No abstract

No abstract

Objective Extreme ultraviolet (EUV) lithography has been introduced into highvolume manufacturing (HVM) of chips with a technology node of 7 nm and below. As the technology nodes of chips decrease, the structure of the EUV mask is becoming more and more complex. The defects in EUV masks degrade the mask imaging quality, which is one of the most critical problems affecting the yield of EUV lithography. Phase defects refer to the deformation of the EUV mask multilayer caused by the defects situated at the bottom of the multilayer. Phase defects of nanometer size can lead to a distinct phase shift of the reflected field and seriously degrade the aerial images. Defect compensation methods can be adopted to indirectly compensate for the degradation of imaging quality caused by the phase defects. Accurate inspection of the type, location, and profile of phase defects is the prerequisite for effective defect compensation. A method to inspect the type, position, and surface profile of phase defects in EUV masks on the basis of aerial images is proposed in this paper. The accuracy of the proposed method is verified by simulations.Methods Deep learning models are adopted to construct the mapping between aerial images of defective mask blanks and defect information. After that, the type, location, and profile of phase defects can be obtained from the aerial images of defective mask blanks by the trained models. The inspection model for the type and location of defects is built by the construction of the relationship between the type and location of defects and the aerial images of defective mask blanks with the convolutional neural network (CNN) model. On this basis, the aerial images are intercepted according to the obtained location of defects. The inspection model for the surface profile parameters of defects is constructed with the spectrum information of the intercepted aerial images and the multilayer perceptron (MLP) model. Results and DiscussionsA test group containing 256 defective mask blanks is utilized to verify the accuracy of the proposed method. The phase defects in the multilayer can be accurately classified into bump defects and pit defects by the trained CNN models (Fig. 6). The mean absolute error (MAE) of the x coordinates of the phase defects is 1. 38 nm, and the MAE of the y coordinates is 0. 74 nm, which indicates that the inspection accuracy of the y coordinates is higher than that of the x coordinates. The simulations show that the inspection accuracy of the location of bump defects is higher than that of pit defects (Fig. 7). For bump defects, the MAE of the surface height is 0. 06 nm, and the MAE of the surface full width at half maximum (FWHM) is 0. 55 nm. For pit defects, the MAE of the surface height is 0. 12 nm, and the MAE of the surface FWHM is 0. 57 nm (Fig. 8). Noise is added to the aerial images in the test group to examine the robustness of the trained models. The results reveal that noise lowers the accuracy of the trained models, and the inspection model for the type and location of de...

No abstract

No abstract