Deep Learning-Assisted Trap Extraction Method from Noise Power Spectral Density for MOSFETs

A novel Deep Learning (DL)-Assisted Trap Extraction method is proposed to extract trap spatial and energetic distribution from the noise power spectral density (PSD). The method comprises two steps: 1) rough extraction using a customized multi-scale receptive fields Convolutional Neural Network (CNN) and 2) refinement using Backpropagation Optimization (BO) network based on Gradient Descent (GD). This automated approach achieves precise trap extraction with a mean difference as low as 1.5x, and the extracted traps' noise PSD matches well with the original PSD. Finally, the method is applied to a practical case, and successfully extracts the trap distribution.