The statistical dependence of oxide failure rates on V/sub dd/ and t/sub ox/ variations, with applications to process design, circuit design, and end use

We describe here a new methodology which demonstrates the power of understanding the statistical dependence of the average failure rate (AFR) from dielectric wearout on statistical variations of the oxide electric field (E/sub ox/) arising from statistical variations of the power supply voltage (V/sub dd/) and oxide thickness (t/sub ox/). It has quantitative applications to process flow design, manufacturing, circuit design, and end use. For process flow designers, it allows about 4 /spl Aring/ thinner gate oxide for the 1.8 V example technology studied here, compared to a simpler worst-case corner approach at maximum power supply voltage V/sub dd/ and minimum t/sub ox/. This reduction in oxide thickness enables improved device and circuit performance. For manufacturing, it also shows that tighter control of oxide thickness would be unwarranted for the specific example given here. For circuit designers, it allows transient overshoot voltages which can be about 10% higher than those based on a worst-case corner field. For end users pushing the performance envelope, it provides understanding of the trade-offs between reliability and power supply parameters.