November_EDFA_Digital

ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 20 NO. 4 28

the RDL. Confirmation of this location was obtained by PFA, which identified a crack in the copper trace 100 µm from its end (Fig. 8, right). 3D PACKAGES This final case study demonstrates that EOTPR can be used to identify every key component of a 3D device architecture, and thus is an ideal tool for failure analysis of 3D packages. To isolate each key feature in the EOTPR waveform, a sample device, shown schematically in Fig. 9, wasmechanically ground from top to bottom in steps, with an EOTPRmeasurement performed at each step. For example, in the first step, the top diewas removed, result- ing in themeasured net being terminated by the exposed µC4 bumps. In the next grinding step, these bumps were removed, moving the termination to the topof the bottom die TSV. In this manner, a complete understanding of the EOTPR waveform can be generated, as depicted in Fig. 9. As individual features of thewaveformcanbe unam- biguously identified, it is possible to scale the obtained waveformand extract accurate distance-to-defect values. OUTLOOK In recent years, a range of technological advances to improve the EOTPR instrument have been implemented. An upgrade to a new delay line system has enhanced the

already impressive SNR, reducedmeasurement time, and extended the achievable scan range. In addition, active laser stabilization has been added to minimize already small variations inmeasurement power. Finally, the devel- opment of semi and fully automatic prober systems has greatly improved instrument usability and productivity. The immediateoutlook focuseson thepotentially chal- lenging process of data analysis. Previously, it was deter- mined that full 3Dmodeling of a DUT can greatly improve an understanding of how an EOTPR pulse interacts with Fig. 10 1D simulation results from the bare substrate and failed device of the 2.5D case study.

an advanced IC package. [8,9] However, this is a time-consuming process that requires significant computing power anddetailedknowledgeof DUTdesign. As a result, TeraView researchers are currently in the process of developing an alternative simulation methodol- ogy that uses a simpler linear model to streamline the modeling process. Figure 10 shows the measured and simulated waveforms for two of the 2.5D devices described above. Here, the measured bare substrate waveform (solid black curve) is used to optimize the linear model param- eters. The resulting simulated bare substrate waveform is shown by the dashed black curve in Fig. 10. Various features of the measured waveform are clearly depicted in the simulated model. The optimized bare substrate model was used to find the best fit to the measured failed unit waveform

Fig. 9 Schematic of the 3D device architecture and the corresponding EOTPR waveforms recorded at each grinding step.

edfas.org