Unsupervised spatial pattern classification of electrical-wafer-sorting maps in semiconductor manufacturing

Multi-step ART1 algorithm for recognition of defect patterns on semiconductor wafers

“…Chen and Liu (2000) compared the performance of ART1 with that of SOMs and reported that the ART1 is superior to the SOMs in terms of pattern recognition performance on defect maps. Conversely, Di Palma et al (2005) reported that the ART1 technique is not adequate for handling a large number of wafers, largely due to use of an 'AND' logic, while the SOM architecture provides very sensible classification for the simulated and real wafer map data, where they used rather simple failure models to generate electrical failures over a wafer. In the present research, we propose a modified version of the ART1 to overcome the weakness of the technique, while mainly taking the advantage of the ART1 on defects classifications.…”

Defect spatial pattern recognition using a hybrid SOM–SVM approach in semiconductor manufacturing

“…In addition, they also employed the filtering algorithm to discard those wafermaps without systematic patterns and then introduced the particle swarm optimization algorithm to build a RBF neural network that could solve the defect classification problem. Palma, Nicolao, Miraglia, Pasquinetti, and Piccinini (2005) compared the unsupervised neural network classifier, namely, SOM and ART1, to validate and recognize the spatial pattern on a wafer. They concluded that ART1 was not adequate, whereas SOM provided completely satisfactory results including a visually effective representation of the pattern's spatial probability classes.…”

“…This spatial measure can be identified in the spatial effects in such processes and parameters that influence the spatial clustering of functional or non-functional chips on the wafer-map (Taam & Hamada, 1993). Other discussions about spatial statistics may be found in Palma et al (2005). A wide range of bin classification is employed during the testing process.…”

“…In fact, increasing the number of clusters will usually lead to an improvement in the chosen measure due to the identification of a new cluster or a split in a true cluster (Palma et al, 2005). Herein, we conduct the unsupervised neural network SOM combined with SVM classifier for validation.…”

Adaptive burn-in time decision system based on pattern recognition for intelligent reliability control

“…According to the research of Palma, Nicolao, Miraglia, Pasquinetti, and Piccinini (2005), the SOM classifies better and more efficiently compared to the Adaptive Resonance Theory Network 1, which uses unsupervised learning algorithm. For the active decision of burn-in time based on the pattern recognition using SOM, we present a decision support system based on intelligent agent technology called adaptive burn-in time decision system (ABITDS).…”