Recognition of diffraction-grating profile using a neural network classifier in optical scatterometry

Abstract : Optical scatterometry has been given much credit during the past few years in the semiconductor industry. The geometry of an optical diffracted structure is deduced from the scattered intensity by solving an inverse problem. This step always requires a previously defined geometrical model. We develop an artificial neural network classifier whose purpose is to identify the structural geometry of a diffraction grating from its measured ellipsometric signature. This will take place before the characterization stage. Two types of geometry will be treated: sinusoidal and symmetric trapezoidal. Experimental results are performed on two manufactured samples: a sinusoidal photoresist grating deposited on a glass substrate and a trapezoidal grating etched on a SiO₂ substrate with periods of 2 m and 0.565 m, respectively.