varRCWA

The basic implementation of varRCWA, RCWA, differential method with both periodic boundary condition and PML in the paper:

VarRCWA: An Adaptive High-Order Rigorous Coupled Wave Analysis Method. Ziwei Zhu and Changxi Zheng ACS Photonics 2022 9 (10), 3310-3317. DOI: 10.1021/acsphotonics.2c00662

Examples

some example codes are in src/tests

• test_cpu.cpp: simple examples on CPU (here I write a few comments, you can start here)
• test_gpu.cpp: simple examples on GPU
• test_gds_cpu.cpp: GDSII examples on CPU
• test_gds_gpu.cpp: GDSII examples on GPU
• test_metasurface_cpu.cpp: metasurface (periodic boundary) examples on CPU
• test_metasurface_gpu.cpp: metasurface (periodic boundary) examples on GPU

Some other tests

• test_cpu_differential_method.cpp: test the stability of differential method

• test_oe_model_gpu.cpp: test the optics express mode converter on GPU

• the rest are for development only, please overlook them

• The main algorithm of varRCWA is located in src/core/RedhefferIntegrator.cpp

• The main algorithm of RCWA is located in src/core/RCWAIntegrator.cpp

• The main algorithm of the differential method is located in src/core/DifferentialIntegrator.cpp

• The GPU version of the above algorithms are in src/core/XXXXIntegratorGPU.cpp

Some other comments:

• For comparison with RCWA, the order of the element in the scattering matrix can be sensitive depending on the computational architecture. (A problem with most eigenvalue decomposition algorithm.) If you are really comparing the full scattering matrix including all the unguided modes, please compare the results from GPU with GPU and CPU with CPU. If you are compare some elements, please compare the non-degerated modes.
• The mode normalization in RCWA is different from most other algorithms as it is in the Fourier space. The energy of different modes with normalized vectors in Fourier space can be different. Please refer to the function varRCWA_benchmark_guidedmodes in test_cpu.cpp for energy-based normalization.

We are working to improve the building and usage process, please email ziweizhu95@gmail.com if you have any feedback or suggestions.

Build and Dependencies

We use Cmake 3.25.0 and Make (the standard build system on Linux(Ubuntu)) for building the library. Please use g++-10 to build as we use some C++20 features (like numbers). The CUDA compiler should be 11.4. You can check the version of your installation by the following commands:

cmake --version
gcc --version
g++ --version
nvcc --version

A tutorial on CMake is available here

There are some options in CMakeLists.txt

option(WITH_OPENMP "enable OpenMP acceleration or not" ON)
option(BUILD_DEBUG "Turn on the debug mode" OFF)
option(BUILD_TESTS "Build unit test cases" ON)
option(BUILD_GPU "Build GPU examples" OFF)

Here I set the BUILD_GPU to be off, you can turn it on if CUDA and Magma is correctly installed. It takes some time but will be worthwhile!

Eigen (header only)

Download from here

OpenMP

sudo apt-get install libomp-dev

Intel OneAPI (MKL, TBB)

Download from here

Magma (Optional for GPU code)

Download from here

After installing Magma, please take a look at the following line of CMakeLists.txt

if (BUILD_GPU)
  set(MAGMA_INCLUDE_DIR /usr/local/include)
  set(MAGMA_LIBRARIES libmagma.so;
      libiomp5.so)
endif()

You may need to change the directory (/usr/local/include) of where you install Magma header files.

CUDA 11.4 (Optional for GPU code)

Download from here

G++-10

tutorial here

To build

For CPU

cmake -D CMAKE_CUDA_HOST_COMPILER=gcc -D CMAKE_CXX_COMPILER=g++ .
make -j8

For GPU (use CUDA 11.4 please!)

cmake -D CMAKE_CUDA_HOST_COMPILER=gcc -D CMAKE_CUDA_COMPILER=/usr/local/cuda-11/bin/nvcc -D CMAKE_CXX_COMPILER=g++ .
make -j8

The address of the CUDA compiler nvcc may need to be changed depending on where you install it.

Citation

Any publications resulting from the use of this software should cite the following papers:

@article{zhu2022VarRCWA,
  author = {Zhu, Ziwei and Zheng, Changxi},
  title = {VarRCWA: An Adaptive High-Order Rigorous Coupled Wave Analysis Method},
  journal = {ACS Photonics},
  volume = {9},
  number = {10},
  pages = {3310-3317},
  year = {2022},
  doi = {10.1021/acsphotonics.2c00662},
  URL = {https://doi.org/10.1021/acsphotonics.2c00662},
  eprint = {https://doi.org/10.1021/acsphotonics.2c00662}
}