InsightFab
Knowledge Base/Time Series Analysis: Trends and Seasonality in Process Data
Statistical Analysis6 min read

Time Series Analysis: Trends and Seasonality in Process Data

This article offers practical insights into a past incident where a new process's CPK report was abysmal, frustrating management due to erratic data fluctuations. The author reminds us that engineers often focus solely on single data points or averages, overlooking the inherent 'life' and 'temperament' of process data, where what often appears to be random fluctuation is, in fact, attributable to discernible trends and seasonality. By understanding how to analyze data beyond isolated points, readers can identify the true root causes of issues, rather than being misled by superficial numbers.

The Day the CPK Report Dropped, and Three Seconds of Silence Fell

Do you remember? Years ago, we introduced an advanced process, full of confidence. But when the first weekly CPK report came out, everyone was stunned. The target was 1.33, but we got 1.08, and DPMO skyrocketed to 6210! The boss's face instantly turned colder than an icebox. I was scared too, thinking if some parameters weren't tuned correctly. I pulled the data from the machine, and wow, the data fluctuation was like an electrocardiogram, wildly oscillating, with no discernible pattern.

Where Did the Problem Lie?

To be honest, what we semiconductor engineers do most often is look at data. But many times, we only look at "point" data like the current Cpk or yield, or compare this week's "average" with last week's. However, process data actually has a "life" of its own. It evolves over time, with its own temperament and rhythm. You might think it's just random fluctuation, but frankly, many times it's not random at all; it's discernible trends and seasonality. It's like your daily commute: traffic is definitely jammed at 7 AM and 5:30 PM – that's seasonality. If a new bridge is suddenly built on that road, traffic flow might undergo a structural change – that's a trend. When we look at data, we often don't consider the dimension of time, resulting in mistaking trends for noise and seasonality for anomalies, which naturally leads to misjudgment.

How to Practically Apply This?

To understand the temperament of the data, the simplest way is to perform "time series analysis." Frankly, this isn't some profound, mysterious concept; you can create a basic version using Excel.

  1. Plot a Line Chart, Emphasizing the Time Axis: Don't just look at averages; connect the data points in chronological order. You'll then intuitively see whether the data is moving upwards, downwards, or oscillating around a certain value.
  2. Identify "Trends": If the line chart shows the data generally slowly increasing or decreasing, there might be a trend. For example, if your etching rate has been slightly slower each month since startup, this could indicate an equipment wear-and-tear trend.
  3. Identify "Seasonality": Process data often exhibits periodic fluctuations. For instance, if yield slightly drops every 8 hours (a shift) and then recovers, this might be related to shift-change operating habits. Or, if the yield of the first batch on Monday mornings is consistently lower, this could be due to machine idleness over the weekend. If you notice a peak in wafer surface particle count every 24 hours, it might be related to cleanroom personnel entry/exit or air circulation cycles.

The key, therefore, is to first grasp the "character" of your data through visualization, and then decide on the appropriate method for improvement.

The Most Common Pitfall

The biggest pitfall I've ever encountered was treating seasonality as an anomaly. Once, with a new process, our film thickness would deviate slightly from spec every day at noon and midnight. Being young and impetuous back then, I immediately assumed the machine was faulty upon seeing the data. After much effort, engineers and equipment suppliers disassembled and reassembled the machine thoroughly. What happened? It was only later discovered that the facility system adjusted the air conditioning temperature every day at noon and midnight, causing slight changes in the machine's internal environmental temperature, which affected process stability. This was a classic "seasonal" variation, not a machine malfunction at all. I wasted so much time and manpower for nothing.

One Thing You Can Do Today

Take your available process data, plot it as a line chart with a time axis, and observe it first.

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.