一種採用非成像散射偵測技術的極紫外微影光罩缺陷檢測新方法

Abstract

Photolithography is a crucial process to extend the feasibility of Moore’s Law which indicates that the number of transistors on a chip doubles about every 18-24 months. According to International Technology Roadmap of Semiconductors (ITRS), extreme ultraviolet (EUV) lithography with a short wavelength of 13.5 nm is a promising candidate as next generation of lithography. Due to the property of EUV, all the optical elements including the mask need to be reflective. A silicon/molybdenum multilayer structure has been widely adopted for the EUV reflective masks. According to the ITRS, defect inspection is essential to the manufacture of defect-free mask blanks. Existing non-actinic defect inspection or defect review tool may fail to detect defects buried deep in the multilayer stacks because of rapid intensity attenuation through penetration. Actinic inspection in which incident light penetrates deep into the multilayer stacks is well accepted to be sufficient. However, it is difficult for EUV optics to achieve high numerical aperture (NA), thereby resolution is limited. A non-imaging coherent scatterometry microscope (CSM) with a less complicated optics-detector geometry can achieve higher spatial resolution by simply increasing measurement solid angle. Such kind of lensless system is aberration free and thus in theory the resolution is only diffraction limited. A coherent diffraction imaging (CDI) technique is used to reconstruct the mask image from its diffraction signal. Since the pupil image contains only intensity information of the diffraction signal, an iterative phase retrieval process is usually required. Most algorithms developed up to date remain too computational intensive for full-mask inspection because the iterative phase retrieval process has a large number of unknown while providing redundant information. In this work, a new inspection method for such kind of lensless system is proposed. CDI is replaced with a direct defect feature extraction from the diffraction signal. This leads to a small number of unknown corresponding to only the key defect features. Therefore, the computation complexity can be significantly reduced. Preliminary simulation results indicate that even a defect with 4 nm full width at half maximum (FWHM) and 0.5 nm height referring to the 11 nm half-pitch node EUV mask blank defect requirement are detectable in about 20 minutes of computation time. In addition, with statistical averaging, the method has a good robustness with respect to some type of system noise. As the phase information remains lost, defect location determination is still difficult since a shift in defect location corresponds to a shift in phase of the diffraction signal. Even so, the theoretical difficulty can become technically manageable by an inspection strategy combining fast scan and detail inspection. The fast scan judges the existence of defect in the beam spot with a relatively simple criterion. If the deviation between a local diffraction signal and the ideal exceeds a predetermined threshold, this beam spot location is to be marked as suspicious. In detail inspection, other inspection tools such as Atomic Force Microscope (AFM) can be used to assist regular defect extraction by accurately locating defect within a small metrology window of only a few microns. Moreover, relative location among multiple defects within the metrology window can be estimated with defect feature extraction, which indicates that locating one defect is sufficient to reveal all other defect locations. In return, it can accelerate the defect localization process drastically by saving AFM scan time.