Yield Trend Dashboard

Yield is never a flat line — it varies lot to lot from countless small causes, and the central skill of yield management is distinguishing that normal wiggle from a real change that demands action. Overreact to noise and you chase ghosts and disrupt a healthy line; miss a true shift and you scrap wafers. Statistical process control exists precisely to draw that line, and it does so with the data's own variation: the control limits at mean ± 3σ mark where normal variation ends and a signal begins.

But control limits alone have a blind spot. A single dramatic excursion — a lot crashing to 75% among 92s — inflates the computed standard deviation so much that the limits widen and the crash falls just inside them, unflagged. That's why a robust dashboard pairs the statistical band with an absolute yield floor: the floor catches any point below a hard threshold regardless of what it does to sigma, so excursions can't hide. Using both is the difference between a chart that looks rigorous and one that actually catches problems.

The other thing control charts miss is the slow killer: a gradual decline that stays inside the limits for weeks while quietly bleeding yield and money. A point-by-point chart never alarms, yet the line is sinking. A trend slope — a least-squares fit across the series — surfaces that drift and projects where it's heading, turning a creeping loss into a visible, actionable number before it becomes a crisis.

And every bit of this is ultimately about speed, because detection time is scrap. Each lot run after an excursion begins and before it's caught is wafers at risk, so the value of fast trend and excursion detection is literally the cost of the material it saves. When an alert fires, the investigation moves to the spatial signature in the Wafer Map Analyzer and the dominant category in the Root Cause Finder, with the saved material priced in the Wafer Scrap console.