Machine-learning-assisted design of depth-graded multilayer x-ray structure

Depth-graded multilayer structures are widely used in X-ray related applications. In this paper, we propose an optimization approach using machine learning principles to accelerate depth-graded multilayer structures design. We use Monte Carlo tree search (MCTS) to find optimal thickness for each layer in the structure that achieves maximum mean reflectivity in an angular range at a specific beam energy. We obtained 0.78 mean reflectivity in an angular range 0.4~0.55° for Cu Kα radiation using this approach. For a at top structure, we could achieve a small standard deviation of 0.016 within the same range. MCTS is an iterative design method that employs tree search with guided randomization that showed exceptional performance in computer games. MCTS expands towards the promising areas of the search space making it able to search large spaces efficiently and systematically. This approach offers flexibility for multiple design purposes without the need to data availability in advance.