Acceleration of FDTD mode solver by high-performance computing techniques.

Area of Science:

• Computational Electromagnetics
• Optical Waveguide Theory
• High-Performance Computing

Background:

• Accurate mode calculation is crucial for designing optical waveguides.
• Conventional eigenmode solvers face memory limitations for complex structures.
• Finite-difference time-domain (FDTD) methods offer an alternative approach.

Purpose of the Study:

• To develop a novel 2D compact FDTD mode solver.
• To leverage the matrix pencil method (MPM) for enhanced accuracy.
• To implement the solver on GPUs for significant computational acceleration.

Main Methods:

• Wave equation formalism combined with the matrix pencil method (MPM).
• Implementation of the FDTD algorithm on graphics processing units (GPUs) using CUDA.
• Validation against benchmark finite-difference (FD) eigenmode solvers for real and complex modes.

Main Results:

• The GPU-accelerated FDTD solver achieves over 30x computational efficiency improvement compared to CPU-based FDTD.
• Computational efficiency is comparable to standard FD eigenmode solvers.
• The new method requires significantly less memory (less than 10%) than conventional solvers.

Conclusions:

• The developed 2D FDTD MPM solver is an efficient, accurate, and robust tool for optical waveguide mode calculations.
• It overcomes memory limitations of conventional methods, enabling analysis of complex waveguide structures.
• High-performance computing with GPUs provides substantial speed-up for electromagnetic simulations.