Novel technique for aluminum thin film thickness measurement using top view SEM-EDX in conjunction with electron beam simulation and machine learning

Abstract

A novel method for determining aluminum thin film thickness using top view SEM and EDX measurements has been developed. Electron beam simulations are used as the reference training data to feed into a machine learning algorithm, which once trained can predict the thickness of the aluminum thin film from EDX characteristic x-ray count measurements for a set of three accelerating voltages. Unlike previous techniques which rely on a reference pure material sample or substrate signal to compare to, this method compares instead using ratios of EDX x-ray signals using different accelerating voltages. Since no substrate signal is required, the layer(s) below the aluminum thin film may be any material. High prediction accuracy was obtained for the training and test data for most data points, below 10 % for thicknesses above 40 nm on average, though some large errors remained. Investigation of the lateral dispersion of the incident electron beams showed that lateral dispersion increased with accelerating voltage. Since measurement of higher thicknesses requires higher accelerating voltages, the minimum feature size that can be accurately measured increases for higher thicknesses. Limitations include the requirement for aluminum to be the top layer, the requirement for consistency of beam current, low signal and excessive noise at low values of accelerating voltage, and the need to make many measurements at different voltages if the approximate range of the thin film thickness is not initially known.