November_EDFA_Digital

ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 20 NO. 4 6

TCT stress test. Using thresholding, voids were detected and marked as red pixels by a custom software tool. The micrographs in Fig. 7 were recorded from the reference sample, which did not receive the TCT. No voids can be observed in the reference samples. Figure 8 also con- tains a combination of GHz-SAM images and electron micrographs of FIB cross sections through selected voids of a sample that contained a TiN interlayer in the AlCu sandwich structure. Two features should be noted here. First, significantly fewer voids are found in samples containing a TiN layer even though they received the same TCT as the samples shown in Fig. 6. In Table 1, the results of a statistical

sandwiched in theAlCumetal lines (Fig. 2). Allmetallization layerswere deposited by physical vapor deposition (PVD). Two sets of sampleswere exposed toa thermal cycling test (TCT) mimicking the aging process under application con- ditions. This treatment consisted of periodically heating and cooling the samples between - 65° and 175°C for 1000 cycles, and was performed according to JESD22- A104E: Temperature Cycling standards, with the tem- perature range adapted according to AEC-Q100 Revision G. Also, an unaltered reference sample was investigated to assess each sample’s initial condition. As previously mentioned, the penetration depth of large numerical aperture acoustic lenses is significantly decreased compared to standard acoustic microscopy transducers. With an 80-µm focal length, penetration depth is limited to a fewmicrons beneath the surface. For this reason, sample preparation was required to provide backside access through 0-2 µm of the remaining silicon. To avoid sample preparation artifacts, provide a flat surface, and leave the AlCu lines unaltered, the acoustic inspection was performed through the backside as illus- trated in Fig. 3. Samplesweremounted on a dummywafer prior tomechanical treatment to ensure secure handling. The silicon substrate was then removed by highly parallel chemical mechanical polishing (CMP) down to the oxide layer acting as end point detection using a MultiPrep Polishing System (Allied High Tech Products, Rancho Dominguez, Calif.). Figure 4 shows an acoustic GHz micrograph recorded through the SiO 2 layer, according to Fig. 3. The AlCu power lines can be clearly resolved and voids in the metalliza- tion appear as bright spots. In Fig. 5, an acoustic GHz micrograph and a secondary electron (SE) micrograph of a focused ion beam (FIB) trench of the same area can be compared. Using the FIB, the surface of the AlCu lines

Fig. 3 Illustration of sample preparation and acoustic inspection.

Fig. 4 Acoustic GHz micrograph recorded through the SiO 2 layer after removing the Si die.

was milled to reveal the voids and confirm the findings obtained by GHz-SAM. Corresponding voids are marked by identically colored circles in Fig. 5. Wi t h t h i s pe r f o rman ce , GHz-SAM can be used to inves- tigate the influence of Ti or TiN interlayers on the formation of hydrogen triggered voids. Figure 6 contains two acoustic GHz micro- graphs recorded at two different defocus positions to place the imaging plane at adjacent depths within a sample that received the

Fig. 5 Left: Defocused GHz-SAM micrograph. Right: SE micrograph after exposing the AlCu lines by FIB. Corresponding voids are indicated by colored circles.

edfas.org