Substrate defect inspection apparatus, substrate defect inspection method and storage medium

Abstract

The present invention provides a technique capable of appropriately classifying defects of a substrate on the basis of an image of an object to be inspected obtained by imaging the substrate. A solution is to obtain binarized data including a defect region by performing deep learning and a two-class classification unit (3) on a captured image obtained by capturing an image of a substrate. A rectangular color image including a defect region is generated from the binarized data, and the type of the defect is inferred using deep learning based on the image. Furthermore, attributes such as the shape of the defect region and the position on the wafer (W) are determined on the basis of the binarized data, and the type of the defect is estimated using a rule base on the basis of the information.And comprehensively judging the type of the defect based on an inference result of deep learning and an inference result of the rule base.

Description

technical field [0001] The present invention relates to a technique for determining what kind of defect has occurred on a substrate using image data obtained by photographing a substrate. Background technique [0002] In the photolithography step in the manufacturing process of a semiconductor device, a resist film is formed by applying a resist to the surface of a semiconductor wafer (hereinafter referred to as wafer) as a substrate, exposing the resist film, developing , to form a resist pattern. The wafer W after a series of steps is inspected by an inspection device to see whether a good resist pattern is formed on the wafer surface, whether there is a scratch or foreign matter is attached, and the like. Such a defect is called a defect, and the defect has various forms depending on the cause thereof. Therefore, from the viewpoint of semiconductor device manufacturing management, the inspection apparatus is required not only to inspect whether or not a defect occurs on...

AI Technical Summary

Problems solved by technology

Defect classification based on deep learning is learned by using teaching data. However, in the case of insufficient learning, etc., misclassification cannot be avoided.

Images