Machine learning-based novelty detection for faulty wafer detection in semiconductor manufacturing

Since semiconductor manufacturing consists of hundreds of processes, a faulty wafer detection system, which allows for earlier detection of faulty wafers, is required. statistical process control (SPC) and virtual metrology (VM) have been used to detect faulty wafers. However, there are some limitations in that SPC requires linear, unimodal and single variable data and VM underestimates the deviations of predictors. In this paper, seven different machine learning-based novelty detection methods were employed to detect faulty wafers. The models were trained with Fault Detection and Classification (FDC) data to detect wafers having faulty metrology values. The real world semiconductor manufacturing data collected from a semiconductor fab were tested. Since the real world data have more than 150 input variables, we employed three different dimensionality reduction methods. The experimental results showed a high True Positive Rate (TPR). These results are promising enough to warrant further study.     

A literature review and classification of recommender systems research

Recommender systems have become an important research field since the emergence of the first paper on collaborative filtering in the mid-1990s. Although academic research on recommender systems has increased significantly over the past 10 years, there are deficiencies in the comprehensive literature review and classification of that research. For that reason, we reviewed 210 articles on recommender systems from 46 journals published between 2001 and 2010, and then classified those by the year of publication, the journals in which they appeared, their application fields, and their data mining techniques. The 210 articles are categorized into eight application fields (books, documents, images, movie, music, shopping, TV programs, and others) and eight data mining techniques (association rule, clustering, decision tree, k-nearest neighbor, link analysis, neural network, regression, and other heuristic methods). Our research provides information about trends in recommender systems research by examining the publication years of the articles, and provides practitioners and researchers with insight and future direction on recommender systems. We hope that this paper helps anyone who is interested in recommender systems research with insight for future research direction.     

Uniformly subsampled ensemble (USE) for churn management: Theory and implementation

The present paper explores the possible application of a new ensemble model. The model, which is based on multiple SVM classifiers, is employed to address churner identification problems in the mobile telecommunication industry, a sector in which the role of customer retention program becomes increasingly important due to its very competitive business environment. In particular, the current study introduces a uniformly subsampled ensemble (USE) model of SVM classifiers, not only to reduce the computational complexity of large-scale data, but also to boost the reliability and accuracy of calibrated models on data sets with highly skewed class distributions. According to our experiments, the performance of the USE SVM model is superior compared to all single and ensemble models. It is more scalable than well-known ensemble models as well.     

Localizing slab identification numbers in factory scene images

This paper deals with algorithms for text localization and character segmentation in images for process automation in the steel-making industry. Each character which comprises slab identification numbers may be corrupted severely before it is captured by network cameras. Therefore, proper processing is required to localize the target texts successfully. In this paper, we propose (1) a method to evaluate the closeness of an edge patch to the form of a closed contour, (2) an edge inspection method to determine character colors and estimate font thickness, and (3) three reasonable binarization methods to increase the performance of the algorithm for the detection of the left and right boundaries of the text rectangle. The experimental results show that the proposed algorithms are reliable.     

Measuring the success of retention management models built on churn probability,retention probability,and expected yearly revenues

We claim that often marketers have not all the information to develop various marketing campaign models. For example, marketers may have sufficient information to build a model for predicting possible churners, while they may have no clues of which customers are most likely to accept a retention campaign. In this paper, we first show that the information useful for a successful churner prediction model alone is not sufficient to develop a successful retention marketing program. In such a case, we claim that only theory-based simulation approach is feasible. In particular, it is claimed that optimal retention management models should consider not only churn probability but also retention probability and expected revenues from target customers. To validate our claim, we develop and compare five retention management models based on churn probability, retention probability, expected revenues, and combination of these models along with different evaluation metrics. Our experimental results show that the retention management model with the highest accuracy in predicting possible churners is not necessarily optimal because it does not consider the probability of accepting retention promotions. In contrast, the retention management model based on both churn and retention probability is the best in terms of predicting customers who are most likely to positively respond to retention promotions. Ultimately, the model based on expected yearly revenue of customers accrues the highest revenues across most target points, making it the best model out of five churn management models.     

Removal efficiencies of endocrine disrupting chemicals by coagulation/flocculation, ozonation, powdered/granular activated carbon adsorption, and chlorination

Removal efficiencies of endocrine disrupting chemicals (EDCs), bisphenol A and nonylphenol, during various types of water treatment processes were evaluated extensively using laboratory- and pilot-scale experiments. The specific processes of interest were coagulation/flocculation sedimentation/filtration (conventional water treatment process), powdered activated carbon (PAC), granular activated carbon (GAC), ozonation and chlorination. Batch sorption tests, coagulation tests, and ozone oxidation tests were also performed at higher concentrations with 14 EDCs including bisphenol A. The conventional water treatment process had very low removal efficiencies (0 to 7%) for all the EDCs except DEHP, DBP and DEP that were removed by 53%, 49%, and 46%, respectively. Ozonation at 1 mgO₃/ L removed 60% of bisphenol A and 89% of nonylphenol, while chlorination at 1 mg/L removed 58% and 5%, respectively. When ozone and chlorine doses were 4 and 5 mg/L, respectively, both EDCs were not detected. PAC removal efficiencies ranged from 15% to 40% at 3 to 10 mg/L of PAC with a contact time of 15 minutes. In the high concentration batch sorption tests, EDC removal efficiencies by PAC were closely related to octanol-water partition coefficient (K_ow). GAC adsorption was very effective water treatment process. The type and service time of GAC did not affect EDC removal efficiencies. The combination of ozonation and GAC in series appears to remove EDCs effectively to safe levels while conventional water treatment could not.     

Enhanced electrochemical properties of SnO2 anode by AlPO4 coating

A SnO₂ anode material undergoes severe capacity loss, which is mainly associated with cracking/crumbling of the material by the large volume change between the Li_xSn and Sn phases, and the intensive reactions with the electrolyte solution. However, AlPO₄ nanoparticle coating showed drastically improved electrochemical properties with decreased surface cracks. The AlPO₄-coated SnO₂ exhibited a capacity of 781 mAh/g, approaching its theoretical capacity at the first cycle, with 44% capacity retention after 15 cycles between 2.5 and 0 V at a relatively high C rate of 105 mA/g. In contrast, the bare SnO₂ showed an initial capacity of 680 mAh/g, with only 8% capacity retention after 15 cycles.     

Tracking of moving interfaces in sedimentation processes using electrical impedance tomography

Sedimentation monitoring is widely used to control and optimize industrial processes. In this paper we propose a novel computational method for sedimentation monitoring using electrical impedance tomography (EIT). EIT measurements consist of electric current and voltage measurements that are made on the surface of the sedimentation tank and therefore they do not interfere with the sedimentation process. The proposed computational method is based on shape estimation and state estimation formulation of the EIT problem. The sedimentation is parameterized by the locations of the phase interfaces and conductivities of the phase layers. Three different evolution models for the state parameters are considered and the state estimates are computed using the extended Kalman filter algorithm. The performance of the method and the models are evaluated using simulated data from a six electrode EIT measurement configuration. From the results a promising performance of the method can be seen.