MBIR reflectance spectrometry for deep trench structure with ANN and Levenberg-Marquardt combined algorithm

Abstract

Model-based infrared (MBIR) reflectance spectrometry has been introduced for characterization of the depth and profile of deep trench structures in dynamic random access memory (DRAM). Modeling the complex trench structure as a multilayer optical film stack with effective medium approximation (EMA) allows the determination of both trench depth and width from Fourier-transfer infrared (FTIR) reflectance spectrum. In this paper an algorithm combining artificial neural networks (ANN) and Levenberg-Marquardt (LM) is proposed to extract the geometric parameters from the measured reflectance data. An initial estimate of the geometric parameters is obtained by the ANN, and then it is used as an input for the LM algorithm which converges to a final solution with a few iterations. The combined algorithm has been implemented on our own experimental platform, and it has been demonstrated to achieve very high accurate results as well as fast enough computation ability.