August_EDFA_Digital

edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 22 NO. 3 10 FAILURE ANALYSIS, STATISTICAL RISK ASSESSMENT, AND ADVANCED MODELING (continued from page 8) failure latency: latency is facedwhen a customer reports a field return, but there is not latency if the failure is revealed in an assembly line at manufacturer’s or at customer’s. Always inside this Is-Is Not tool, ‘Who’ question refers to customers facing the studied failure, but also to the one or several products impacted by this same failure. ‘Why’ and ‘How’ questions potentially deal with root causes. With the specific topic managed by a risk assessment about the number of likely-to-be-faced failures, this type of analysis is oriented towards prediction, and aims to prevent reoccurrence. In an 8D template, it is included in the seventh step that concludes the 8D work, just before team congratulations. A typical question, linked to prediction work, is about prediction validity. Prediction accuracy strongly depends on all the questions previously listed and included in the Is-Is Not Tool: their answers identify failure context and set prediction method. It is the reason why a structured problem solving approach is so important, to question all the failure aspects and evidences to validate all the assumptions that may be set: an exhaustive questioning and search for evidences are indeed the two main basis of a structured problem solving, and many other tools can be implemented, beyond the first two tools roughly presented here, Is-Is Not and 8D templates (see Table 1). Furthermore, in quantification of pending failures, prediction validity concern is fitting with error and uncer- tainty management: statistics are the main tool for this. The following section deals with statistics and confidence intervals. CONFIDENCE INTERVALS TO MANAGE RISK AND UNCERTAINTY From a start situation in which all the questions were asked and answered about a specific failure, impacted products, suspected parts, ...and for which evidences were highlighted, it deals now about quantification and prediction: a failure rate estimation is targeted (expressed in ppm), within a potential time baseline if latency is one of the failure features (typically expressed in ppm/year). If failure is revealed on manufacturer or customer assembly lines, that is to say if the failure is not latent, statistical analysis implemented to predict failure rate is relatively easy and fits onlywith a confidence interval built on the failure rate observed, from the risk level accepted (ppmbetween high and lowvalues at statistical Alpha risk or type I error risk). In case of latency, as previously stated, when failing parts are reported as occurring in field, a time period is added to this failure rate (ppm/year). Furthermore, if some additional data are available for each field return, an advanced modeling can be implemented, resulting Table 1 8D framework and structured problem solving tools possibly embedded Step in 8D framework Main questions Typical SPS tools 1 Team Roles determined? 2 Problem description Problem clear and verified? Logistic impact clear and lot list available? Is-Is Not 3 Containment Is customer safeguarded? 4 Root cause All possible root causes investigated and verified? (technical and systemic root causes on 3 axis: occurrence/root cause escape/failing product escape) Fishbone / Ishikawa Fault Tree Analysis 5 Whys in each of the 3 axis, upstream (systemic root causes) and downstream (technical root causes) 5 Determine corrective actions Possible corrective actions determined? Effectiveness verified? 6 Implement corrective actions Implementation date clear? Effectiveness validated? PDCA (Plan Do Check Act) 7 Prevent reoccurrence FMEA and Control Plan updated? Risk assessment and lessons learned performed and shared? Poka Yoke 8 Conclude and congratulate Conclusion