That day, the CPK report from the metrology tool came out, and the air in the entire fab froze for three seconds.
I still remember it was many years ago. That evening, I was about to leave work when my phone suddenly rang. The PM's tone was a bit urgent: "Sam, are you still in the fab? Come quickly to the metrology station. That... the Via CD report for Etcher M4, the CPK value is only 1.08, and DPMO has skyrocketed to 6210!" My heart sank halfway when I heard that. This was an alarm among alarms! You know, our usual target is a CPK of 1.33 or higher, and DPMO is ideally zero. Such a drop meant the yield must be terrible. I quickly rushed to the tool, looked at the data plot, and my face turned green. The PE next to me was also helpless, asking me, "Brother Sam, what should we do? Do we have to stop the entire line?"
What went wrong? What exactly is SPC looking at?
To put it bluntly, SPC (Statistical Process Control) helps us monitor whether the process has drifted. Imagine you're cooking; if you add roughly the same amount of salt every time, the taste of your dishes will be consistent. But if one day you suddenly add too much or too little salt, the taste will be off. The semiconductor process is the same; we want every photolithography, etch, and thin-film deposition to stably produce wafers that meet specifications.
So, what exactly is SPC looking at? Frankly, it's looking at "variation." Just like adding salt, you might be slightly off each time, but as long as that "slight deviation" is within an acceptable range, it's fine. Once that "slight deviation" becomes too large, or the entire mean shifts, SPC will trigger an alarm. And what we engineers need to do is find out where this variation is coming from. So, the key is that SPC shows you anomalies, but you have to interpret them.
How is it done in practice? How do you identify critical parameters?
You must be asking, with hundreds of process parameters in the fab, do we have to monitor every single one? Of course not! That would exhaust you. In practice, our engineers identify "critical parameters." These parameters typically are:
- Parameters that directly impact product yield: Like the Via CD (post-etch hole diameter) we just mentioned, this directly affects electrical properties, so it must be monitored. Also, film thickness and photoresist Line Width directly affect chip functionality.
- Parameters most prone to going out of control in the process: Some tools, you just know they are more "stubborn" and prone to issues. For example, the Etcher's RF Power, Chamber Pressure, or the Stepper's Dose—even small fluctuations in these can easily cause the process to drift.
For instance, in the case where our Via CD dropped to a CPK of 1.08, we later discovered that there were minor fluctuations in the etcher's RF Power, leading to an unstable etch rate. Although the RF Power readings appeared to be within specifications each time they were measured, the SPC chart told us that its "variation" had increased. In other words, even though the amount of salt added each time was still within the "salty" and "not salty" range, this time the amount added varied greatly from batch to batch, causing significant differences in the taste of each dish. So the key is not just to see if a single value is out of spec, but more importantly, to look at its "stability."
The most common pitfalls, don't blindly fall into them.
The most common pitfall I've seen is "lots of data, but don't know what to look at." Often, engineers get overwhelmed by the vast amount of data generated by the tools and end up "like blind men touching an elephant." They think that putting all parameters on SPC charts will be foolproof. The result? A barrage of alarms, but they're all "false alarms," wasting time investigating processes that had no actual issues.
Another major pitfall is only looking at CPK and DPMO. While these two metrics are indeed important, they are "outcomes." More importantly, you need to go back and look at the SPC charts of the "process parameters." If you only look at the results, it's like only looking at final exam scores without reviewing your regular study methods. When you see CPK drop, you need to quickly check which critical process parameter's SPC chart is showing problems so you can prescribe the right remedy. To be honest, there are no shortcuts; it's all about accumulating experience and learning from those "old timers."
One thing you can do today
Go back and review the processes you are responsible for. Select three critical parameters that you find "most impactful," pull up their SPC charts, and observe their recent trends.