Lee-Ing Tong

Multi-response robust design by principal component analysis

Abstract Most previous Taguchi method applications have only addressed a single-response problem. However, more than one correlated response normally occurs in a manufactured product. The multi-response problem has received only limited attention. In this work, we propose an effective procedure on the basis of principal component analysis (PCA) to optimize the multi-response problems in the Taguchi method. With the PCA, a set of original responses can be transformed into a set of uncorrelated components. Therefore, the conflict for determining the optimal settings of the design parameters for the multi-response problems can be reduced. Two case studies are evaluated, indicating that the proposed procedure yields a satisfactory result.

The optimization of multi‐response problems in the Taguchi method

The Taguchi method is the conventional approach used in off‐line quality control. However, most previous Taguchi method applications have dealt only with a single‐response problem. The multi‐response problem has received only limited attention. Proposes an effective procedure on the basis of the quality loss of each response so as to achieve the optimization on multi‐response problems in the Taguchi method. The procedure is a universal approach which can simultaneously deal with continuous and discrete data. Evaluates a plasma‐enhanced chemical vapour deposition (PECVD) process experiment and a case study, indicating that the proposed procedure yields a satisfactory result.

Optimization of multiple quality responses involving qualitative and quantitative characteristics in IC manufacturing using neural networks

Abstract The optimization of product or process quality profoundly influences a manufacturer. Most studies have focused primarily on optimizing a quantitative (or qualitative) quality response, while others have concentrated on optimizing multiple quantitative quality responses. However, optimizing multiple responses involving both qualitative and quantitative characteristics have scarcely been mentioned, largely owing to the inability to directly apply conventional optimization techniques. In this study, we present a novel approach based on artificial neural networks (ANNs) to simultaneously optimize multiple responses including both qualitative and quantitative quality characteristics. Two neural networks are constructed: one for determining the ideal parameter settings and the other for estimating the values of the multiple quality characteristics. In addition, a numerical example from an ion implantation process employed by a Taiwan IC fabrication manufacturer demonstrates the proposed approach’s effectiveness.

Statistical testing for assessing the performance of lifetime index of electronic components with exponential distribution

The electronics industry has heavily prioritized enhancing the quality, lifetime and conforming rate (conforming to specifications) of electronic components. Various methods have been developed for assessing quality performance. In practice, process capability indices (PCIs) are used as a means of measuring process potential and performance. Moreover, most PCIs have been developed or investigated under the assumption that electronic components have a lifetime with a normal distribution. However, PCIs for non‐normal distributions have seldom been discussed. Nevertheless, the lifetime of electronic components generally may possess an exponential, gamma or Weibull distribution and so forth. Under an exponential distribution, some properties of the PCIs and their estimators differ from those in a normal distribution. To utilize the PCIs more reasonably and accurately in assessing the lifetime performance of electronic components, this study constructs a uniformly minimum variance unbiased (UMVU) estimator of their lifetime performance index under an exponential distribution. The UMVU estimator of the lifetime performance index is then utilized to develop the hypothesis testing procedure. The purchasers can then employ the testing procedure to determine whether the lifetime of the electronic components adheres to the required level. Manufacturers can also utilize this procedure to enhance process capability.

Lower confidence limits of process capability indices for non‐normal process distributions

When the process probability distribution is non‐normal or is unknown, the process mean and standard deviation may not properly describe the distribution’s shape. Consequently, the traditional process capability indices (PCI) Cp, Cpk, Cpm and Cpmk cannot express the actual process capability. This paper presents a procedure to construct lower confidence limits for PCIs when the process distribution is unknown. First, the order statistics are utilized to find the estimators of Cp, Cpk, Cpm and Cpmk. Bootstrap simulation method is then utilized to construct the lower confidence limits of PCIs, thereby allowing the process’s capability to be evaluated. A numerical example demonstrates the effectiveness of the proposed procedure.

Forecasting field failure data for repairable systems using neural networks and SARIMA model

Purpose – To propose an accurate product reliability prediction model in order to enhance product quality and reduce product costs.Design/methodology/approach – This study proposes a method for analysing and forecasting field failure data for repairable systems. The novel method constructs a predictive model by combining the seasonal autoregressive integrated‐moving average (SARIMA) method and neural network model.Findings – Current methods for analysing and forecasting field failure data for repairable systems do not consider the seasonal effect in the data. The proposed method can not only analyse the trends and seasonal vibration of the data, but can also forecast the short‐ and long‐term reliability of the system based on only a small amount of historical data.Research limitations/implications – This study adopts only real failure data from an electronic system to verify the feasibility and effectiveness of the proposed method. Future research may use other products failure data to verify the propose...

Using a neural network-based approach to predict the wafer yield in integrated circuit manufacturing

In integrated circuit (IC) manufacturing, defects on wafer tend to cluster. As the wafer size increases, the clustering phenomenon of the defects becomes increasingly apparent. When the conventional Poisson yield model is used, the clustered defects frequently cause erroneous results. In this study, we propose a neural network-based approach to predict the wafer yield in IC manufacturing. The proposed approach can reduce the phenomenon of the erroneous predictions caused by the clustered defects. A case study is also presented, demonstrating the effectiveness of the proposed approach. In addition, the proposed approach can be written as a computer software to accurately predict the wafer yield in IC manufacturing.

A non‐parametric method for experimental analysis with censored data

Considers that, occasionally, only part of an experiment can be completed owing to some uncontrollable causes such as the damage to the instrument, power failure during the experiment, and time and cost limitations. States that such incomplete data are generally referred to as censored data. Shows that conventional approaches for analysis of censored data are computationally complicated and often difficult to explain to practitioners. In this work, an effective procedure based on the rank transformation of the responses and the regression analysis is proposed for analysing an experiment with singly censored data. Proposes the procedure is simpler than conventional methods such as maximum likelihood estimation and Taguchi’s minute accumulating analysis. Verifies the proposed procedure by a numerical example.

A neural network-based procedure for the process monitoring of clustered defects in integrated circuit fabrication

In integrated circuit (IC) fabrication, a wafers defects tend to cluster. As the wafer size increases, the clustering phenomenon of the defects becomes increasingly apparent. When the conventional control chart (c chart) is used, the clustered defects frequently cause many false alarms. In this study, we propose a neural network-based procedure for the process monitoring of clustered defects in IC fabrication. The proposed procedure can reduce the phenomenon of the false alarms caused by the clustered defects. A case study is also presented to show the effectiveness of the proposed procedure.

Average type‐II censoring times for two‐parameter Pareto and Rayleigh distributions

Reliability engineers must not only consider the consumption of energy, capital and material resources, but also seek more economic means of completing experiments effectively. This study derives formulae for computing ratios of expected type‐II censoring times and expected complete sampling times when the lifetime adheres to two‐parameter Pareto and Rayleigh distributions. Utilizing such formulae allows the construction of tables providing information about how much experiment time can be saved by employing a type‐II censoring plan instead of a complete sampling plan. Engineers can employ the proposed tables to determine the censoring number, the initial sample size and the other relevant parameters for reducing the total experiment time. Illustrative examples demonstrate the effectiveness of the proposed procedure.