That day, when the CPK report came out, there was three seconds of silence across the room, and DPPM surged.
"Allen, what's going on with this batch of wafers? DPPM has exceeded 1000 for two consecutive weeks!" The production manager's call was like a relentless barrage, and my mind started racing. DPPM over a thousand? This is simply a disaster. I rushed to the site, only to see the production line machines stopped, hundreds of wafers awaiting testing piled up, and engineers looking pale. The supplier kept insisting that their outgoing quality met specifications, yet our production line continuously faced problems. Honestly, every time something like this happens, I want to hit my head against a wall. Where exactly did things go wrong?
What's the Problem? What Exactly Does DPPM Measure?
Often, you might hear a supplier say, "Our CPK is above 1.33!" Sounds great, right? But frankly, CPK is a metric that evaluates process capability stability; it does not equate to the defect rate. If, for instance, a parameter from your supplier has a specification upper limit of 100 and a lower limit of 90, and their average is 92 with a very small standard deviation, their CPK will naturally look excellent.
However, if your process encounters issues whenever that parameter exceeds 95, then even within the 90-100 specification range, your yield will be significantly impacted. This is where DPPM (Defective Parts Per Million) becomes crucial. It directly tells you how many defective items there are per million products. Frankly, DPPM is your most critical pain point because it directly impacts your production costs and delivery times.
How to Do It in Practice? Redefining Your "Defect"
Since DPPM is so important, how should it be designed and tracked?
- Redefine "Defect": You must sit down with your supplier and clearly articulate your process's sensitive points. For instance, the thickness of our wafers has a supplier specification of 500±10um. However, if the thickness exceeds 505um, our machines have an 80% chance of causing subsequent process defects. In this scenario, your internal "acceptable range" is no longer 500±10um, but rather 500±5um.
- Establish an Early Warning Mechanism: We later negotiated with our supplier that, in addition to standard specifications, their outgoing inspection must specifically monitor sensitive parameters that affect our yield. If the values of these sensitive parameters fall outside our internal acceptable range, even if they are still within the standard specification, they should be considered "pre-warning defects."
- Tiered DPPM Tracking: We categorize DPPM into two levels.
* DPPM-B: These are pre-warning defects, indicating that while these components are still usable, they exceed our internal acceptable range and carry a higher risk of future problems.
Therefore, the key is to clearly define what constitutes an unacceptable "defect" for you and quantify it with numbers.
The Most Common Pitfall: Numbers Lie, and So Do People
The most outrageous thing I've encountered before is suppliers "adjusting" sampling standards to make DPPM numbers look better. For example, instead of sampling 5% per batch as originally required, they might only sample 1%, and specifically choose those that appear problem-free. This "head-in-the-sand" approach makes the data look good in the short term, but ultimately the mess lands back on your plate.
Another pitfall is focusing solely on final yield without dissecting which component caused the problem. Sometimes, when process yield is poor, you blame the machines or operators, only to discover upon investigation that a specific parameter of a critical material, though still within the supplier's specifications, has drifted into your process's "edge zone," causing daily issues.
One Thing You Can Do Today
Pick up the phone and redefine "defect" with your supplier.