JMP platforms and statistical methods

Graph Builder:
Comparative dot plots and line graph

Fit Model:
Factorial ANOVA models

Distribution:
Histograms, summary statistics, bar chart, frequency table, capability analysis, confidence interval for a proportion

Formula Editor:
Creating a new variable

Objective

Use multifactor ANOVA to analyze data from an experiment to identify and quantify the impact that factors have on the resulting thickness of a thin film in a semiconductor chemical vapor deposition process.

Problem statement

Plasma-enhanced chemical vapor deposition (PECVD) is an essential process in semiconductor fabrication. In a high-temperature PECVD reactor furnace, gaseous chemicals are added to form a solid film on the surface. Stable control of temperature, gas flow, pressure, power, and other variables is essential for creating a film with consistent thickness to desired specifications.

An engineering team is looking to better understand the impact that gas flow and power inputs have on the resulting thickness of the thin film. The team designs an experiment that will vary gas flow between 240-300 standard cubic cm per minute, high-frequency power between 750-850 watts, and low-frequency power between 180-220 watts. To study the impact each factor might have, along with their possible interactions, and a potential curvature effect, a three-factor, two-level design with center points was chosen. There are two experimental runs at each of the eight corner points and four runs at the center point, resulting in a total of 20 experimental runs. The order of the runs was performed randomly. For each run, five thickness measurements were made on the wafer. The average and the standard deviation in thickness will be used as the outcome variables for the analyses.

The target thickness in this process is 6,650 angstroms, though it is still considered acceptable if the average thickness is between 6600-6700. Consistency across the wafer is also an important goal. The specification on the variation in thickness within a wafer is a standard deviation <220 angstroms.