InsightFab
Knowledge Base/Computer-Aided DOE: Minitab vs JMP vs Python Practical Comparison
DOE6 min read

Computer-Aided DOE: Minitab vs JMP vs Python Practical Comparison

This article recounts a past incident of plummeting wafer yield and machine failure. The manager's directive to "run this DOE" highlighted the crucial role of Design of Experiments. DOE is presented as a highly practical, systematic method to identify root causes and critical parameters affecting product quality with minimal effort. Readers will learn to approach issues strategically rather than through trial and error, mastering the essence of DOE for enhanced efficiency.

That day the machine crashed, and the supervisor just said one thing: "You run this DOE."

Remember the end of last year, when the wafer yield for that new process plummeted like an avalanche? Let me tell you, that day a machine on the production line crashed completely. We worked overnight to repair it, and the next day, the engineers stared at that terrible Cpk report, and the whole room was truly silent for three seconds. The Cpk dropped from the original 1.35 to 1.08, and the DPMO figure soared to 6210. The supervisor looked at it for a long time, then finally patted my shoulder and said, "Alright, you run this DOE and see who's behind this." Damn, my scalp tingled after hearing that.

Where the problem lies: "Shooting in the dark" only makes things worse.

To put it bluntly, DOE (Design of Experiments) is a systematic method for identifying the root causes of problems. Imagine, when a machine malfunctions and the yield tanks, you can't possibly adjust every single parameter, right? That would be like shooting in the dark, guaranteed to waste both time and wafers. The essence of DOE is to use the minimum number of experiments to find out which parameters have the greatest impact on your product quality, and whether there are any interaction effects between these parameters. So, the key is that DOE helps you quickly filter out critical factors from a multitude of possible variables, allowing you to stop adjusting blindly.

How to actually do it? Minitab, JMP, and Python each have their strengths.

In the past, when we ran DOE, we either used Minitab or JMP; both are industry-standard statistical software.

  1. Minitab: The interface is intuitive and feature-rich, making it very user-friendly for beginners. You just input your experimental data, and it automatically generates main effect plots, interaction plots, and even optimization tools, allowing you to easily find the optimal parameter combination. For that yield problem, I first ran a full factorial design with Minitab, and it quickly pinpointed an issue with a certain etch time and temperature, and they even had an interaction effect!
  2. JMP: This is a product under SAS, with stronger graphical capabilities and better interactivity. If you're someone who enjoys dragging, clicking, and intuitively exploring data, you'll love JMP. It excels particularly in visual presentation, displaying complex statistical results with beautiful charts, making it convenient for reporting to your boss.
  3. Python: The rise of Python in recent years has also led many to use it for DOE. While it doesn't have a ready-made GUI where you can just click around like Minitab and JMP, Python offers tremendous flexibility. You can use libraries like `scipy.stats` or `statsmodels` for regression analysis and analysis of variance (ANOVA), or even use the `pyDOE` package to generate experimental design matrices. Honestly, Python has a steeper learning curve, requiring some programming foundation, but once you get the hang of it, you can customize almost any analysis you desire.

In other words, Minitab and JMP are like point-and-shoot cameras—powerful but with fixed frameworks; Python is like a DSLR camera—complex to operate, but you can capture any picture you want.

The most common pitfall: Getting flustered by P-values and forgetting practical significance.

Let me tell you, the most common annoying thing is when novice engineers run a DOE, see a bunch of P-values less than 0.05, and get incredibly excited, thinking they've struck gold. But then they actually go and adjust the parameters, and the yield doesn't improve at all. Why? Because they forget that a P-value is only about statistical significance, not practical significance. Sometimes, a parameter might be statistically significant, but its impact on yield might only be 0.01%, which is practically useless! Even worse, if you adjust the parameter, the process stability might deteriorate instead, and your Cpk will plummet right before your eyes. So, the key is that after data analysis, you must combine it with your process knowledge to determine which parameters are truly worth acting upon.

One thing you can do today:

Open Minitab or JMP on your computer, find some historical data, and run a regression analysis to see if you discover anything new.

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

DOE: Not Just Running Experiments, But Hitting the Most Targets with the Fewest Bullets

A full factorial experiment with four factors and three levels each requires 81 runs. Using a Taguchi L9 orthogonal array, only 9 runs are needed, saving 89%, while simultaneously estimating the main effects of all four factors without mutual contamination.

DOE Taguchi Method: Finding Optimal Parameters in 9 Experiments, Not by Luck

A full factorial design for 4 factors at 3 levels requires 81 runs, while a one-factor-at-a-time approach, though shorter at 36 runs, misses crucial factor interactions. The Taguchi L9 method offers a highly efficient alternative, completing the analysis in just 9 experiments and achieving an 89% reduction in experimental effort.

Response Surface Method: Finding the Optimal Process Sweet Spot

After DOE identifies the significant factors, RSM (Response Surface Methodology) helps pinpoint the precise parameter combination that optimizes the outcome. This marks a crucial transition from understanding 'which factors are important' to determining 'what settings are best'.