Scatter And Histogram

Scatter

Use scatter when you need correlation, clustering, or direct interaction with limits.

Good use cases:

• compare two parametric tests
• isolate outlier clouds
• inspect site or wafer-separated clusters
• drag limits and watch local impact before committing

Key points:

• WebGPU handles the large-point path when available
• brushing is rectangular in the full scatter workflow
• live what-if previews update the linked context before commit

Per-wafer trends

The DUT-index Trend card and expanded view mark wafer boundaries and show a separate median line for each wafer. Use Per-wafer overlay to hide the medians or add the middle 50% band. This is the interquartile range, not a confidence interval. The existing By Wafer toggle changes point colors independently.

Medians and bands use all finite fetched results before canvas point sampling. When the dataset is capped, Fetched sample means the overlay also describes that sample, not exact full-population quantiles. Null results do not enter the quantiles, but keep their row-index positions. With only one finite result, the band collapses to the median line. Brushing and what-if limits do not change this reference population.

The horizontal axis is row order, not time. Lines never connect one wafer’s median to the next. Filtering to one wafer keeps its median and removes the inter-wafer boundaries.

In the expanded trend, zoom on the main plot, then drag the overview viewport to pan horizontally and vertically. The cursor changes from grab to grabbing. Click elsewhere inside the overview plot to recenter. Overview navigation stays within the data bounds and never changes die selection or what-if limits, even with Shift or Alt held. Releasing outside the overview ends the drag. At 1x or below, the whole domain fits, so there is no smaller viewport to move.

Histogram

Use histogram when you need shape, tails, limit context, or a simpler explanation of a single test.

Good use cases:

• detect a fat tail or split distribution
• compare the current selection to the lot-wide shape
• decide whether a limit change affects a meaningful population

The CPK table’s Tail diagnostic compares the observed canonical fail rate with the out-of-spec rate predicted by a normal population with the same mean and overall sigma. A large positive ppm difference can expose sparse flyers or heavy tails beside an otherwise healthy-looking CPK. Select the diagnostic to open PAT review for that test; no limit changes are made until you explicitly apply a proposal.

How the two work together

• scatter tells you whether a problem is structured
• histogram tells you how much of the distribution is affected

If you only look at one, you can miss either the structure or the volume.

Practical limit-review pattern

1. start in histogram to understand the failing tail
2. open scatter to see whether the tail belongs to a real population
3. drag candidate limits only after the population makes sense
4. use impact review before treating the change as a proposal