We have developed modules in ARCHER that transport photons and electrons in voxelized geometry phantoms with realistic CT scanner model implemented. Three physical processes for photons including Compton scattering, Rayleigh scattering and photoelectric effect are being considered. Woodcock tracking algorithm is employed to make the simulation more efficent in heterogeneous media contained in human body. The code offers two methods to calculate the material attenuation coefficients: 1) Detailed physics, where atomic form factors are incorporated for various scattering, and the attenuation coefficients are calculated from the raw microscopic cross sections on the fly. 2) Simple physics, where atomic form factors are ignored and pre-tabulated cross section data is used. While the former gives more precise results, the latter provides faster speed. It is up to the user to decide which one to use based on the actual needs.


Electron transport part is based on the class-II condensed history method and continuously slowing down approximation (CSDA). Moller scattering and Bremsstrahlung are modeled for interaction with energy loss greater than the critical value. We handle the coupled electron-photon transport explicitly, that is, all the secondary particles produced in the primary particle transportation are stored and transported as well.


The code is developed by using the CUDA parallel computing platform invented by NVIDIA, and tested on Tesla M2090 GPU cards. For typical CT dose calculations, the GPU code is 1400x faster than the general purpose production CT code MCNP/MCNPX running on an Intel X5660 2.8GHz CPU. A typical dose calculation can be done within one minute on GPU compared with more than ten hours of running on CPU. The code also supports the multi-GPU running mode, and satisfactory scalability is obtained for up to six Tesla GPUs. Difference between results from the GPU code and MCNP-6 is in general less than 1%.


ARCHER is being designed to be a versatile testbed for studying Monte Carlo software design for emerging heterogeneous computing architectures that have uncertain specifics. Applications to x-ray CT imaging and radiation dosimetry programs have demonstrated impressive results. On-going efforts will develop more complex physical models and apply optimization techniques to improve both the accuracy and efficiency for different hardware/software platforms.