1. Introduction

The increasing interest in renewable energy sources has placed considerable interest in improving efficiency and reliability of power modules for conversion of electrical power. Reliability issues are becoming increasingly important due to long expected lifetimes which go as far as 20-30 years in connection with wind power generators. Due to the high quality of Si based semiconductor components as well as intelligent module design of device protection the bottleneck with regard to module lifetime is closely connected with thermo-mechanical degradation of module interconnections [1,2].

Heavy Al wires ($>100\mu m$ in diameter) ultrasonically bonded to the Si chip topside is one of the commonly used interconnections in high power modules [3]. Due to a difference in coefficient of thermal expansion between the Al wire and Si chip, temperature variation induces stress in the bond interface. Under normal operation the module is subjected to pulsed current conditions with active semiconductor switching, leading to temperature and thus strain variations [1]. The wire bond robustness under thermal cycling is directly connected with the interface microstructure. When the polycrystalline Al wire is bonded to the Si chip the Al grains in the bottom of the wire are refined in order to form a bond with the substrate [3]. The grain size and orientation in the refinement region thereby dictates the bond robustness with respect to thermo-mechanical fatigue and thereby module lifetime [2]. Accordingly, proper characterization of bond microstructure is essential to understand failure mechanisms and increase component robustness.

Optical imaging of surfaces with nonlinear contrast mechanisms allows for investigation of properties that are not accessible through traditional linear optics. Microscopy based on optical second-harmonic generation (SHG) has been demonstrated as a versatile tool in several research fields from biological systems [4] to crystal growth [5,6], and analysis of electronic components [7]. The higher order of the SH response function makes the technique very sensitive to surfaces, interfaces, and inhomogeneous structures. Furthermore, the second-order response is sensitive to crystal orientations, even for cubic crystals. It is this sensitivity to crystal orientations that will be explored for polycrystalline Al in this work. Scanning second harmonic generated (SHG) microscopy is used in the present work for studying the polycrystalline structure of Al wire bonds. The size and shape of the Al grains/domains depend on thermal and mechanical processing and play a crucial role for the macroscopic properties of the material. A smooth metallic surface does not provide contrast between domains in linear microscopy unless the surface is processed to introduce a geometrical surface structure, typically though chemical etching [8]. Using this technique makes it impossible to study interfaces lying under a transparent media. However, in nonlinear optics the response depends on the crystallographic orientation of the microcrystals due to the selection rules of the nonlinear process [9–11]. As the SHG signal for a given combination of input and output polarizations depends on the orientation of the microcrystals, information about crystal orientations may be obtained.

2. Theory

The high surface and interface sensitivity of SHG in reflection from a centrosymmetric media stems from the broken centrosymmetry at the interface that allows dipole contributions to the nonlinear response [12]. In the present work the contrast in the SHG images is caused by the anisotropy of the in-plane nonlinear response as well as the difference in isotropic response among different crystal orientations relative to the surface normal.

Rotational anisotropy of Al crystals is closely connected to interband transitions where the so-called parallel interband transition at 1.5 eV (825 nm) dominates when the Ti:Sapphire laser is used as excitation source. This interband transition derives from a band gap induced by the U200 Fourier component of the lattice potential [13]. The symmetry of the rotational anisotropy will then indicate the orientation of [100] directions of the crystal relative to the surface plane. A detailed discussion of the rotational anisotropy of Al crystal faces, including deviations from perfect low-index surfaces, has been presented by Petukhov et al. [11]. When investigating the samples in this work, a fine grade polishing from the side of the wire interface was performed until reaching the wire of interest. Therefore the SH signal detected in these experiments will originate from crystal planes oriented under many different angles relative to low-index planes, i.e. miscut planes. However, the analysis in this work will be restricted to grains that can be represented by vicinal (100) and (111) surfaces which have been investigated in detail in [10] and [11]. A (111)-oriented surface with symmetrical projections of (100)-directions on the surface will give rise to a 3-fold rotational symmetry while a (100) oriented surface leads to a 4-fold symmetry. However, a perfect 4-fold symmetry cannot give a dipole contribution to surface SHG and only weaker quadrupole effects contribute. On the other hand, any miscut of the (100) surface will reduce the symmetry and introduce dipole-allowed surface contributions. For a (100) surface miscut towards a [110] direction the p- to p-polarized variation of SHG with azimuthal angle $\phi$ can be described by

3. Materials and experimental methods

The regarded sample is a cross-sectional cut of an Al wire bond on an IGBT chip from a standard high power module designed for power conversion in wind power generators [3]. To investigate the microscopic structure of the interface the sample was embedded in epoxy for mechanical protection, and subjected to fine grade polishing from the side of the interface until reaching the wire [3]. The sample was treated in ambient air; therefore a native oxide layer was present during the experiments and SHG thus originates from the Al/oxide interface.

To investigate the variations of the SH signal over the sample, a femtosecond laser system (Spectra-Physics Tsunami Ti:Sapphire oscillator) was used to deliver ~85-fs pulses at a repetition rate of 80 MHz with average power of 0.94 W at 786 nm. The power of the light reaching the sample was reduced to 23mW by neutral density filters and a polarizer, just below the damage threshold of the surrounding epoxy. In order to rotate the linearly polarized light reaching the sample, a half-wave plate was inserted in front of a high numerical aperture aspheric lens (Geltech^TM Molded Glass Aspheric Less, NA: 0.62, focal length: 4.03mm), focusing the beam onto the sample. The sample was mounted onto a 2D stepmotor driven stage, making it possible to raster-scan the surface with a mechanical resolution of 0.2μm. The 2D stage was positioned such that the angle of incidence of the laser light was 38°, see Fig. 1. The SH light leaving the surface was detected by a photomultiplier tube (PMT). In front of the PMT a motor driven polarizer was placed, used for analyzing the SH light from the sample. A colored glass (RG715 filter) placed before the objective to remove any light in the wavelength region of the SH radiation generated by the laser and in the optical components before the objective. After the sample a BG39 bandpass filter was placed in order to remove reflected excitation light as well as any third-harmonic light generated in the beam path. To get information about the crystallographic orientation of the domains, the polarization of the incident light and the analyzing polarizer where rotated either perpendicular or parallel to each other, as demonstrated in [14]. As the rotation of the half-wave plate in the input beam did not shift the beam within the dimensions of the Al grains, local symmetry properties of the grains could be probed.

 

Fig. 1 Schematic of the experimental setup used: P1: Polarizer used to adjust the power reaching the sample. λ/2: Half waveplate mounted on a rotational stage, making it possible to rotate the polarization of the light reaching the sample. RG: RG715 Colored glass filter. AsL: High numerical aperture Aspheric lens. L: Lens. P2: Motordriven polarizer. BG: Colored glass filter (BG39). PMT: PhotoMultiplier Tube.

Download Full Size | PDF

After obtaining the SH microscope images and the rotation scans, the Al sample was electro-chemical etched with Barker’s reagent (electro-etching) in order to promote grain structure contrast for comparison in a bright-field optical microscope, see [3] for additional information.

4. SH microscope investigation of the grain structure

A scan of 450 x 1350 μm² was performed on a selected wire bond interface on the IGBT chip using a 2.4 μm step size [Fig. 2(a)].

 

Fig. 2 a) SHG-image of wire bond interface before electro-etching with p to p-polarization. The arrows marked A and B shows the areas where the beam focus was fixed during the rotational spectra recording. b) Zoom of an area of interest in Fig. 2(a) c) Optical microscopy image of the interface of the region in (a) after electro-etching. d) Optical microscopy image of the same area in b).

Download Full Size | PDF

The polarizer and λ/2 plate in the incident beam and the analyzer in the reflected beam were adjusted to transmit p-polarized light, since this provides the highest isotropic contribution. Variation in the SH response divide the surface into regions of different signal levels. By comparing the SH-image [Fig. 2(a)] with the image from the optical microscope after electro-etching [Fig. 2(c)], it is clearly observed that the overall grain structure is identical. Additional information can be obtained when comparing individual grains. In Fig. 2(d) two areas marked A1 and B1 show what looks like single grains in the linear optical microscope, and two grains in the SH-image in Fig. 2(b). To investigate the contrast in the SH-images, scans of the same area were recorded for three different polarizations [Fig. 3(b)]. The images are strongly polarization dependent, and several regions can be identified with different intensities in the three scans. For one polarization combination there are large areas without structure that split into domains for another polarization combination. An evaluation of the complete grain structure from SHG microscopy thus requires scans using different polarizations.

 

Fig. 3 a) Optical microscopy image of wire interface after electrochemical etching. Insert shows enlarged image of the same area displayed in c). b) SHG-image reduced area for different polarizations. The arrows marked C shows the area where the beam focus was fixed during the rotational spectra recording (Fig. 4c).

Download Full Size | PDF

The contrast observed in the SH-images is associated with the rotational anisotropy of the nonlinear response. To reveal the rotational anisotropy of a single grain, the laser focus (~2μm diameter) was fixed at the center of a grain, and the polarization of the SHG was rotated parallel with the pump polarization. At normal incidence this corresponds to azimuthal sample rotation. However, the rotational anisotropy part of the signal is hardly affected by the 38ᴼ angle of incidence used in these experiments as can be seen by evaluating the expressions for rotational anisotropy in [9].

Useful information on grain structures can thus be obtained by evaluating rotational spectra recorded with polarization rotation on the basis of the equations above. In Fig. 4(a) and Fig. 4(b) the laser focus was fixed onto the grains marked A and B rep. in Fig. 2(a). In Fig. 4(a) the SH intensity detected from grain A, was fited with Eq. (1) with $B_2=A_4=0$ representing a surface miscut towards a [100] direction (see Table 1). In this plot the C_1V surface symmetry represented with the ${\cos }^6\phi$ dependence of the rotational angle of the sample, describes the experiments well. For the less intense B grain in Fig. 2(a), the plot shows a different angular dependence with 90° between maxima. The solid line in this case is obtained by fitting Eq. (1) with both B₁ and B₂ different from zero, representing a (100) surface miscut towards [110]. In Fig. 4(c), the SH intensity detected from grain C in [Fig. 2(a)], the rotational anisotropy has 60° between peaks indicating a (111)-type surface that may be represented by Eq. (2).

 

Fig. 4 Variation of the SHG signal with the polarization parallel to the direction of the pump light, recorded for three different grains: a) point A in Fig. 2(a), b) B in Fig. 2(a) and point C in Fig. 3(b). In a) the data were a fitted (red curve) by Eq. (1) with $B_2=A_4=0$representing a surface miscut towards a [100] direction. For graph b) the data were fitted by Eq. (1) with both B₁ and B₂ different from zero, representing a (100) surface miscut towards [110]. In c) the data were fitted by Eq. (2), indicating a (111)-type surface.

Download Full Size | PDF

Table 1. Fitting Parameters to the Graphs Shown in Fig. 4

View Table

A large number of crystals were investigated and the majority of them revealed similar results as those presented in Fig. 4, with the Al(111)-type as the most common face. In some rare cases more complex rotational anisotropy of the SHG was detected (not shown). However, it is always possible to identify the symmetry of the rotational spectra and thus obtain qualitative information about the grain orientation. Exact determination of grain orientations will be a challenging task that cannot be addressed before more detailed investigations of the dependence of the rotational anisotropy on the size of miscut angles has been performed for the low-index single crystal surfaces.

6. Conclusions

The experiments show that the recorded second harmonic microscopy images can provide information about the domain/grain structure of metals. Moreover, additional information of the individual grain orientation could be achieved by using simple interpretations of the recorded rotational anisotropy. Contrary to linear optical microcopy studies of domains, where surfaces are etched to introduce a geometrical surface structure, it is possible study interfaces buried under a transparent media (oxide), since no surface processing is needed.