August_EDFA_Digital

edfas.org 15 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 22 NO. 3 delamination (see Table 2 and Fig. 5). Whatever the design, a risk is set as medium if self-heating level is high, which may prove that prediction risk seems more critical. In time, when a first risk assessment is performed for one product, the first failure quantity prediction may define product risk level as low in the extent that failure rate is low (p low) and if amaximumfailure rate event is observed (early max failure rate). Otherwise, risk level may be high until maximum failure rate event was predictable. Now, at this step, it becomes possible to model field for merged products from each risk level. The three field models obtained are fitting with risk model. A risk level that is finally fitting with a cluster by design and applica- tion can be assigned to any product for which design is characterized in term of number of metal layers, contact density in edge seal and application self-heating, and its risk model is implemented to predict number of failures that may be expected, with the confidence interval of risk level. CONCLUSION An inter dielectric delamination defect, generated by a manufacturing equipment during a short period of time, and impactingmore than adozenproducts, was subjected to a failure analysis, then to a risk assessment. These two analyses are typically part of an 8D framework. A risk assessment estimates the number of failures pending in field, and prediction validation is always challenged since it leads to decide about actions to prevent failure reoccurrence. This case study clearly demonstrated that prediction quality is asmuch linked to structured problem solving approach rigor that is imple- mented in the issue faced as statistical method choice and implementation. Actually, there may be a higher risk than a high failure rate predicted: the one of a wrong prediction. For this defect, an advanced life distribution modeling in defec- tive subpopulation Frechet was implemented on all the impactedproducts, and it was possible touse two features in each product model to rank risk in terms of failure quantity and prediction validity. Lastly, a linkwas highlighted between design sensitiv- ity-to-defect and failure rate risk, and between product application and prediction validity risk. So that, it may become possible to estimate a risk level for any product impacted by this defect for which no data are known, from its design and application characteristics. ACKNOWLEDGMENTS The author would like to thank NXP worldwide teams who regularly share with her some tricky cases requiring her expertise. All these issues are always strong oppor- tunities for her to still grow in data analysis and risk assessment field. Let them consider author’s papers as a knowledge transmission in returnof these cases provided. REFERENCES 1. C. Bergès, Y. Chandon, R. Gubian, “InnovativeMethodology for Failure Rate Estimation from Quality Incidents, for ISO26262 Standard Requirements,” IPFA 2012. 2. C. Bergès, “Failure Rate Estimation in Field for a Defect, in Function of Manufacturing Defectivity Density: Case Study for a Gate Oxide Rupture on Valve Driver in Automotive Semiconductor,” ICTACT Journal on Microelectronics, April 2018, Volume: 04, Issue: 01. 3. M. Castignolles, T. Zirilli, E. Cattey, J. Lewenstein, S. Schauer, W. Liu, J. Chen, J. Hammett, and S. Subramanian, “Failure Analysis of ILD Delamination: Uncovering Multiple Root Causes,” Proc. Int. Symp. Test. Fail. Anal. (ISTFA), 2015. 4. C. Bergès, S. Jinrong, H. Chi, Q. Xuejian, R. Jun, T. Li, W. Gaojie, and Z. Haus, “Provocation Tests, Design of Experiments and Advanced Statistical Modeling to Estimate Product Sensitivity to a Defect: Delamination Failure Case Study for Automotive Semiconductors,” IPFA 2019. 5. C. Bergès, J. Goxe, “Statistical DefectiveSubpopulationModel for Risk Analysis: Usage Case Study to Compare Sensitivity to Delamination Defect, for Automotive Semiconductors,” ESREF 2019. ABOUT THE AUTHOR Corinne Bergès is an engineer in Electrotechnics, Automatics and Electronics from ENSEEIHT/ INP engineering school in Toulouse, France. She has a Ph.D. in Electronic Systems. She started her career as a design leader for spatial and seismic prospection, then was a project leader in CNES spatial agency in France. Twenty years ago, she joined Motorola as a Reliability and Failure Analysis. Passionate aboutmathematics and statistics, she started a cursus in Six SigmaQuality, reachingGreen Belt, then Black Belt certifications. Committed in innovative statistical analysis for design, quality or test, she is an expert in risk assessments and data analysis, and a training leader in Six Sigma and Statistics for EMEA. Always in Motorola that became Freescale then finally NXP Semiconductors, she recently joined an organization that aims to spread themost advancedmethodologies in Structured ProblemSolving and Data Science in NXPworldwide, for NXPmanufacturing, subcontractors and suppliers. She has authored and co-authored tens of articles in statistics and risk assessments, data analysis, and data science.