Highly parallelized RCWA with optimized eigenvalue problem for efficient simulation of curvilinear mask structures

The paper presents the further development of our established and well-proven RCWA based full 3D mask solver which was restructured to enable highly parallelized fast simulations of larger mask areas with curvilinear structures. The restructuring of the original algorithm with the goal to create parallel executable code blocks that are interrupted as less as possible in their sequence is essential for an efficient parallelization. The paper shows the realization of this concept for the RCWA algorithm on a graphics processing unit (GPU). The computation of the mask eigenmodes is the most time-consuming part of the algorithm. An optimized eigenproblem with significant speedup of the calculations is presented and the integration into the RCWA algorithm is shown. Simulation examples of curvilinearly structured larger EUV mask areas and the optical performance analysis of a meta-surfaces are presented. The superior speed of our method and the advantage over the decomposition technique are demonstrated.