Extreme ultraviolet (EUV) lithography is the most advanced lithographic technology. However, during exposure processes, even small defects in masks can cause significant changes in the critical dimensions (CD) of wafers. Therefore, detecting and repairing defects of a certain size in masks is crucial. As the chip technology nodes decreases, detecting and repairing mask defects become increasingly challenging. EUV mask defects are classified into amplitude and phase defects, based on their positions and effects on reflection fields. Amplitude defects are typically located at the top or absorption layers of multilayer films. They significantly reduced the overall reflectivity of the multilayer films to EUV light at the defect sites, directly affecting the amplitude of the reflected field. However, most phase defects are found in the substrate or multilayer film, deforming of the multilayer film and affecting the phase of the reflected light. Although the total reflectivity of the multilayer film for EUV light remains high, the phase shift of the reflected field has a small impact on the amplitude. Compared with amplitude defects, phase defects are more challenging to detect and almost impossible to repair. Owing to the difficulty in detecting phase defects through optical and electron beam inspections, a photochemical inspection is necessary for EUV masks. Samsung developed an extreme ultraviolet lithography mask defect inspection system (EMDRS) that uses higher harmonics to generate EUV light sources capable of detecting both amplitude and phase defects. However, imaging processes resulted in blurred scanned images and reduced resolution of the defect detection system, owing to the sizes and shapes of light spots. Hence, a binary image deblurring algorithm is proposed in this study to improve the resolution of EUV mask defect detection systems.

The EMDRS employs a zone plate to focus coherent EUV light onto the mask surface, creating an illumination spot of 82 nm. The reflection intensity within the receiving angle of the detector is measured. The focused EUV light irradiates the mask surface at an incident angle of 6°, and factors like 3D shadowing help narrow line widths after scanning. However, for simulation purposes, this study assumes the EUV mask as a two-dimensional surface. The sample image primarily consists of space/line images. A transfer function f(x,y) is established to represent the reflection field of the sample. Currently, EUV mask patterns can be simplified into two cases: reflection or non-reflection of EUV light. Hence, the function f(x,y) is a binary distribution under ideal conditions. For calculation convenience, amplitude defects are assumed to completely block the reflection of the EUV light by the multilayer film, thereby reducing the amplitude of the reflected field to zero. In the case of phase defects, the effect of the multilayer film on EUV light reflectivity is assumed to be negligible, affecting only the phase while maintaining a constant amplitude at the defect site. The sample image is scanned point by point using a Gaussian spot with a beam waist diameter of 81 nm and scanning step of 1 nm. Then, the scanned image of the sample is subjected to deblurring. Broken line, pinhole, and phase defects were used as examples, and it was observed that employing a deblurring algorithm to process the scanned images considerably enhanced the resolution of the EUV mask defect detection system.

In the simulation of the broken-line defects (Fig. 3), a line with a 4 nm segment missing in the middle was introduced into the space/line image with a period of 120 nm to emulate a real broken-line defect in the mask. Deblurring of the scanned image caused the defects to become clearer, making them easier to identify, and the contrast of the defect signal was improved by 52.03%. Pinhole defects showed the most significant enhancement (Fig. 4). A square pinhole defect measuring 10 nm×10 nm was added to each sample pattern. After deblurring, the defect signal in the image was enhanced significantly, resulting in a 64.89% improvement. Broken-line and pinhole defects are amplitude defects, whereas phase defects are more challenging to detect. In the simulation of a phase defect (Fig. 5), a phase defect with dimensions of 50 nm×50 nm was designed with an amplitude of 1 and Gaussian-shaped phase changes. The phase at the center was π/2. Deblurring of the scanned image amplified the intensity of the defect signal, improving the contrast of the defect signal by 44.96%. We analyzed the defect signals under different system noises (Fig. 6) and found that the signal contrast of the three defects increased by 58.75%, 62.17%, and 44.48% by the addition of Gaussian noise with an average amplitude of 1%. The enhancement effect of the deblurring algorithm on the defect signals became weak when the system noise amplitude increased to 4%.

This study applied a text image deblurring method based on L₀ regularization intensity and gradient prior to the resolution improving of a EUV mask defect detection system. A blind deblurring method that leverages prior knowledge and iterative optimization was used to obtain a clear image. The deblurring process could be completed solely by inputting a blurred image, without requiring additional information, such as faculae. Simulation research on broken line, pinhole, and phase defects demonstrated that deblurring scanned images result in clearer defect images, with improvements of 52.03%, 64.89%, and 44.96% in signal contrast, respectively. This enhancement significantly improves the detection capabilities of mask defect detection systems, thereby satisfying the demand of smaller lithography process nodes.