Method based on golden section search for reconstruction of nano layer structure directly from depth profiling data

In this paper, a method for the dynamic optimization of the iteration step τ in Newton’s method is proposed, based on the golden section search algorithm. This method combines the total variation-Tikhonov regularization, Newton’s method, and the L-curve criterion by the simulated annealing algorithm for reconstruction of the original nano layer structure directly from the measured depth profiling data. The golden section search algorithm finds the optimal value for each iteration step and makes the iteration procedure to converge quickly with high computational efficiency. The deconvolution results show that the total deconvolution speed is improved by nearly 154 times with a deconvolution error of 8.7 × 10−14% as compared to the previous studies with a constant of τ, indicating an excellent reconstruction of the original “square-wave like” layer structure. In addition, this method could deal with the depth profiling data containing sputtering-induced roughness and/or Gaussian noise. Finally, this method is applied for the reconstruction of the original layer structure from the measured Auger electron spectroscopy depth profiling data of 4 × Si(15 nm)/Al(15 nm)/Si(100).