That Day, Measurement Data Was a Mess, and the Boss's Face Was Whiter Than the Cleanroom Walls
I still remember years ago when our factory acquired a batch of new etching machines. Everyone was excited, thinking our production capacity would skyrocket. But what happened? After the first batch of products ran, we took them for measurement. When the data came out, wow, some parameters floated into outer space, while others directly went out of spec. The CPK report was released, and the whole room fell silent for three seconds; the boss's face was whiter than the cleanroom walls. Everyone exchanged glances, no one knew where the problem was. It was only later that we realized we hadn't properly established the "performance baseline" for these new machines, and we had no idea of their true characteristics.
Where the Problem Lies: Do You Really Understand Your Equipment?
Simply put, your machine is like your girlfriend (or boyfriend)—you think you understand her well, but in reality, you only see her outward behavior and don't know her "normal" limits. An equipment performance baseline, in plain language, is setting a standard for a piece of equipment's "health status" and "stable performance." When you say this machine is stable, what do those stability numbers look like? What is the tolerance range? Is it a CPK of 1.33 or 1.0? This isn't based on feeling; it must be spoken through statistical figures.
For instance, with our previous batch of etching machines, we initially failed to identify their etching rate fluctuation range. We thought ±5% was acceptable, but when running stably, it could actually achieve ±2%. This ±3% difference is critical. Without a baseline, you wouldn't know if the machine is "truly unstable" or "that's just how it is by nature."
How to Actually Do It: Observe First, Then Define
1. Run Enough Data:
Frankly, there's no shortcut here; you just have to run it. You need to let the equipment run enough batches under "ideal" and "stable" conditions to accumulate sufficient data. We usually recommend at least 30 batches, or even more; if it's a critical process, 100 batches is not too many. For example, for a machine's temperature control, you can't just look at one day; you need to observe its performance over several weeks under different loads.
2. Analyze with Statistical Tools:
Once there's enough data, you can start analysis. The simplest and most commonly used approach is to calculate the Mean and Standard Deviation. These two numbers give you the central point of the equipment's "normal" performance, as well as its "fluctuation range" around this central point.
- Example: We previously had a CVD machine, and after stably running 50 batches, its film thickness was measured. The mean was 1500 Ångströms, and the standard deviation was 5 Ångströms.
- The Key Point Is: This tells you that the film thickness of this machine will, most of the time, fall within the range of 1500 ± 3*5 = 1500 ± 15 Ångströms (i.e., 1485 to 1515 Ångströms). This is its "inherent characteristic," its performance baseline.
3. Set Control Upper and Lower Limits:
With the mean and standard deviation, you can then set the upper and lower limits for your control chart. Typically, the mean ±3 standard deviations are used as Control Limits. If your data points fall outside this range, you should be alerted, as it indicates the machine might be experiencing issues.
- In Other Words: For the CVD machine just mentioned, its film thickness control upper limit is 1515 Ångströms, and the lower limit is 1485 Ångströms. Next time you run a product, if the film thickness is 1520 Ångströms, you'll know, "Oh, something's not right here," and you'll need to check if there's anything unusual with the machine.
The Most Common Pitfall: Trying to Define a Baseline Right After Installation? Overthinking It!
The biggest pitfall I've experienced is rushing to define a baseline right after equipment installation or maintenance, when the machine is still in its "break-in period" and the data is fundamentally unstable. At that time, data was fluctuating wildly, and I naively thought that was its "normal performance," which, of course, led to setting an incorrect standard. I later learned that you must wait until the machine operation is stable and all parameters are optimized before beginning data collection.
Another pitfall is trying to define it with insufficient data. Claiming a baseline after only five batches is simply gambling. The power of statistics comes from a large number of samples; an insufficient sample size will lead to misjudgment.
One Thing You Can Do Today
Choose one piece of equipment you are responsible for, identify one of its critical parameters, collect data from at least 30 consecutive batches, and calculate its mean and standard deviation.