August_EDFA_Digital

edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 22 NO. 3 14 tests may be arranged using a design-of- experiments methodology. Two types of risk were previously described: risk of high failure quantity, for one first type; and risk of wrong predic- tion, due specifically here to a maximum failure rate event happening late, for a second type. From this risk categoriza- tion, three zones are defined on the prob- ability function graph. Schematically, a low-self heating in application increases prediction risk (late max failure rate time), while failure quantity risk (p high) is increased by a design with less metal layers and less contacts in edge seal, which makes it more sensitive to Fig. 4 DS Frechet probability function for global field model superimposed (red) with the elementary model for each product (all the other colors) (left) and for the only one global field model (right) (probability axis is zoomed). Table 2 Risk level in term of prediction and failure quantity, according to product design and customer application, and fitting features for life distribution Risk level Design features Customer application Life distribution (probability density and hazard functions) High Minimum edge seal Two metal layers Low self-heating High failure prediction (p high) And High prediction risk: -Late max failure rate -No max failure rate event predicted yet -Failure rate slowly decreasing after maximum failure rate event Medium Minimum edge seal Two metal layers High-self heating Medium failure prediction and early max failure rate event (low prediction risk) Low Extended edge seal >Four metal layers High self-heating Low failure prediction (p low) and failure rate quickly decreasing after maximum failure rate event, whatever max failure rate event (low prediction risk) Fig. 5 For one product, prediction and failure quantity risk is set by location of its field model among probability functions for the other impacted products (p parameter and max failure rate time).