Design of Experiments for Thin-Film Engineers

Why OFAT falls short

Changing one variable at a time cannot reveal interactions, such as how the effect of pressure on film stress depends on power. Many useful process windows sit exactly in those interactions.

Classical designs

• Full and fractional factorial designs for screening
• Response surface designs (central composite, Box-Behnken) for optimization
• Mixture designs for composition problems

Sequential and Bayesian designs

When each run is expensive, sequential designs choose the next experiment based on everything learned so far, balancing exploring unknown regions against refining promising ones.

Practical tips

• Randomize run order where the tool allows it
• Include replicates to estimate noise
• Record chamber state and maintenance history
• Agree on the response metrics before the first run