I am planning to evaluate a new process which is supposed have a better defectivity performance (semiconductor industry). The question that I have is lets say that my process failure is not repeatable during a certain runs of experiment. For example in my 100 wafers run, i have observed that the failure happens on the 20th and 50th wafer run. A repeat on another 100 wafers run the failure happen on the 10th, 31st and 70th wafers. In this situation how do I design my experiment in order to determine whether my new process does indeed improves the defectivity of my baseline process
Hope someone is able to help me on this
It seems to me, that one of the possible ways is to compare the inverted (rises from bottom to top) Kaplan-Meier curves (which reflect the cumulative probability of failure). As a time variable (X-axis) the "conditional time" can be used (the unit of which is wafer).
Create, and open in JMP data file with columns (name):
Time: just a sequence of numbers from 1 to 100
Censor: with code 0 and 1 (failure "YES" – 0 (sic!), failure "NO" - 1)
Grouping: with code 1, 2 etc (comparison groups)
Analyze->Reliability and Survival->Survival
and activate (tick): "Plot Failure Instead of Survival"
Sorry for brevity (because of my weak English) :)
These curves reflect the (cumulative) probability of the failure. So
a process that has better defectivity performance has a lower curve.
Also need to consider (but not fetishization) p-value. But all this applies
to the theory of statistics.
Able to share with me what does the number stands for, is the alpha also set at 0.05?which means in this case we are not able to reject the null hypothesis that the two groups are the same?
Irfan, your alpha is 0.4138 (Log-Rank test). Since 0.4138 is greater than 0.05, we are not able to reject the null hypothesis that the two groups are the same. So formally! But this fact should not prevent you from stating that your (cumulate) plot lower (If it is so), and hence the process has a better defectivity performance. Generally speaking, getting a statistical significance difference is still quite difficult. But this is my subjective attitude (perhaps, experts here will find it possible to make their corrective comments).