That Cpk 1.08 Report Sent Shivers Down the Production Line's Spine
I still remember a few years ago when a new process finally completed its P-Run, and the report came out with Cpk 1.08. All the engineers in the room exchanged glances, followed by three seconds of silence. This number, neither good nor bad enough, was sufficient to get you stuck but not enough to scrap and restart. The boss's face turned grim, and he directly asked, "Do you even have Gage R&R results?" The QA lady softly replied, "Yes, RR% is 15%." The boss frowned, turned to me, "A-Zhong, didn't you say Measurement Uncertainty was more important? What's your take now?"
You see, this is everyday life in a semiconductor fab. A single number can impact millions, tens of millions in output value. Today, let's discuss Measurement Uncertainty (MU) and Gage R&R—these two terms you've surely heard of but might find a bit confusing—and what their specific practical differences are in a semiconductor fab.
Where's the Problem: What ISO Says Doesn't Count, But What AIAG Says Does?
To put it plainly, Measurement Uncertainty is promoted by the ISO system. It emphasizes "how much confidence interval your measurement result has." For example, if you measure 100 nanometers, then the true value of this 100 nanometers might be between 99.5 and 100.5. MU considers a wide range of factors, from instrument calibration, ambient temperature, operator skill, to even the error of the standards you use—all are included in the calculation. The goal is to provide a complete description of "measurement result ± uncertainty."
Gage R&R, on the other hand, is the methodology from AIAG (Automotive Industry Action Group). It is more practical and focuses on "whether your measurement system can distinguish differences between different products." Simply put, it wants to know how much of the variation you measure comes from the product itself, and how much comes from your measurement instrument and operator. That's why you see the RR% figure; it gives you a percentage, telling you how much of the total variation is caused by the measurement system.
So the key point is, MU tells you "how accurate your measurement value is," while Gage R&R tells you "how stable and discriminating your measurement system is." In fields like semiconductor manufacturing, which pursue extreme precision, both are important, but their application scenarios and focus points differ slightly.
How It's Actually Done: First Gage R&R, Then Measurement Uncertainty
In our fab, it's usually handled this way:
- When introducing new equipment or developing new processes, Gage R&R is a must-run. You must first ensure your measurement system is "usable." If the Gage R&R result shows an RR% exceeding 30%, then I'm sorry, your measurement system is fundamentally unstable, and the data measured has no reference value whatsoever. At this point, let alone Cpk 1.08, no one would believe Cpk 0.5 either. We typically require RR% to be below 10%; the lower, the better.
- Only after Gage R&R passes do we further consider Measurement Uncertainty. Especially for processes hovering at the edge of specifications, or products requiring extremely high precision from customers. For example, if you measure a critical dimension, and although the Cpk is 1.33, the customer requires DPMO to be less than 100. In this situation, if your Measurement Uncertainty is too large, even if the Cpk meets the target, your actual product defect rate might still be very high. We would then evaluate how much this Measurement Uncertainty impacts Cpk, and even go back to request instrument suppliers to improve accuracy.
Take the Cpk 1.08 case from last time as an example. Our Gage R&R was indeed 15%, which seemed acceptable. However, when we later considered Measurement Uncertainty, we found that due to the influence of ambient temperature, the actual Cpk could drop to 0.95. This was bad; what was originally a minor roadblock turned into a major problem. So, we later installed a temperature control system to reduce environmental variation, and only then stabilized the Cpk.
The Most Common Pitfall: Focusing Only on RR%, Not MU
The most common pitfall I've encountered is when engineers only know about Gage R&R, or even just know to produce a "qualified" RR% and consider their job done. The result is that the measurement system appears stable, but because Measurement Uncertainty was not evaluated, some products on the edge of specifications, although measured as OK, were actually NG (Not Good) products, eventually flowing to customers and leading to complaints.
There's also another situation: new engineers, whenever they see "accuracy ±X%" written on instrument specifications, assume that this is the Measurement Uncertainty. Honestly, what's written in instrument specifications is usually just the instrument's own repeatability, which is far from true Measurement Uncertainty. True MU requires incorporating all possible variations that could affect measurement results, such as environment, operation, and calibration. This requires a more comprehensive and systematic evaluation, not something that can be resolved by just looking at an instrument's specification.
One Thing You Can Do Today
Go back and look at your measurement data. Besides Gage R&R, have you evaluated Measurement Uncertainty?