<title>SEM defect review and classification for semiconductor device manufacturing</title>

Abstract

Manufacturing of semiconductor devices is an extremely complicated, time consuming, multi-stage process. In order to maintain acceptable yield levels, the production line must be continuously monitored. Scanning Electron Microscopes (SEMs) are employed by all advanced wafer fabrication lines (fabs) for imaging and review of sub-micron defects which may result in faulty circuits. Designed to greatly enhance the functions of a defect review SEM, the Applied Materials SEMVision cX is a fully automated defect review and classification system, that is able to generate a novel type of defect image, and to use the information in these images for automatic defect classification. A unique technology that was developed for this purpose is Multiple Perspective SEM Imaging (MPSI ), an electron detection method in which multiple detectors that differ in their spectral and spatial response, are employed such that multiple images (perspectives) containing complementary information are generated simultaneously. Optical inspection tools, such as the Applied Materials Compass , first scan semiconductor wafers, and generate a list of suspected defect locations, with their approximate coordinates. At the defect review SEM, this information is used for driving to each of the approximate locations, imaging the vicinity of the suspected defect, and comparing this MPSI image to a reference image, which is typically taken from a neighboring die. This comparison is known as redetection, and results in precise location and sizing of the defect, as well as elimination of inspection false alarms. After the defect is found, a third image may be generated in a smaller field of view and increased resolution, to be used in the automatic classification algorithms. The Automatic Defect Classification (ADC) system contains a primary layer of rule based classification, and a secondary customization layer. In the rule based classifier, a predefined set of analysis algorithms is employed for each defect, categorizing it into one of several universal classes for semiconductor defects, so called Core Classes, such as a particle, a scratch, missing pattern and more. The customization layer may be used to further refine these into user-defined classes, according to requirements which maybe specific to the process, the product or local conventions.