That Day, the Cpk Report Came Out, and the Room Fell Silent for Three Seconds
I still remember years ago, our factory had a batch of new products going into mass production, and everyone was incredibly busy during that time. When the yield report for the first batch of wafers came out, the entire meeting room instantly fell quiet, with only the hum of the air conditioning audible. The PM's face was ashen, asking, "What the hell is this? The yield dropped so much, what went wrong?" The process engineer fumbled, eventually pointing the finger at the "reticle." It was then that everyone realized how critical reticle quality, especially CD and Overlay, was to mass production yield.
Where Was the Problem? The Reticle's "Myopia" and "Astigmatism"
To put it simply, a reticle is like the "negative film" in semiconductor manufacturing. We etch circuit patterns onto it, and then use an exposure machine to shrink and transfer these patterns onto the wafer. So, what is CD (Critical Dimension)? Simply put, it's the "width" of your circuit pattern. Imagine your design calls for a line 100 nanometers wide, but the line on the reticle turns out to be 105 nanometers or 95 nanometers—that's CD deviation. This is like the reticle having "myopia," making lines appear thicker or thinner than they should be.
What about Overlay? It refers to the alignment precision between patterns of different layers. To make a chip, we stack dozens, even hundreds, of pattern layers, and each layer must be perfectly aligned. If the Overlay is inaccurate, it's like the reticle having "astigmatism," causing the patterns to be misaligned. A line that should be stacked directly on top of another might end up shifted to the side, rendering the chip useless (GG). So, the key is that CD affects line width, and Overlay affects alignment. If either of these parameters drifts, it can lightly impact component performance or severely turn the product into scrap.
How Is It Actually Done? Let the Numbers Speak
In the factory, how do we determine if a reticle's CD and Overlay are acceptable? By looking at the numbers, of course!
- CD Control: We set a target value and an allowable range. For example, if the CD target for a certain line is 50nm, its specification might be set as 50nm ± 5nm. This means the actual measured value must be between 45nm and 55nm to be considered acceptable. We also look at the Cpk value, which represents the process capability index. If Cpk is only 1.08, it means approximately 6210 parts per million (DPMO 6210) will have issues, and a drop in yield is inevitable. Therefore, generally, we require Cpk to be at least 1.33 or higher, which would reduce DPMO to 63.
- Overlay Control: Overlay specifications are typically stricter than CD specifications. For instance, the Overlay specification for a certain layer might require ±5nm. This means the alignment error between your upper and lower layers of patterns cannot exceed 5 nanometers. If the measurement shows 8nm, it's immediately deemed NG (Not Good), and that wafer is usually scrapped directly. Even a slight Overlay deviation can lead to short circuits or open circuits. So, the key takeaway is that these numbers directly translate to real money in terms of yield; if the numbers aren't good, money is lost.
The Most Common Trap: Supplier Delivers Wrong Version, Engineers Don't Notice
The most outrageous incident I encountered was when a supplier delivered a batch of new reticles, and for one of them, the CD value was already shown to be at the edge of the acceptable range on the acceptance reports from the first few batches we received. At that time, the production line was rushed, and everyone, seeing the numbers within range, approved it for use. As a result, after hundreds of wafers were mass-produced, the yield suddenly plummeted! Upon investigation, it was discovered that the CD of that particular reticle, after prolonged use and due to slight changes in process conditions, had accumulated errors that exceeded specifications.
Honestly, this happens frequently. Sometimes, suppliers rush to meet deadlines and send out "marginal" reticles; other times, our own acceptance SOPs aren't strict enough, only checking "pass/fail" without evaluating "how well it passed." Frankly, if we had taken a little more time then to use the Cpk value as one of our acceptance criteria, instead of just relying on single measurement values, we might have avoided that disaster.
One Thing You Can Do Today
Next time you receive a reticle acceptance report, in addition to checking "pass or fail," take an extra look at the "Cpk value."