Proceedings Article | 5 April 2012
Proc. SPIE. 8324, Metrology, Inspection, and Process Control for Microlithography XXVI
KEYWORDS: Lithography, Diffraction, Metrology, Data modeling, Calibration, Image processing, Scatterometry, Scatter measurement, Process modeling, Diffraction gratings
Optical scatterometry is crucial to advanced nodes due to its ability of non-destructively and rapidly retrieving accurate
3D profile information.<sup>1, 2, 3</sup> In recent years, an angle-resolved polarized reflectometry-based scatterometry which can
measure critical dimensions, overlay, and focus in single shot has been developed.<sup>4, 6, 20</sup> In principle, a microscope
objective collects diffracted light, and pupil images are collected by a detector. For its application of calibrating
lithography models, the pupil images are fit to a database pre-characterized usually by rigorous electromagnetic
simulation to estimate dimensional parameters of developed resist profiles.<sup>5</sup> The estimated dimensional parameters can
then be used for lithography model calibration. In this work, we propose a new method which directly utilizes the pupil
images to calibrate lithography models without needing dimensional parameter estimation. To test its feasibility and
effectiveness by numerical simulation, a reference lithography process model is first constructed with a set of parameter
values complying with ITRS. A to-be-calibrated process model is initialized with a different set of parameter values from
those of the reference model. Rigorous electromagnetic simulation is used to obtain the pupil images of the developed
resist profiles predicted by both process models. An optimization algorithm iteratively reduces the difference between
the pupil images by adjusting the set of parameter values of the to-be-calibrated process model until the pupil image
difference satisfies a predefined converging criterion. This method can be used to calibrate both rigorous first-principle
models for process and equipment development and monitoring, and fast kernel-based models for full-chip proximity
effect simulation and correction. Preliminary studies with both 1D and 2D aperiodic and periodic layouts indicate that
when the pupil image difference is minimized, the lithography model can be accurately calibrated.
