Elsevier

Microelectronics Reliability

Volume 81, February 2018, Pages 159-166
Microelectronics Reliability

Using GA-SVM for defect inspection of flip chips based on vibration signals

https://doi.org/10.1016/j.microrel.2017.12.032Get rights and content

Highlights

  • This paper presents a nondestructive method for defect detection of flip chips.

  • Thirty-four features were extracted from raw vibration signals.

  • Cross validation and genetic algorithms were utilized to optimize the parameters of SVM.

  • CV-SVM and GA-SVM were adopted for classification.

  • The results demonstrate the feasibility of GA-SVM for defect inspection.

Abstract

Flip chip technology has been widely used in IC packaging, and the combination of flip chip technology and solder joint interconnection technology has been utilized in the manufacturing of electronic devices universally. As the development of flip chip towards high density and ultra-fine pitch, the inspection of flip chips is confronted with great challenges. In this paper, we developed an intelligent system used for the detection of flip chips based on vibration. Thirty-four features including 18 time domain features and 16 frequency domain features were extracted from the raw vibration data. The support vector machine was employed to implement the recognition and classification of flip chips. In order to improve the classification accuracy of SVM, cross validation (CV) and genetic algorithm (GA) were utilized to optimize the parameters of SVM respectively. SVM, CV-SVM and GA-SVM were applied to classification separately and the results were obtained. By comparison, GA-SVM can recognize and classify the flip chips rapidly with high accuracy. Thus, GA-SVM is effective for the defect inspection of flip chips.

Introduction

In the early 1960s, flip chip technology was originated by IBM. Combined with solder joint interconnection technology, flip chip has been widely applied to the manufacturing of electronic devices. The bare die is turned over and the solder joints are put in order between the chip and substrate to guarantee the thermal conduction and electronic connection. Due to the faults occurring frequently such as fatigue failures and manufacturing defects of flip chips, nondestructive inspection of flip chips is significantly imperative [1], [2]. As the development of flip chip towards high density and ultra-fine pitch, some new requirements of packaging materials have appeared, such as lead free and low K, which bring great challenges to the nondestructive inspection of solder joints beneath chips [3], [4], [5]. Therefore, the technologies of electronic testing, infrared thermography, X-ray inspection and acoustic microscopy imaging are introduced for the detection of flip chips.

Electronic testing is utilized to detect the solder joints according to the change of resistance [6]. The flip chip is fixed on the pre-designed test board, and a probe contacted with the pre-designed board to put through small current is used for solder joint inspection. The pre-designed board is complex and each kind of flip chips need a board, which brings high expenditure and time consumption. Besides, any mechanical contact may lead to defective solder joints misrecognition [7]. According to the temperature change of the flip chip surface, infrared imaging technology has been applied to solder joint inspection [8]. Lu [9] realized the inspection of flip chips with missing solder joints in terms of the pulsed phase thermography. However, as the solder joint size decreases, the infrared device will be limited by the infrared wavelength. X-ray techniques were also applied to flip chip defect inspection, which contain X-ray photography and tomography. X-ray photography is an effective method on the inspection of solder joint voids and bridge connections, and a fuzzy rule-based system was proposed to recognize the defective solder joints. However, X-ray photography is difficult to detect the tiny cracks and laminations [10], [11]. X-ray tomography can inspect defects in vertical direction, but it is not suitable to be applied to on-line detection for the long time-consuming and harmful radiation.

Acoustic microscopy imaging (AMI) is widely utilized for flip chip detection and scanning acoustic microscope has been applied to manufacturing industry. Ultrasound is utilized to scan the sample interface and the internal conditions of the component are detected based on reflected waves [12]. The reflected wave shows the weak bonding joints and voids clearly [13], [14], [15]. When there are defects appearing in the periphery of flip chips, the inspection accuracy will be impacted by marginal effects. Semmens used high frequency ultrasound to detect flip chip faults and obtained the internal conditions of the component by analyzing the reflected wave [16]. Zhang [17] improved scanning microscopy and evaluated microelectronic packages by means of sparse representation. Neural network combined with cross-correlation method was employed to detect the defects in flip chip solder joints. Sebastian Brand introduced wavelet and pulse separation analysis to the ultrasonic inspection of solder joints [18]. High frequency ultrasonic scanning microscopy is very efficient for flip chip inspection, while the signal attenuation and diffraction between solder joints and laminated substrates are the major drawbacks, which influence the application for flip chips bonding to the laminated substrates. Recently, an inspection method for flip chips based on vibration analysis was applied extensively, which can detect the chips rapidly without destruction [19], [20], [21]. Laser ultrasonic inspection system was developed for flip chip inspection [22], [23], and finite element analysis, correlation coefficient, error ratio and modal analysis were also introduced to diagnose the defects.

In this paper, we developed an intelligent device used for the detection of flip chips. The device contains an ultrasonic transducer, a laser scanning vibrometer and a set of vibration analysis system. The ultrasonic transducer transmits ultrasounds to excite the flip chip, and the micro displacements of the chips are gauged by the laser scanning vibrometer. The relevant data acquired by the laser scanning vibrometer is processed by the vibration analysis system. We inspected good chips and chips with missing joints, and 34 features were extracted, including 18 time domain features and 16 frequency domain features. The support vector machine (SVM) was introduced for recognition and classification. For the improvement of the recognition accuracy, cross validation (CV) and genetic algorithm (GA) were employed to optimize the critical parameters of SVM.

The rest of the paper is organized as follows. Section 2 introduces the theoretical basis of the proposed method. Section 3 states the experimental investigation. In Section 4, we demonstrate the features extracted for recognition. We analyze and discuss the classification results in Section 5. The conclusion is depicted in Section 6.

Section snippets

Principle of support vector machine

Support vector machine (SVM) is widely used in pattern classification and nonlinear regression. The main idea of SVM is to establish a classification hyperplane serving as the decision surface, which maximizes the distance of the separation edge between the positive cases and the negative cases [24]. SVM is based on the theory of statistical learning and approximately fulfils the minimization of structure risks. The geometric margin can be defined as follows:yiwTxi+b1,i=1,,mwhere x is the

Experimental investigation

Fig. 2 shows the device used for the inspection of flip chips and the flip chip is fixed on vibration isolation table. First, the signal generator (Polytec, GEN-M2) produces period chirp signal to act as the excitation signal source. Then, the power amplifier (NJFNKJ, HFVA-62) magnifies the original signal to enhance its energy to guarantee that flip chips can be fully excited. After that, the capacitive air-coupled ultrasonic transducer (ULTRAN Group, CAP3) fixed in the manual X-Y motion stage

Feature extraction

For the thorough study of vibration signals of the flip chips, features that can represent the characteristics of flip chip need to be extracted to constitute the vector for data analysis. The features mainly contain information from waveforms in time domain and frequency domain. Time domain features are shown in Table 1, where x(n) represents the real time domain vibration signals and N is the number of sampling points. The k is an integer, xn¯k is the average of k sampling points and Nk is

Results and discussion

Each chip was inspected 55 times, and 55 vibration signals were captured. For eliminating abnormal interference, we selected 50 signals according to the mean and standard deviation of signal series and abandoned the others which have significant differences. Eventually, 800 data sets of raw vibration signals were acquired, including 150 data sets of good flip chips and 650 data sets of flip chips with missing joints. Feature extraction was conducted and 800 vectors including 34 features were

Conclusions

In this paper, we selected PB08 as the samples and ultrasound was employed to exciting flip chips. We extracted 34 features from the vibration signals, including 18 time domain features and 16 frequency domain features. SVM, CV-SVM and GA-SVM were utilized to train and testify the acquired data respectively. The corresponding classification accuracy reaches up to 82.33%, 91%, 92.67% separately, and the time expenditures of three methods are little. The comparison results indicate that SVM

Acknowledgments

The authors are grateful for the supports by the National Key Basic Research Special Fund of China (Grant No. 2015CB057205), the National Natural Science Foundation of China (Grant Nos. 51705203 and 51675250), the Natural Science Foundation of Jiangsu Province (Grant no. BK20160183) and the project funded by China Postdoctoral Science Foundation (Grant no. 2017M611690).

References (25)

  • P.A. Burdett et al.

    Techniques for the inspection of flip chip solder bonded devices

    Microelectron. Int.

    (1989)
  • T.I. Shih et al.

    Electrical characteristics for Sn–Ag–Cu solder bump with Ti/Ni/Cu under-bump metallization after temperature cycling tests

    J. Electron. Mater.

    (2006)
  • Cited by (41)

    • Surface defect characterization and depth identification of CFRP material by laser line scanning

      2022, NDT and E International
      Citation Excerpt :

      Genetic algorithm (GA) was used to optimize the parameters of SVM. Experiments show that this optimization algorithm can identify and classify flip chips quickly and accurately [45]. To solve the problem of not being able to diagnose the internal insulation defects of oil-impregnated paper bushings in detail, Wang et al. proposed a multiclass least-square support vector machines (LS-SVMs) based method optimized by the cuckoo search (CS) algorithm to diagnose the internal insulation defects of oil-impregnated paper bushings [46].

    • Intelligent detection technology of flip chip based on H-SVM algorithm

      2022, Engineering Failure Analysis
      Citation Excerpt :

      The method based on machine learning features for feature extraction of original image information has more effective implementation and less training period than traditional CNN. Most machine learning methods for FC defect detection were non-destructive [24,25]. SVM was one of the typical machine learning methods, and others have optimized and improved SVM in different aspects.

    View all citing articles on Scopus
    View full text