Yield Loss Classification: Random Defects vs. Systematic Defects

That Day OEE Dropped Below 70%, and the Boss Was Furious

That day was truly a "turbulent" day. In the morning, the boss rushed in, his face ashen, asking: "Why has the OEE for machine X dropped below 70%? Yield is also falling, go investigate!" The air in the conference room froze; everyone looked at each other, no one dared to speak. To be honest, we've all encountered this situation: yield starts to become unstable, but no one can explain why, so we can only resort to trial-and-error investigations from scratch. Yield, to put it plainly, is money. Every percentage point drop could mean hundreds of thousands or millions in losses. At this point, your mind should start running through whether the problem is random or systematic.

Where's the Problem? It's Either Bad Luck or a Broken Machine

Simply put, yield loss can generally be divided into two types: Random Defects and Systematic Defects.

1. Random Defects: You can imagine it like throwing darts today; although most hit the bullseye, occasionally your hand slips, hitting the outer ring or even missing the board entirely. This type of "occasionally occurring," without a fixed pattern, is a random defect. It could be a particle falling onto a wafer, a sudden electrostatic discharge, or minor fluctuations during the process. Its occurrence location is random, and while its quantity might vary, there's no fixed pattern.
2. Systematic Defects: Imagine the tip of your dart is bent. Every time you throw it, no matter how you aim for the bullseye, it consistently veers towards a specific point on the right. This is a systematic defect. It has a fixed pattern, a fixed location, and usually recurs consistently. It could be due to a worn-out machine part, a sensor reading drift, or a process parameter consistently deviating from its set value over a long period.

The key point, therefore, is that these two types of defects have entirely different "personalities," and their solutions are vastly different. Random defects might require improving cleanliness and reducing particle generation; systematic defects, on the other hand, require identifying the root cause and fixing it from the source.

How to Actually Do It? Looking at Patterns is Most Accurate

To identify these two types of defects, the most commonly used methods are looking at "Maps" and "Trend Charts."

1. Looking at Maps: Plot the defect locations. If defect points are scattered across the entire wafer without specific concentrated areas, it's very likely a random defect. However, if defects are concentrated in a particular quadrant, along an edge, or consistently appear at the same location on the wafer every time, it's almost certainly a systematic defect. We once encountered a situation where yield suddenly dropped from 98% to 95%. Upon checking the map, behold! All defect points were concentrated at the 12 o'clock position on the wafer, consistently. An investigation revealed that the machine arm scratched that specific location when gripping the wafer.
2. Looking at Trend Charts: Suppose the Cpk of a critical parameter suddenly drops from 1.67 (yield DPMO of only 3.4) to 1.08 (DPMO as high as 6210), and this occurs for several consecutive batches. This is definitely a red flag. If the yield fluctuates up and down without a clear direction, it might be due to random fluctuations; but if the yield consistently drops or remains at a low point, it indicates a systematic problem at play.

In other words, random defects are like a common cold, occurring occasionally and remedied with medicine and rest; systematic defects are like a chronic illness, which will keep recurring unless the root cause is identified and addressed.

The Most Common Trap: Treating the Symptom, Not the Cause

Frankly, many times we make the mistake of treating systematic defects as random defects. I recall one instance where the OEE report showed a particular machine's yield was 2% lower than average for three consecutive weeks. Initially, everyone said, "Ah, it's probably just too many particles, let's try a few more cleans." However, after several cleans, the yield still showed no improvement.

Later, upon careful retrospective analysis, it was discovered that a gas flow controller (MFC) on the machine had drifted by 5% in its reading, leading to a long-term insufficient supply of process gas. Cleaning the machine's surface was useless for this problem, as the root cause lay within the internal control system. This "treating the symptom, not the cause" approach not only wastes time and resources but also allows the real problem to continue worsening. Therefore, when yield shows abnormality, the first step is truly to determine whether it is random or systematic. If this initial judgment is wrong, all subsequent efforts might be in vain.

One Thing You Can Do Today

Open your yield report and check if the defect Map shows any patterns.