Hi All,
In semiconductor industry, from the lot size of 12000 units, 600 units are rejected units in Test. Now I would like to draw sample from 600 rejected units for X-ray analysis. How many samples i should take for X-ray to conclude the defect mode in these rejected units? (Example: offpad wire bonding).
Pls advice.
Thanks and regards,
veness
I am going to suggest three different approaches, with three different sample sizes. You need to assess which scenario fits you best.
Scenario #1
How many unique defect modes are there? If there are less than 10 modes, you can take 5 times the number of modes to estimate the percent of each mode in the defective tests. If the number of modes is higher, then you will need to test a substantial amount to ensure all modes are in the sample.
Scenario #2
You can use the square root of N+1 leading to a sample size of 26 using N=600. This would be used to test if all 26 had a defect. You can also use regular attribute sampling plans for different AQL and LTPD values if the purpose is to confirm the test with X-ray analysis.
Scenario #3
Would require doing X-ray on both the "good" and "bad" project to confirm the inprocess testing and X-ray are consistent. Since you have a 5% defective rate, you can take a sample if 50 "bad" and 50 "good" using a 2x2 contigency table. You will need to use a test like Fisher's Exact since the expected value in some of the cells will be less than 5.
Hope that helps some....