The present invention relates to a deep learning-based control variable setting device in a semiconductor vision inspection system, which is capable of automatically setting an optimal control value, and a method thereof. According to the present invention, the method comprises the steps of: receiving information about a plurality of control variables controlling operation of an image acquisition device; using the number of control variables included in the received information about the control variables and the number of variable values of the control variables to form a control variable set; receiving images acquired by the image acquisition device reflecting the variable values of the control variables corresponding to each element of the formed control variable set with respect to one or more vision inspection samples; and using a previously trained deep learning-based neural network to determine a final image among the received images, and acquiring final variable values of the control variable for acquiring the determined final image. Each element of the formed control variable set includes all of the control variables and two arbitrary elements different from each other of the formed control variable set are corresponded to have different variable values among one or more identical control variables.;COPYRIGHT KIPO 2019
展开▼