AUTOMATIC DRAM CELL POSITIONING SYSTEM

Abstract

The scanning electron microscope (SEM) is an essential tool for DRAM failure analyst to view failure sites and to investigate its failure mechanism. However, manually locating the failure site in the DRAM cell array using the SEM is a tedious task and prone to human error. An automatic failure site positioning system would facilitate the localization process. This project aims to develop an image-based system that simulates the manual operation of the failure analyst with a vision computer. By tracking the movement of the DRAM die and counting the number of patterns passed out of a detection window, the system can drive the SEM stage to position the DRAM failure site in the SEM. Two main algorithms, feature location and feature tracking, are developed to locate the position of the tracked pattern and to track the movement of the DRAM image, respectively. Cross correlation by template matching is the classical feature location technique that is widely used. Although it has optimal signal-to-noise ratio (SNR), it suffers from broad peak points. Since the Phase-Only Filter can produce much sharper peak points and images derived from a field emission SEM have relatively good SNR, the Phase-Only Filter is selected as the practical feature location algorithm in this project. A template is needed to identify the tracked pattern. The basic requirement for the template is that it should be unit cell of the DRAM cell array. The feature tracking algorithm is divided into normal feature tracking and feature re-tracking. While normal feature tracking only considers the tracking situation where the tracked pattern is within the detection window, the feature re-tracking algorithm copes with tracking situations where the tracked feature moves outside the detection window. According to different movement situations of the DRAM image, the feature re-tracking algorithm is further subdivided into normal feature re-tracking and abnormal feature re-tracking algorithms. Special consideration has also been given to irregular features encountered during the tracking process. This system has been successfully implemented on a DEC Alpha 400/500 UNIX workstation in C in an X environment. The system is proven to be reliable in a noisy environment with various image parameters, and its tracking speed is comparable to manual operation. Apart from DRAMs, this system is applicable to samples made up of repetitive patterns with distinguishable features.