TOWARD A-PRIORI SELECTION OF ELLIPSOMETRY ANGLES AND WAVELENGTHS USING A HIGH-PERFORMANCE SEMANTIC DATABASE

Modern ellipsometers only barely resemble those of only a decade ago. In situ measurements are commonly spectroscopic and ex situ investigat ions frequently add an additional variable, incidence angle. In either case, the number of measurements for an analysis can be 10000 or more . Experience in a recent international round robin experiment showed n o particular agreement in the selection of angles and wavelengths. Sel ecting a fewer number of 'good' angles of incidence and wavelengths is important because at the same time it would reduce the measurement ti me, reduce computational load and improve the solutions. Currently we choose to say 'good' points have high 'resolution' (a change in the de sired parameter results in a measurable change in measured parameters) and a solvable 'condition' (the inverse Hessian matrix is well condit ioned for variably damped least squares). In this work, values of reso lution and condition are simulated for approximately 1000 points in wa velength-angle space for a large number of sample configurations and m aterials (over 10000). These results are stored in a high performance semantic database which can be more than a Gigabyte. Three-dimensional images of the data and movies over thickness, for example, can be pul led from the database in real time allowing selection of incidence ang les and wavelengths for an analysis. (C) 1998 Published by Elsevier Sc ience S.A.