提出了一种极紫外光刻掩模多层膜相位型缺陷的形貌重建方法。采用表面与底部形貌参数表征相位型缺陷的三维形貌;采用原子力显微镜测量缺陷表面形貌参数;采用傅里叶叠层成像技术重建含缺陷的空白掩模空间像复振幅;采用卷积神经网络与多层感知器两种深度学习模型构建空间像振幅/相位与缺陷底部形貌参数之间的关系,建立缺陷底部形貌参数重建模型;利用训练后模型从空间像的振幅与相位信息中重建出缺陷底部形貌参数。仿真结果表明,训练后模型可准确重建相位型缺陷的底部形貌参数。凸起型与凹陷型缺陷的底部半峰全宽重建结果的均方根误差分别为0.51 nm和0.43 nm,底部高度重建结果的均方根误差分别为3.35 nm和1.73 nm。由于采用空间像作为信息载体,本方法不受沉积条件的影响。

This paper proposed a new method for profile reconstruction of phase defects in extreme ultraviolet lithography mask multilayer films. Three-dimensional profiles of phase defects were characterized using the top and bottom profile parameters. The top profile parameters of defects were measured using an atomic force microscope. Moreover, Fourier ptychography technology was used to retrieve the complex amplitudes of aerial images of the defected mask blanks. Using deep learning models, the bottom profile parameter reconstruction model of defects was constructed by determining the relationship between the amplitudes/phases of aerial images and the bottom profile parameters of defects. The deep learning models used herein include a convolutional neural network and multilayer perceptron. The bottom profile parameters of defects can be reconstructed from the amplitudes/phases of the aerial images using the trained models. The simulation results show that the trained models can accurately reconstruct the bottom profile parameters of phase defects. The root-mean-square errors of bottom full-width-half-maximum reconstruction results of bump and pit defects are 0.51 and 0.43 nm, respectively. The root-mean-square errors of bottom height reconstruction results are 3.35 and 1.73 nm, respectively. The proposed method is immune to the deposition conditions because it captures aerial images as an information carrier.