Parallel Overlapping-Domain Decomposition FDFD Models Large-Scale Nanostructures with Enhanced Efficiency

Electromagnetic simulations of increasingly complex nanoscale devices present a significant computational challenge, demanding new approaches to efficiently model their behaviour. Zhanwen Wang, Chengnian Huang, and colleagues from Zhejiang University, along with Yuntian Chen from Huazhong University of Science and Technology, now present a novel computational method that dramatically accelerates these simulations. Their work develops a parallel, overlapping domain decomposition technique built upon the finite-difference frequency-domain formulation, allowing researchers to model large and intricate nanostructures with unprecedented speed. The team’s method achieves substantial reductions in computation time, up to an order of magnitude faster than existing techniques, and delivers results that closely match both analytical calculations and established commercial software, marking a significant advance in the field of electromagnetic modelling.

The increasing complexity and scale of photonic and electromagnetic devices demand efficient and accurate numerical solvers. This work develops a parallel overlapping domain decomposition method, based on the finite-difference frequency-domain formulation, to model the electromagnetic response of large-scale complex nanostructures. The computational domain partitions into multiple overlapping subdomains, each terminated with perfectly matched layers, enabling seamless transfer of information between adjacent regions. A multi-frontal preconditioner accelerates the iterative solution process, and an OpenMP-based parallel implementation further enhances computational efficiency.

Domain Decomposition for Photonic Crystal Simulation

This research details a new method for accelerating large-scale electromagnetic simulations, specifically for analysing topological photonic crystals. Scientists present a domain decomposition method combined with a multi-frontal solver to efficiently handle the computational demands of these complex structures. Simulating their behaviour requires solving complex electromagnetic problems, but the computational cost limits the size and complexity of designs that can be analysed. This research aims to develop a method to significantly reduce computational time, enabling the analysis of larger and more complex designs.

The authors propose a domain decomposition method combined with a direct solver, the multi-frontal method, to tackle this computational challenge. The large computational domain, representing the photonic crystal, divides into smaller, non-overlapping subdomains, each processed independently, allowing for parallelization. A direct solver efficiently solves the electromagnetic equations within each subdomain and then combines the solutions across the subdomains. Key features include a source transfer technique to efficiently exchange information between subdomains, reducing communication overhead, and absorbing boundary conditions, utilising perfectly matched layers and complex coordinate stretching to accurately model infinite periodic structures. Optimized ordering algorithms and parallel processing further accelerate the solution process.

Efficient Electromagnetic Modeling via Domain Decomposition

Scientists developed a parallel domain decomposition method to model electromagnetic responses in large-scale, complex nanostructures, achieving significant improvements in computational efficiency. The work utilizes a finite-difference frequency-domain formulation and partitions the computational domain into overlapping subdomains terminated with perfectly matched layers, enabling seamless transfer of information between adjacent regions. This approach allows for the efficient solution of problems involving intricate geometries and diverse excitation types. Experiments demonstrate excellent agreement between the new method and both analytical solutions and commercial software.

Crucially, the team achieved up to an order of magnitude reduction in computation time when compared to existing methods. This breakthrough delivers substantial performance gains for modelling complex nanostructures, addressing a key limitation in the design and optimization of next-generation photonic and electromagnetic devices. The method incorporates a multi-frontal preconditioner, which significantly reduces the computational cost of each iteration by focusing on modifying only the source terms. Furthermore, the researchers implemented an OpenMP-based parallelization strategy, systematically analysing the impact of subdomain partitioning on parallel efficiency to provide practical guidelines for large-scale simulations. The results confirm the method’s ability to accurately model the scattering of electromagnetic waves from dielectric cylinders and to simulate propagation through large-scale trapezoidal-shaped topological optical waveguides.

Parallel Solver Accelerates Nanostructure Electromagnetic Simulations

This work presents a new parallel computational method for modelling electromagnetic responses in complex nanostructures, addressing the need for efficient and accurate solvers as devices become increasingly intricate. The researchers developed a domain decomposition method, building upon the finite-difference frequency-domain technique, which divides a large problem into smaller, overlapping subdomains. This approach, combined with a multi-frontal preconditioner and parallel processing using OpenMP, significantly accelerates the solution of the electromagnetic equations. The results demonstrate that this method achieves a substantial reduction in computation time, up to an order of magnitude faster than existing techniques, while maintaining a high degree of accuracy, as confirmed by comparison with analytical solutions and commercial software. The team also systematically analysed how the partitioning of subdomains impacts parallel efficiency, providing practical guidance for optimising simulations. While acknowledging that the computational cost increases with the complexity of the modelled structures, the authors highlight the method’s potential for large-scale electromagnetic modelling.