O.R. Applications
Optimization of TQFP molding process using neuro-fuzzy-GA approach

https://doi.org/10.1016/S0377-2217(02)00258-8Get rights and content

Abstract

This paper focuses on an integrated optimization problem that involves multiple qualitative and quantitative responses in the thin quad flat pack (TQFP) molding process. A fuzzy quality loss function (FQLF) is first applied to the qualitative responses, since the molding defects cannot be simply represented by the relationship between molding conditions and mathematical models. Neural network is then used to provide a nonlinear relationship between process parameters and responses. A genetic algorithm together with exponential desirability function is employed to determine the optimal parameter setting for TQFP encapsulation. The proposed method was implemented in a semiconductor assembly factory in Taiwan. The results from this study have proved the feasibility of the proposed approach.

Introduction

The manufacturing of integrated circuits has become the heart of the electronic industry that is now the second largest basic industry (behind agriculture) and the fastest expanding manufacturing industry in the world. In the past, product is the only determinant of the profitability of a semiconductor company. However, over the last decade, the ever-increasing competition had led to the need for IC companies to also be able to manufacture their products in an efficient and cost effective manner. Increasingly, these companies have turned to data intensive operational modeling and analysis tools and techniques because of their potential to significantly improve the bottom line performance [14].

Recently, IC packages have been diversified due to increasing handy or high performance electronic applications, which require smaller body, lightweight, and high I/O connection. Thin quad flat pack packages (TQFP) are applied to accommodate higher I/Os and faster production cycle times. The TQFP assembly process includes wafer mounting, die sawing, die bonding, wire bonding, molding, marking, plating, trimming and forming, and electrical functional testing.

This study is presents an integrated optimization approach of neural networks, generic algorithms, exponential desirability function, and fuzzy theory for solving multiple qualitative and quantitative response problems in the molding process of TQFP.

Section snippets

The molding process of TQFP

TQFP packages are plastic die encasements with lead contact distribution around the perimeters of the packages and can be referred to as “Gull Wing” packages due to the shape of the very fine contact leads (≦0.5 mm pitches) (Fig. 1). TQFP can be seen in a wide area of applications includes mobile communications, portable consumer electronics, portable computers, and PCMCIA cards with pin counts ranging from 44 leads to 256.

The TQFP molding process is used to provide mechanical support, connect

Background information

This section will briefly introduce some portions of fuzzy quality loss function, neural network, genetic algorithms and exponential desirability function used in this study.

Proposed optimization procedure

This study proposes an integrated optimizing algorithm of the parameter settings in the TQFP molding process that involves multiple qualitative and quantitative responses. The proposed approach consists of three major stages. The first stage of the procedure defines the linguistic fuzzy term, defective category, and membership function for each qualitative response. The value of fuzzy quality loss function is then computed in the organized experiment. The next stage involves using a BP network

Conduction of the experiment

This study focuses on three types of TQFP molding defects, i.e., void, resin bleed and warpage, which are most often encountered during the operation. In order to optimize the molding process with respect to each response, an engineering experiment on the 1.4 mm TQFP100 molding process is conducted. Table 3 lists the process parameters and values for each level. Thirty-six trials (36 strips of lead frames) are conducted by a well-structured orthogonal array L18.

A universal set of uU

Confirmation experiment and benchmark

Many practitioners have applied Taguchi’s approach and sound engineering knowledge with experience to tackling multiresponse problem [12]. This study finally conducted a comparison between the Taguchi method and the proposed approach under the optimum conditions. The analysis of the 18 original trials shown in Table 3 by using the Taguchi method results in the optimal settings for six control factors are as listed in Table 9. For benchmarking purpose, the Taguchi method and the proposed

Conclusion

This study is presents an integrated optimization approach of neural networks, generic algorithms, exponential desirability function, and fuzzy theory for solving multiple qualitative response problems, which has received little attention among the engineering fraternity. Fuzzy set theory incorporates imprecision and subjectivity into the model formulation and solution process. The method of formalizing linguistic evaluation based on fuzzy sets proposed by Zadeh [2], [15] is used in this study

Acknowledgements

The authors would like to thank the National Science Council, Taiwan, ROC for partially supporting this research under Contract No. NSC 91-2213-E-159-011. The authors acknowledge Mr. David Chen, the Product Manager of TQFP Group of ADPrecision Taiwan Ltd., for sharing his expertise and support during this study.

References (15)

  • A.E. Ames et al.

    Quality loss functions for optimization across multiple responses surface

    Journal of Quality Technology

    (1997)
  • R. Bellman et al.

    Decision-making in a fuzzy environment

    Management Science

    (1970)
  • E. Castillo et al.

    Modified desirability functions for multiple response optimization

    Journal of Quality Technology

    (1996)
  • D.E. Goldberg

    Genetic Algorithms in Search, Optimisation, and Machine Learning

    (1989)
  • K.L. Hsieh, Applying neural networks and fuzzy sets to achieving the off-line quality optimization, Ph.D. Dissertation,...
  • C.M. Hsu et al.

    Multi-objective machine-component grouping in cellular manufacturing: A genetic algorithm

    Production Planning and Control

    (1998)
  • K.J. Kim et al.

    Simultaneous optimization of mechanical properties of steel by maximizing exponential desirability functions

    Applied Statistics

    (2000)
There are more references available in the full text version of this article.

Cited by (25)

  • Optimization of pin through hole connector in thermal fluid-structure interaction analysis of wave soldering process using response surface methodology

    2015, Simulation Modelling Practice and Theory
    Citation Excerpt :

    Therefore, optimization of the wave soldering process is important to optimize each independent variable, thereby eliminating defects and maintaining solder joint reliability. An optimization is a fundamental process that is systematically applied in most engineering applications by using various optimization techniques such as Taguchi method [1,2], response surface methodology (RSM) [3], artificial neural networks [4], genetic algorithm [5] and neuro-fuzzy genetic algorithm [6]. Optimization often produces solutions that are better than the conventional solution.

  • Optimization of IC encapsulation considering fluid/structure interaction using response surface methodology

    2012, Simulation Modelling Practice and Theory
    Citation Excerpt :

    RSM had been utilized for the investigation of the ball grid array (BGA) package by van Driel et al. [5] using the three-dimensional finite element (3D-FE) virtual prototyping. Besides, the optimal process parameter setting of the thin quad flat package (TQFP) [4] and plastic dual-inline package (PDIP) [6] moulding process was achieved through the neuro-fuzzy-GA approach and Taguchi method. Chen et al. [7] used Taguchi method to optimize the design of plastic quad flat packages (PQFP) in the curing process.

  • A novel approach for optimization of correlated multiple responses based on desirability function and fuzzy logics

    2012, Neurocomputing
    Citation Excerpt :

    In this method, similar to Chang’s [21] work, optimization phase is performed by GA. Chiang and Su [23] focused on an optimization problem that involves multiple qualitative and quantitative responses in the thin quad flat pack (TQFP) modeling process. First a fuzzy quality loss function is performed on qualitative responses.

  • Profiling effects in industrial data mining by non-parametric DOE methods: An application on screening checkweighing systems in packaging operations

    2012, European Journal of Operational Research
    Citation Excerpt :

    Nevertheless, data handling with the SNR operator has surfaced several limitations (Box, 1988). The data mining capabilities of Taguchi experimental designs have been further developed and applied in intensive expert-system programs where neuro-fuzzy and genetic algorithms are employed to aid in solving difficult replicated OA problems (Chang, 2008; Chiang and Su, 2003; Tan and Tang, 2001). From an operational research point of view, Taguchi theory has been applied to a wide spectrum of subjects ranging from human resource selection optimization to supply chain design improvement (Tiwari et al., 2010; Tsai et al., 2003).

  • Optimization of tile manufacturing process using particle swarm optimization

    2011, Swarm and Evolutionary Computation
    Citation Excerpt :

    However, these studies share a number of associated problems due to various limitations of GA which include low convergence efficacy and difficulty in finding proper value for parameters such as population size, crossover rate, and mutation rate. Chiang and Su [7] and Hsu et al. [8] proposed approaches which integrate neural networks, exponential desirability functions, and GA to solve process parameter design problems for the electronic-circuit and fiber-optic industries, respectively. The desirability function is an intuitive and simple approach for converting a multi-objective optimization problem into a single objective problem.

View all citing articles on Scopus
View full text