COMSOL Multiphysics has integrated NVIDIA’s CUDA Direct Sparse Solver (cuDSS) into its standard solver framework to utilise NVIDIA GPUs for multiphysics simulations. Semiconductor manufacturing models often require detailed geometries, fine meshes and coupled physics, making them computationally demanding. The integration targets large sparse linear systems arising in nonlinear iterations, implicit time-stepping and parametric studies. It aims to reduce computation time by using GPU memory bandwidth and parallel processing while retaining physical detail.
Benchmarks compare NVIDIA GPUs with Intel Xeon CPUs across chemical mechanical polishing, copper electrodeposition, trench filling, photoresist spin coating, electrostatic discharge testing of circuit boards, and residual stress and warpage during chip-package underfill curing. These examples demonstrate the application of GPU computing across fluid flow, electrochemistry, electromagnetic and thermal-mechanical simulations.
Image generated by: ChatGPT