Ariel Ben-Porath

Automatic defect classification using topography map from SEM photometric stereo

As the industry moves to smaller design rules, shrinking process windows and shorter product lifecycles, the need for enhanced yield management methodology is increasing. Defect classification is required for identification and isolation of yield loss sources. Practice demonstrates that an operator relies on 3D information heavily while classifying defects. Therefore, Defect Topographic Map (DTM) information can enhance Automatic Defect Classification (ADC) capabilities dramatically. In the present article, we describe the manner in which reliable and rapid SEM measurements of defect topography characteristics increase the classifier ability to achieve fast identification of the exact process step at which a given defect was introduced. Special multiple perspective SEM imaging allows efficient application of the photometric stereo methods. Physical properties of a defect can be derived from the 3D by using straightforward computer vision algorithms. We will show several examples, from both production fabs and R&D lines, of instances where the depth map is essential in correctly partitioning the defects, thus reducing time to source and overall fab expenses due to defect excursions.

SEM review and discrete inspection of optically invisible defects in a production environment

Timely detection and prevention of defect excursions is the primary goal of any yield enhancement methodology in a wafer processing fab. Contamination killer defects are traditionally at the focus of such efforts. However, systematic process-induced defects are increasingly impacting the yield in the contemporary manufacturing environment, which is characterized by rapid modifications, constantly decreasing design rules and narrowing process windows. While the treatment of killer random defects remains important, the emphasis on yield-limiting problems is shifting to subtle treatment of killer random defects remains important, the emphasis on yield-limiting problems is shifting to subtle precursor phenomena, before the tool yield is severally damaged. Common light-scattering inspection is limited when applied to early symptoms of process deterioration, such as defects below 100nm or those lying in trenches. Electrical defects are even more challenging, since in most cases they are entirely undetectable by optical means. Disconnected metal plugs, under-etch of contact and via holes, implant variations undetectable by optical means. Disconnected metal plugs, under-etch of contact and via holes, implant variations on active area, gate to active area shorts, junction leakage - all of them can produce an electrical signature that can be detected with electron beam microscopy using Voltage Contrast imaging. In this paper, we describe a methodology implemented at the STMicroelectronics fab in Crolles, France, that addresses the above issues. SEM reviews and discrete inspections with a SEMVision cX were used for characterization of a variety of optically invisible defects on several critical inspection steps: Shallow Trench Isolation after etch and CMP, Polysilicon after etch, Via after etch, Tungsten plugs after CMP, Aluminum metal lines after etch and Copper metal lines after CMP.

Paradigm for selecting the optimum classifier in semiconductor automatic defect classification applications

The automatic classification of defects found on semiconductor wafers using a scanning electron microscope (SEM) is a complex task that involves many steps. The process includes re- detecting the defect, measuring attributes of the defect, and automatically assigning a classification. In many cases, especially during product ramp-up, and when multiple products are manufactured in the same line, there are few training examples for an automatic defect classification (ADC) system. This condition presents a problem for traditional supervised parametric and nonparametric learning techniques. In this paper we investigate the attributes of several approaches to ADC and compare their performance under a variety of available training data scenarios. We have selected to characterize the attributes and performance of a traditional K-nearest neighbor classifier, probabilistic neural network (PNN), and rule-based classifier in the context of SEM ADC. The PNN classifier is a nonparametric supervised classifier that is built around a radial basis function (RBF) neural network architecture. A basic summary of the PNN will be presented along with the generic strengths and weakness described in the literature and observed with actual semiconductor defect data. The PNN classifier is able to manage conditions such as non-convex class distributions and single class multiple clusters in feature space. A rule-based classifier producing built-in core classes provided by the Applied Materials SEMVision tool will be characterized in the context of both few examples and no examples. An extensive set of fab generated data is used to characterize the performance of these ADC approaches. Typical data sets contain from 30 to greater than 200. The number of classes in the data set range from 4 to more than 12. The conclusions reached from this analysis indicate that the strengths of each method are evident under specific conditions that are related to different stages within the VLSI yield curve, and to the number of different products in the line.