In the semiconductor industry, within wafer non-uniformity (WIW-NU) has been the most critical indicator to quantify film performances. Produced wafers should meet the desired film properties homogeneously. We are facing a huge challenge to improve process tuning due to lack of smart WIW-NU metrics.

The baseline WIW-NU metric can only provide the wafer map and basic stats (e.g., mean, stdev., and range), which cannot identify the true root cause. There is an increasing demand to establish a smarter WIW-NU metric to shorten process tuning cycles. 

JMP is used to breakdown the total WIW-NU variations into four process failure modes: radial, angular, off-centering, and outlier. The following methodology was deployed:

• Transform raw data to create desired descriptive statistics.
• Define WIW-NU criteria.
• Conduct hypothesis mean test, normality test, and normality violation modes.
• Conduct GRR and Cpk/Ppk (wafer is the rational subgrouping).
• Establish I-MR by Phase (wafer), Xbar-R to capture special variations.
• Utilize modern data mining to extract WIW-NU insights.
• Derive predictive models to guide film process tuning.

This JMP WIW-NU project has significantly improved process tuning efficiency and shortened the process development cycle time. We are proliferating these methods across all internal organizations to improve overall performances and enhance competitive advantage.