Causal analysis of systematic spatial variation in optical lithography

The central theme of this thesis is the systematic analysis of the sources of variation in optical lithography. This work pursues data collection and data analysis techniques that are used to identify and model the various sources of variability in pattern transfer.; The first contribution of the thesis is that an automatic focus/exposure control method has been proposed based on digitized SEM scan using statistical feature extraction and neural network classification. Digitized CD-SEM traces are used to study die-to-die spatial variation and wafer-to-wafer or lot-to-lot temporal variation, induced by focus/exposure fluctuations for a 0.35 mum i-line process. Seventy-four perfect classification or ninety-six percent +/-5mJ match has been achieved for automatically detecting exposure, with 1sigma prediction accuracy approximately equal to 1/18 of exposure window; forty perfect classifications or eighty-three percent +/-0.1mum match has been achieved for automatic detecting defocus settings, with 1sigma prediction accuracy approximately equal to 20% of depth of focus window. This automatic focus/exposure control method can be extended to further extracting pattern profile information. Through intelligent data analysis techniques, more process information can be obtained besides routine CD measurements, without resorting to complicated modeling of the interaction of the electron beam with samples. The work is of practical interest and research for implementation is being conducted in industry.; The second contribution of this thesis is a new technique to measure full-field lens aberrations using printed linewidth patterns. The basic concept of this method is that the dependency of the linewidth on individual aberration terms can be approximated by a Taylor series expansion about aberration free imaging. This approximation is valid, since the aberration of a state of the art lithography system is typically small. The expansion can be conducted under different process conditions. Therefore the aberrations can be deduced by means of numerical analysis. The experiment of this part of work utilizes electrical test patterns for off-line analysis, combined with reticle measurements to decompose the linewidth variability to the contributions of the optics, the reticle, the resist, and other random sources for a 0.22 mum process using 248 nm DUV lithography. This data set captures the spatial distribution of CD variation across die as well as across wafer. It is assumed that the deterministic within field variation is a major variation component, which consumes a large portion of the error budget. Experimental results indicate that the across wafer variation is at the order of 2.7nm; while across field variation is about 5.9nm for isolated vertical feature for this standard 0.22mum process.; The third contribution of the thesis is a compact formulation of the mask error factor. Mask error magnification, as another important source of systematic spatial variation, has been studied theoretically. The rigorous formulation of mask error factor has been derived based on closed form aerial image calculation for coherent illumination condition. The results agree very well with first principle simulation and experiment. (Abstract shortened by UMI.)...