A neural-network-based local-field-effect correction scheme for quantitative voltage contrast measurements in the scanning electron microscope

Abstract

A local-field-effect correction scheme, using a neural-network-based approach, is proposed for quantitative voltage contrast measurements (QVCM) in the scanning electron microscope (SEM). This technique showed some (though modest) improvement over an iterative correction scheme proposed previously. The correction technique also gives reasonably accurate voltage measurements on a multi-electrode test structure, even under low-extraction-field conditions for which local field effects are especially serious. The neural network employed is a back-propagation network with an adaptive learning rate to decrease the training time of the correction scheme. A momentum constant is also added to the back-propagation learning rule to minimize the chances of the network becoming stuck in a local minima of the error surface curve. The addition of momentum has a low-pass filtering effect on noise in the training data set and this could possibly account for the modest improvement in performance of this approach over the earlier iterative approach.