Materials Science Forum Vol. 594

Abstract: Estimating the cycle time of every job in a semiconductor manufacturing factory is a critical task to the factory. Many recent studies have shown that pre-classifying a job before estimating the cycle time of the job was beneficial to the forecasting accuracy. However, most pre-classification approaches applied in this field could not absolutely classify jobs. Besides, whether the pre-classification approach combined with the subsequent forecasting approach was suitable for the data was questionable. For tackling these problems, an artificial neural network (ANN) approach that equally divides and post-classifies jobs is proposed in this study in which a job is post-classified by a BPN instead after the forecasting error is generated. In this novel way, only jobs which cycle time forecasts are the same accurate will be clustered into the same category, and the classification algorithm becomes tailored to the forecasting approach. For evaluating the effectiveness of the proposed methodology and to make comparison with some existing approaches, some data were collected from an actual semiconductor manufacturing factory. According to experimental results, the forecasting accuracy (measured with root mean squared error (RMSE)) of the proposed methodology was significantly better than those of the other approaches in most cases by achieving a 16%~44% (and an average of 29%) reduction in RMSE over the comparison basis – multiple-factor linear combination (MFLC). The effect of post-classification was also evident.

Abstract: Product yield is undoubtedly the most critical factor to the competitiveness of a semiconductor manufacturing factory. Therefore, evaluating the competitiveness of a semiconductor manufacturing factory based on the yields of all its products is a reasonable idea. For this purpose, a systematic procedure is established to evaluate the mid-term competitiveness of a semiconductor manufacturing factory in which the yield learning processes of all its products are considered simultaneously. After that, for products that are not competitive enough but have strong demand, a correction mechanism is designed to enhance their competitiveness at the expense of the other products that are competitive but have no market. The whole process is therefore called collaborative yield learning planning. To evaluate the advantages or disadvantages of the proposed methodology, it has been applied to the data collected from a semiconductor manufacturing factory. Experimental results supported the practicability and usefulness of the proposed collaborative yield learning planning approach.

Abstract: Engineering assets are fundamentally important to enterprises. Thus, making the best use of engineering assets attracts equipment and system engineers’ attention. The state-of-the-art researches contribute to asset condition monitoring, asset symptom diagnosis, asset health prognosis, and the integration of above knowledge. However, they still lack the combination with enterprise resources to determine the best maintenance/renewal time for the optimization of total enterprise benefits. Consequently, this paper proposes the integrated architectural framework, activity and process models of a multi-agent system called agent-based integrated engineering asset management (AIEAM) based on agent techniques to build collaborative environment for asset manager, diagnosis expert, prognosis expert and enterprise resource manager. An engineering asset management case (for repair and maintenance of automatic parking tower) applying the proposed architecture and models is depicted in the paper.

Abstract: An investigation about the robust estimation of a class of systems with noise coupling input saturation is presented in this study. In general, the existed estimation algorithm is based on the exactly known of the input saturation, but in fact, this is not always true in some practical cases because of the coupling of input saturations and input noises. For treating these kinds of coupling problems in the state estimations of systems, in this study, one fuzzy-based robust estimation algorithm is proposed. The proposed robust estimator includes two parts: firstly, a regression form fuzzy system that is adopted to approximate the unknown input saturation, and then, a robust estimator that combines the above fuzzy system by robust filter design concept for eliminating the effects of noises and modeling uncertainties is proposed. This combination of fuzzy approach and robust filtering technologies successfully offers one a more simple and practical method for treating the estimation problem of a class of systems that input saturations and input noises couple together.

Abstract: This research utilizes a 2D printer mechanism from a commercial available inkjet printer, combines with hardware and software to build up a new concept 3D printer. This 3D printer can receive printing information from personal computer, and setting the required layer thickness to stack those layers into a solid model. The largest build volume is 300x420x350 mm, print head resolution is as high as 1200x2400 dpi, and the life of printer head is longer than 12 weeks which is far longer than 4 weeks of Z Corp’s. There are many specifications are better than or equivalent to those of Z Corp’s Z-510 which is the state-of-art machine of this powder based rapid prototyping.