I have an MSA question which I’m sure you can help me with, it’s seems like a simple one but for some reason I’ve convinced myself that I’m confused. We ran an MSA here where the model was Y = Day + Load[Day] + Site*Load[Day] + Repeat[Site,Load,Day], the measured response was oxide thickness. This was 3 days, 3 loads/day, 9 sites/load, 5 repeats/site; one wafer was used for MSA. We do have an SPC chart for this process and plenty of historical data, it’s an Xbar-S chart (Xbar of wafer mean calculated from 9 measurement sites per wafer; 1 wafer measured per process run). Using the variability platform we calculate %GR&R as a % of process, when prompted to enter the historical sigma I started to question which sigma I should be using. I’ve been entering the historical sigma calculated from the wafer averages of historical data for this process…..but this time I questioned whether I should be using the historical sigma calculated with the site level data (i.e. raw data) from historical data. There is a 2X difference in sigma between wafer level and site level calculations (site level data has 2X higher sigma as you would imagine). Any input on which one is more appropriate in this case? Since we chart wafer means should I be using the sigma calculated from wafer means or site level data? Thanks in advance.
I'm going to attempt to answer my own question as I believe I have the answer. Hopefully this will help others that may have the same question in the future. Short version of the answer is that I should be using the sigma calculated from the historical individual site measurements. Long version now....I'm not measuring wafer means in my SPC chart, I'm measuring individual sites and calculating Xbar and Std Dev from these individual site level measurements. My process variation isn't variation of means, it's variation of the individual sites. Hopefully that makes sense. Feel free to chime in with input.