InsightFab
Knowledge Base/Variance Analysis: Decomposing Process Variation
Statistical Analysis6 min read

Variance Analysis: Decomposing Process Variation

This article introduces Variance Analysis, a crucial method for dissecting process variation when performance deviates from targets. It explains how to systematically identify the sources of variation—such as machines, materials, or operators—providing a clear path to problem-solving and informed decision-making in manufacturing.

That Day the CPK Report Came Out, and the Room Fell Silent for Three Seconds

Remember a few months ago? Our newly opened Fab line encountered issues right after commencing mass production. During a management meeting, the CPK report was presented: 1.08. The entire room was silent for three seconds, then I heard Ah-Hong from the next department whisper, "Didn't they say all the equipment had been validated?" Frankly, 1.08 isn't terrible, but it's a significant deviation from our target of 1.33. The boss's face turned green, and he directly called out, "Ah-Ming, explain where exactly this variation is coming from?" At that moment, you know it's time for Variance Analysis.

Where Does the Problem Lie?

Simply put, Variance Analysis is about breaking down and examining every source of variation in a process, one by one. Imagine you're brewing a cup of coffee; each time, the strength comes out differently. Why? It could be inconsistent grind size, unstable water temperature, or inaccurate brewing time. These "unstable" factors are the sources of variation. In a semiconductor fab, this "coffee strength" might be your film thickness, resistance, or CD value. We need to know whether the greatest difference is caused by the machine itself, material batches, or operator technique.

How Is It Actually Done?

The most common application is ANOVA (Analysis of Variance). We typically select several factors that you suspect might affect quality, such as:

  1. Machine: Is there a difference in results when the same batch of wafers runs on different machines?
  2. Lot: Is there a difference in products made from different batches of raw materials?
  3. Operator: Is there a difference in results when engineers or OPs from different shifts perform the operation?

Suppose today we are focusing on a critical parameter, the Refresh Time of DRAM, with a target of 500ns. We ran three machines, used three batches of material, and processed five wafers for each combination. After collecting the data, we use statistical software to perform calculations. It will tell you what proportion of the total variation comes from the machine, how much from the material, and how much is random error. If the results show that the machine's P-value (p-value) is less than 0.05, this indicates that the "Machine" factor has a significant impact on Refresh Time. For instance, if your report shows that 70% of the variation comes from machine #3, then you know you should investigate the condition of machine #3.

Common Pitfalls

To be frank, when I first started learning, the most common mistake I made was trying to "analyze everything." I indiscriminately crammed every conceivable factor into the analysis, resulting in a deluge of data with large P-values, which only added to my confusion. I recall one time, while tracking an anomaly with a DPMO of 6210, I included machines, recipes, and even the day's temperature and humidity in the ANOVA. The result? No significant effects were found, and my boss sarcastically remarked, "Did you perhaps include what you had for lunch today too?"

Honestly, the key is to have a sufficient understanding of the process beforehand, to first filter out the factors you believe are most likely to have an impact, and to design a reasonable experiment. Otherwise, no matter how much data you have, it will just be "garbage in, garbage out." Also, don't forget to examine the Residual Plot; if the residuals are not randomly distributed, it suggests your model might be missing some important variables, or there might be issues with the data itself.

One Thing You Can Do Today

Go back and look at your latest anomaly report; think about where you believe the greatest variation might be coming from.

Want to try it yourself?

Every tool mentioned in this article is available on InsightFab — just upload a CSV to analyze.

Go to Tools

More in this category

Correlation Coefficient: Pearson vs Spearman – Linear vs Monotonic Relationships

This article sheds light on a common pitfall in engineering data analysis: the misapplication of correlation coefficients. It explains why relying solely on Pearson correlation, which primarily detects linear relationships, can lead to incorrect conclusions when data exhibits non-linear, monotonic trends, especially when addressing issues like unexpected product yield drops.

Post-hoc Test: Choosing Between Tukey vs Bonferroni

This article addresses a common production line challenge: identifying the specific source of a problem when overall metrics like CPK fall short. It introduces Post-hoc Tests as a crucial tool to pinpoint significant differences between individual machines or groups, enabling precise problem-solving and immediate action.

Statistical Power: How to Ensure Your Conclusions Are Meaningful

This article addresses a common challenge for engineers: drawing conclusions solely based on observed data changes without accounting for statistical significance. It emphasizes the importance of using 'Statistical Power' to validate findings, ensuring that conclusions are robust and well-supported, particularly when presenting to management.