Thin-Film X-Ray Analysis Using Neural Network Parameter Estimation

The patent applies preliminary action by using a neural network to pre-estimate initial parameter values before performing local optimization. The neural network is trained offline on synthetic data generated from global optimization results, enabling it to predict accurate initial values that are close to the global optimum. This preliminary estimation eliminates the need for time-consuming global optimization during actual analysis, while still achieving accurate parameter optimization through subsequent local optimization.

The patent replaces the mechanical global optimization process with a neural network-based estimation system. Instead of executing computationally intensive global optimization algorithms during analysis, the system uses a pre-trained neural network that has learned the complex relationships between X-ray profiles and optimal parameters. This substitution transforms the mechanical optimization process into an efficient pattern recognition task, dramatically reducing analysis time while maintaining accuracy.

The patent performs preliminary action by pre-training the neural network offline using synthetic data generated from global optimization. During actual analysis, the pre-trained network provides accurate initial parameter estimates that guide local optimization to the correct solution. This shifts the computational burden from online analysis to offline training, ensuring reliable parameter determination during analysis while minimizing real-time computational resource consumption.

The patent uses copying by generating synthetic training data that replicates the characteristics of real X-ray profiles. The neural network learns from these copied synthetic examples, which are created through simulation based on physical models. This allows the system to acquire optimization expertise from synthetic copies without requiring extensive real measurement data or repeated global optimization computations on actual samples.