL layers at panel scale pose significant challenges in RDL lithography techniques. For example, the Photo Imageable Dielectric (PID) or other build-up dielectric materials used in multilayer RDL fabrication are polymers, having low Young’s modulus, high CTE, and big volume shrinkage after curing. These material properties could cause fabrication process induced warpage and surface topography deformations, such as non-planarity, roughness, contamination, defects, and dimensional variations, which could potentially lead to massive yield loss when forming fine features during the multilayer RDL patterning. This paper presents material simulation methodologies based on quantum mechanics (QM), molecular dynamics (MD), and Machine Learning (ML), which are adopted to predict the material properties of a PID material, including glass transition temperature (Tg), CTE, mechanical properties, dielectric properties, as well as volume shrinkage after curing. Comparison between the simulation results and the experimental data is performed to validate the methodology. Similar methodology could be used to predict material properties of other organic packaging materials, which is crucial for building up accurate process, yield, and reliability simulation or digital twin of advanced packaging.
Li et al. (Fri,) studied this question.