Process Optimization with DoE and Design Space Profiler

See how to describe the goal, specify effects for an assumed model, generate a design, collect experimental run data, create, and fit a model incorporating experimental data, and use factors to optimize and improve settings. With Design Space Profiler, interactively find limit of failure regions (operational ranges) inside the factor space where the model(s) predict that the response(s) will be within specification.

See how to:

• Examine data Used in experiment case study
• Examine experiment design possibilities
• Summarizie results of creating Definitive Screening Design (DSD) and profiling Design Space
• Create DSD and interpret results
• Fit and interpret statistical model to design factors using Effective Model Selection for DSD methodology
• Use Design Space Profiler to determine factor settings necessary to keep process within spec limits

Questions answered by @Phil_Kay during the live webinar:

Q: How you get a DSD DOE that used multiple levels? Does the factors in the DOE contain a certain amount of levels unique to each factor? Or, does JMP chose this when selecting the DSD as an approach?

A: Basically, a definitive screening design always has 3 levels for each factor. So, you define the low end and the high end of the range and the definitive screening design will always give you 3 levels. It will always have runs where the factor is set at the low.  It will have runs where the factor is set at the midpoint, and also where it's set at the high point. All you need to do is identify your factors, give the high and low points and JMP will interpolate to find the center point of each factor range for you. A related question might be what if I’ve only got defined levels like you could only set a furnace to 200, 280 and 310. Here they are not evenly space factor levels. In that case, you'd want to use a custom design.

Q: How do you color the prediction profiler within spec?

A: With any graph in JMP, you can double click any axis and add reference lines or ranges.  See below.

Q: Can you have the thickness variability as response and how?

A: We could. In this case to do that, where we were measuring the film thickness all these different points about 25 points per wafer, we could use JMP to calculate the standard deviation or the range or both.  Then just use that as your response.

Q: For some of the factors, the middle value in the design was not the mean value. Was there a reason for that?

A: With a definitive screen design, they always will be absolutely bang in the middle of the continuous range But, it just looks a bit skewed because of the way the distribution bins things and the plot displays. If the actual value of your factor was somewhat off center, you could just put in the real value and it mess up the design properties a bit. You probably can't use the fit definitive screening tool that I showed you, but it will still be essentially a good experiment.

Q: What determines the highest available "portion in spec."? For instance, could one set 100.00%? Or does noise in the process affect this?

A: Basically, you define the highest available portion. 100% might be unrealistic. Statistically, there is always variability and achieving something with 100% certainty is never possible. You need to decide what is an acceptable failure rate in these kind of situations. What's the cost of a failure and how much of that you can afford? This is a business, not a statistical, decision. JMP helps you make that business decision, to find the right compromise between in spec and the cost of trying to control 100%.

Resources

• Mastering JMP video and materials, Maximizing Quality Using Design Space Profiling