Nowadays, more attention has been placed on cost reductions and yield enhancement in the semiconductor industry. During the manufacturing process, a considerable amount of sensor data called Status Variables Identification (SVID) is collected by sensors embedded in advanced machines. This data is a valuable source for data-driven automatic fault detection and diagnosis at an early manufacturing stage to maintain competitive advantages. However, wafer processing times vary slightly from wafer to wafer, resulting in variable-length signal data. The conventional approaches use much condensed data called FDC data made by manually designed feature extraction. Or, recent deep learning approaches assume that all wafers have the same processing time, which is impotent to the variable-length SVID. To detect and diagnose faults directly from the variable-length SVID, we propose a self-attentive convolutional neural network. In experiments using real-world data from a semiconductor manufacturer, the proposed model outperformed other deep learning models with less training time and showed robustness at different sequence lengths. Compared to FDC data, SVID data showed better fault detection performance. Without manually investigating the lengthy sensor signals, abnormal sensor value patterns were found at the time specified by the model.

Wafer maps provide important information for engineers for detecting root causes of failure in a semiconductor manufacturing process. Thus, there has been active research into the automation of wafer map pattern classification. With recent advances in deep learning, a convolutional neural network (CNN) has yielded state-of-the-art performance in wafer map pattern classification. Because a large amount of labeled training data is required, experienced engineers need to annotate large quantities of wafer maps manually which is costly. To construct a well-performing CNN model with a lower labeling cost, we propose a cost-effective wafer map pattern classification system based on the active learning of a CNN. In the system, a CNN model is constructed based on four main steps: uncertainty estimation, query wafer selection, query wafer labeling, and model update. By repetitively performing these steps, the performance of the CNN model is gradually and effectively increased. We compared several methods for uncertainty estimation and query wafer selection in our system. We demonstrated the effectiveness of the proposed system through experiments using real-world data from a semiconductor manufacturer.

Purpose: Quality management of products is an important part of manufacturing process. One way to manage and assure product quality is to use machine learning algorithms based on relationship among various process steps. The purpose of this paper is to integrate manufacturing, inspection and after-sales service data to make full use of machine learning algorithms for estimating the products’ quality in a supervised fashion. Proposed frameworks and methods are applied to actual data associated with heavy machinery engines.Design/methodology/approach: By following Lenzerini’s formula, manufacturing, inspection and after-sales service data from various sources are integrated. The after-sales service data are used to label each engine as normal or abnormal. In this study, one-class classification algorithms are used due to class imbalance problem. To address multi-dimensionality of time series data, the symbolic aggregate approximation algorithm is used for data segmentation. Then, binary genetic algorithm-based wrapper approach is applied to segmented data to find the optimal feature subset.Findings: By employing machine learning-based anomaly detection models, an anomaly score for each engine is calculated. Experimental results show that the proposed method can detect defective engines with a high probability before they are shipped.Originality/value: Through data integration, the actual customer-perceived quality from after-sales service is linked to data from manufacturing and inspection process. In terms of business application, data integration and machine learning-based anomaly detection can help manufacturers establish quality management policies that reflect the actual customer-perceived quality by predicting defective engines.