This paper presents a failure mode detection methodology using a piezoresistive silicon based stress sensor. Data from experiment is used to validate algorithms developed for detection. Dedicated test vehicles are designed and fabricated, where the process parameters and materials are carefully selected to produce delamination between molding compound and PCB after fabrication. The test vehicles are then subjected to thermal cycling of −40°C to 125°C to grow the delamination area. After every 150 cycles, the samples are examined using Scanning Acoustic Microscopy (SAM), and the results are correlated with stress sensor signal. It is demonstrated that the propagation of the delamination area can be detected using the stress sensor. Collected data is also used to examine the applicability of statistical pattern recognition algorithms for detecting the failure. The algorithms considered in the study include Mahalanobis Distance (MD) and Singular Value Decomposition (SVD), which do not require a prior knowledge about failures and are just searching for deviation from norm in the data. The results from the analysis indicate that both techniques are suitable to stress sensor measurements, and thus are capable of detecting failure during reliability testing.

You do not currently have access to this content.