· Table of Contents
· Fundamental & Applied Research
Characterization of Imperfect Interfaces in IC Packaging using Deconvolved Ultrasonic SignalsX. Jian, N. Guo, J.Abdul, and H.C. Yeo
School of Mechanical & Production Engineering
Nanyang Technological University
A multi-spring model of ultrasonic reflection signal is used to describe interface conditions of a three layer (silicon/adhesive/copper) structure in IC packaging. Assume normal incident longitudinal wave, various interface reflection signals are analyzed and related to the interface condition that is modelled by spring stiffness. Samples with varying interface bond condition are fabricated. A plane longitudinal transducer of nominal frequency 100 MHz is employed to measure the ultrasonic response of structure silicon/adhesive/copper. Measured signals in general agree with corresponding prediction of same interface conditions.
The measured ultrasonic response is deconvolved, and then used to optimize the spring stiffness of the two interfaces used in the theoretical model of silicon/adhesive/copper so that the deconvolution from the predicted signal has best match to that from the measured signal. The optimized interface spring stiffness represents the interface quality of the sample, and is also compared with the interface failure strength measured. It is found that good correlation exists between the optimized interface spring stiffness and the failure bond strength, indicating the extracted interface spring stiffness is a good measure of interface condition.
Keywords: Multi-spring model, IC packaging, bond strength, imperfect interface, ultrasonic evaluation
The integrated circuit (IC) plastic packaging has many multilayer structures such as mold compound/ silicon, and silicon/ adhesive/ copper. The total thickness of the structure is small, for example, silicon/adhesive/copper can be less than 1 mm. Weak interfaces in IC packaging often occur prior to other defect types during fabrication process, and could cause failure in the process of reflow and subsequent service. Therefore, the interface bond evaluation in IC packaging plays a key role in the reliability testing of IC packaging. However, it is difficult to establish exact relationship between the acoustical parameters and the interface mechanical characterizations.
The interface condition description is of great importance in bond evaluation of IC packaging. When interface is imperfect, strains and stresses in layers between an interface discontinue. Many models were proposed to describe the strain and stress relationship between an interface, so as to give better agreement between the prediction and actual ultrasonic measurement. Interface Spring model is employed and evolved in various cases by Rokhlin et al (1991). Generally, ultrasonic responses from spring models give a reasonable description of interface between two solids. However, it may be inadequate when multilayer structures and multiple interfaces are involved because ultrasonic reflection from embedded interface of interest is much complicated.
The ultrasonic responses of a multilayer structure are sensitive to interface condition, materials properties of each layer within the multilayer, as well as coupling. Ultrasonic non-destructive evaluation has been widely applied in the characterization investigations of multilayer structures effectively (Betta, 1993; Kazys, 1997; Jian, 1999). Many literatures are involved in imperfect interfaces of a multilayer structure. For example, Lavrentyev and Rokhlin (1998) studied the ultrasonic spectroscopy characterizations.
The actual measurement and theoretical calculations of ultrasonic reflection coefficient and materials attenuation should be carefully treated (Veen, 1986). In the calculations of ultrasonic response, matrix transfer method and global matrix method are employed. Lowe (1995) gave a comparison and review on various matrix techniques.
Many signal processing methods have found their various applications in the multilayer characterization in noise suppression, resolution enhancement, feature extraction, and elastic constant calculation etc. Signal processing application in ultrasonic NDE is also one of its important research direction. Increasing number of papers in recent year gave verifications (Kazys, 1997; Jian, 1999).
In the paper, ultrasonic characterizations of the multilayer is calculated under consideration of material attenuation, interface bond condition, in order to better understand the effect of all physical parameters. Special treat is made on algorithm of matrix transfer method.
Matrix transfer method is proposed by Thomson in 1950 and further developed by Haskell in 1953. When all the interfaces between the layers is welded perfectly, ultrasonic reflection coefficient can be calculated by the method.
Here, imperfect interface is under consideration. If the size and spacing between the imperfect interface are much smaller than the wavelength, the ultrasonic wave interaction with the interface can be described by a boundary spring model (Rokhlin, 1998). There are relationships among displacements and stresses across the interface of j - 1 layer and layer j,
Where z+j and z-j-1 are at bottom surface of layer j - 1 and upper surface of layer j respectively; Kjn and Kjt are the normal and shear spring stiffness of the interface, indicating the bonding quality. When Kjn=0 and, Kjt=0, stress at the interface is zero, corresponding to free interface, representing disbond. When Kjn and Kjt approach infinite, both stress and displacements continue across interface, representing perfect condition.
The relationship can be expressed by matrix explicitly,
Finally, the displacement-stress vector at the bottom boundary of layer 1 relates to that at the upper boundary of layer n+1 by,
Considering a three-layer structure silicon/adhesive/copper in IC packaging and the material property in the structure is shown in Table 1. Main interests are the response characterizations of samples with varying interface quality within the structure. Conditions of the two interfaces within the structure can be changed from perfect to disbond. For discussion convenience, the interface of silicon and adhesive is named as interface 1, described by normal and shear interface spring stiffness Kn1and Kt1 respectively. Accordingly, the bottom interface of adhesive and copper is named as interface 2.
|Longitudinal wave velocity (m/s)||8409||2530||4696|
|Transverse wave velocity (m/s)||5505||1549||2643|
|Table 1: Material characterization of multilayer structures in IC packaging|
Figure 1 shows the calculated ultrasonic reflections of samples in (a)-(c), and corresponding measured responses in (d)-(f). The interface conditions are as follow, in Figure 1(a) and 1(d), interface 1 is disbond hence condition at interface 2 is thus immaterial. In Figure 1(b) and 1(e), interface 1 is perfect and interface 2 is disbond. In Figure 1(c) and 1(f), both interfaces are perfect.
|Fig 1: Calculated and measured ultrasonic responses of Si/Adh/Cu structure.|
Figure 2 illustrates the arrival of the interface echoes, and some general features can be observed. Since the acoustic impedance (longitudinal wave) ratio between silicon to adhesive is about 2.8, while that of copper to silicon is 6.2, echo B1, B2 and B3 have reverse phase compared with echo A. Echo D1 is much smaller in amplitude than echo B1 and C1. Poor bond quality of interface 1 would increase the amplitude of echo B's, and decrease the amplitude of echo C's. The amplitude ratio of echo B2 to B1 and that of B3 to B2 are same and depend on the condition of interface 1. Echo B's are out of phase with echo A but independent on the condition of interface 1. However, imperfect interface 2 causes not only amplitude increase of echo C's but also the phase shift. Imperfect interface 2 would further decrease echo D's.
|Fig 2: Echo analysis from structure si/adh/cu.|
Samples are fabricated to verify the evaluation methods. The procedure to fabricate IC packaging sample is close to that used in industry. Thermally conductive adhesive was cured in an oven to a temperature of up to 175° C and put onto the copper thin plate. The copper and silicon plates are bonded under a constant load/pressure. The bonded sample is then put into oven at 175° C for several minutes. Finally, it is taken out and cooled at room temperature. The thickness of the adhesive is carefully controlled at about 100 micrometer.
|Fig 3: Interface degradation of samples from perfect condition to weak condition. Obvious phase shift can be observed after degradation.|
IC packaging degradation may be introduced during fabrication and assembly by many factors, such as improper curing time, surface contamination by exposure to moisture, and copper oxide growth. IC packaging is liable to degradation in consequent delivery and application process because of thermal recycle and moisture invasion. In this research, samples with varying interface quality are obtained by proper control of curing time, moisture exposure and the thickness of copper oxide layer. Figure 3 shows the measured ultrasonic responses of three samples before and after degradation by moisture exposure. It can be seen that obvious phase shift and amplitude change occur.
Figure 4 shows the typical measured ultrasonic response of samples with change in interface quality marked by two letter labels, where "G" represents perfect interface, "M" for intermediate interface quality, and "B" for disbond interface. The first letter represents condition of interface 1 and the second for interface 2. For example, "GM" means that interface 1 is perfect and interface 2 is of intermediate quality. The interface conditions are obtained by one or combination of following parameters such as moisture exposure, thermal recycle at silicon layer, and peeling.
The measured ultrasonic responses from samples are deconvolved (Jian and Li, 1999; Jian and Guo, 2001). The objective of deconvolution is to get rid of waveform distortion, and to enhance time resolution. The deconvolved curve is called ultrasonic pulse response. By iterating interface spring stiffness used in the model, the predicted ultrasonic pulse response approaches the measured deconvolved ultrasonic response. When the best match is obtained, the optimized interface spring stiffness indicates the interface quality at that particular interface. Consequent shear pull strength is tested on the samples to obtain the relationship between the interface spring stiffness and shear pull strength.
|Fig 4: Typical ultrasonic response pallet for varied interface conditions.|
|Condition of interface 1 and 2|
|Shear interface failure stress (MPa) at interface 2||0||0||4.1||8.9||0||5.2||7.9|
|Spring stiffness of interface 1 (TPa/m)||0||15||40||100||4||4||5|
|Spring stiffness of interface 2 (TPa/m)||0||0||5||9||0||5||6|
|Table 2: Comparison of shear pull strength with interface spring stiffness.|
Test results for samples shown in Figure 4 are listed in Table 2. Small value of spring stiffness corresponds to small shear pull strength as in the case of intermediate interface quality. High spring stiffness corresponds to high shear pull strength as in the case of perfect interface. Similar results are found in Table 3 using the responses of the three samples as shown in Figure 3. It can be seen that after degradation, both the interface spring stiffness and shear pull strength decrease considerately. It can also be observed from Figure 3 and Table 2 that to some extent, the reduction in spring stiffness and shear strength is related to the phase shift.
|Shear interface failure stress (MPa) at interface 2||No||No||no||4.2||5.8||4.8|
|Spring stiffness of interface 1 (TPa/m)||9||10||9||8||9||8|
|Spring stiffness of interface 2 (TPa/m)||9||8||9||4||6||5|
|Table 3: Comparison of shear pull strength with the interface spring stiffness.|
It is known that ultrasonic signal is prone to noise, distortion, echo overlapping and coupling problem, and thus difficult to obtain accurate measurement of echo from ultrasonic signal in time domain or frequency domain. Furthermore, in multilayer interfaces, reflection may be affected by many interfaces whose reflections are not known. As a result, error may exist in the interface evaluation by direct reflection coefficient measurement. Deconvolution can reduce ultrasonic waveform distortion and enhance resolution. Furthermore, interface spring stiffness extraction is based on the matching of the whole ultrasonic response waveform, not just a single echo. As a result, better results can be obtained.
It can be seen that the interface condition of a multilayer interface can be evaluated properly by the optimized interface spring stiffness that best matches the theoretical deconvolution responses with the measured one. It is shown that the extracted interface conditions for the structure in IC packaging agree in general with the shear pull strength tests of the samples.
However, the evaluation method demands huge calculation so it is difficult for fast on-line application. The problem may be resolved by employing neural networks, which has the advantages of nonlinear simulation and parallel algorithm. On the other hand, by proper simplification of algorithm in the multi-spring model can increase calculation speed to meet the demand of on-line application.
The work was supported by a grant from Applied Research Council of Nanyang Technological University.
|© AINDT , created by NDT.net|||Home| |Top||