November_EDFA_Digital

ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 20 NO. 4 8 Sample indent # of voids per 100 µm Size of void appearance [µm 2 ] REF 0 0.00 ± 0.00 S1 144.6 2.54 ± 2.7 S2 207.9 2.1 ± 2.5 S3 12.49 1.1 ± 0.8 S4 15.1 1.13 ± 1.17 Without TiN layer With TiN layer

Table 1 Results of void detection based on the acoustic GHz microscopy data plus electrical resistance. The acoustic focus was approximately placed in the axial center of the metallization.

through 25-µm Cu bond wires on an unstructured Cu metallization. For comparison, the same chip was also investigated by conventional acoustic microscopy at 300 MHz. Figure 10 shows the results of both the SAM and the GHz-SAM analyses on the wire bond interfaces. The leftmost image was obtained at 300 MHz by conventional SAM. The graph in the horizontal center is the result of GHz-SAM imaging, recorded at the same interconnects as imaged by conventional SAM. The rightmost graph shows amagnification of a region in the center image (GHz-SAM). It can be noted that the sensitivity and resolution of the conventional SAM even at 300 MHz is too poor to resolve the structures in the interface between the bondwire and the pad metallization. Although, the GHz-SAM reveals a highdegree of inhomogeneity in thebonds at the right and the bottom bond at the left. The three bonds that appear bright in the conventional SAM image are delaminated across most of the bonding area. However, at the right

R [Ω]

1.11 5.11

12.88

5.3

12.79

As an alternative approach to ball bond quality assess- ments, the intermetallic phase formation between ball and pad can be analyzed. While the selective removal of the ball bonds iswidely established (e.g., for Auor Cuwires on Al pads), [11] thismethod naturally cannot be applied to mono-metallic interconnects formed by the upcoming package technologies (e.g., Cu to Cu wire bonding). Conventional high frequency backside SAM can be employed for analyzing bond interfaces of various geom- etries. [6] Although, the resolution achievable using con- ventional ultrasonic transducers in the frequency range of up to 230 MHz is approximately 20 µm depending on the numerical aperture of the lens and the required analysis depth inside the specimen. However, this is not nearly sufficient for a precise and accurate assessment of wire bond interconnects, especially considering today’s small wire bonds with diameters of a few tens of micrometers. To provide access for the GHz-SAM to image the bonding interface, the lead-frame and die attach materi- als such as solder or glue were removed by employing standard wet chemical etching. Even though GHz-SAM inspection is possible through 1-4 µmof remaining Si, with the samples investigated here all remaining Si was removed by employing a selective SF 6 -based plasma etching process stopping at the oxide layers of the BEOL stack. The reason for this approach is that imperfectly prepared surfaces containing scratches or surface warpage lead to unwanted artifacts in the acoustic micrographs. Special effort and care must be taken to prepare perfectly flat Si surfaces of thinned dies. A schematic of the samples’ cross sections before and after prepara- tion is provided in Fig. 9. The red arrow indicates the direction of acoustic inspection. The sample housing was a standard PG-LQFP-176 package. Electrical interconnection was obtained

(a)

(b)

Fig. 9 (a) Schematic of a common plastic package with ball bond interconnects bonded on a pad or power metallization. (b) Toprovide access for backsidehigh- resolution SAM analysis, the die was exposed and thinned. Red arrow indicates direction of acoustic inspection/ analysis.

Fig. 10 Comparison of acoustic micrographs of ball bonds formed from a 25-µm Cu bond wire, recorded through the back- side access. Images were recorded by (a) conventional 300 MHz transducer with 2.5 mm focal length and (b) ultra-high resolution 1 GHz acoustic lens with 80 µm focal length. (a) (b) (c)

edfas.org