Parameter extraction of advanced semiconductor detectors with artificial neural network

Abstract

A novel method is presented to extract the important parameters of semiconductor detectors using artificial neural networks (ANN). We have designed a feed-forward (FF) Multi-layer Perceptron (MLP) ANN, with supervised training based on Levenberg-Marquardt (LM) back propagation algorithm (BP). In this method the sampling points of the transient current are the inputs of the network and the outputs are some important parameters of the detector, which are selected by sensitivity analysis. The training set is obtained by uniform sampling of the parameter space within the range of typical experimental data. These data are collected from drift-diffusion model for the transient current of laser illuminated hydrogenated amorphous silicon p+-i-n+ diode in reversed bias. At final step the trained ANN is used to extract the parameters of a real detector by using the transient current technique (TCT) measurements. It is observed that the simulated transient current with the extracted parameters by ANN exhibit an excellent agreement with experimental results. The new developments in semiconductor technology and the necessity of having a more accurate estimation of the parameters of advanced semiconductor devices to predict their behavior and also the capabilities of ANN in this field, is explained in section 1. In section 2 a brief review of ANN and its historical applications in designing and extracting the fabrication parameters of semiconductors devices is presented. The new advances in fabrication of amorphous devices, the theoretical bases and advantages of TCT and the signal induction are discussed in section 3. The complexity of hydrogenated amorphous silicon detectors, the trapping model, the simulation steps and also the required equations is the subject of section 4. In section 5 by using the Finite Element Method (FEM), the two dimensional simulation of the transient current of simple p+-i-n+ diode is performed and a method for reduction the execution time is presented. The strategy for extracting the transient current of the detector from the preamplifier output by using the Laplace transform is introduced in section 6. In section 7 the transient current of the detector is used to extract the electric field profile, the electron mobility, and the ionized dangling bonds and electrons collection time. In this section, the effect of lifetime of electrons and holes on the behavior of such detectors is also estimated. In section 8 the ANN structure and modeling scheme and some practical notes for optimizing the performance of ANN structure is presented. The final results and the performance of ANN in validation and testing steps and its excellent ability in extracting the parameters of a real amorphous silicon detector is also discussed in this section. © 2011 by Nova Science Publishers, Inc. All rights reserved.