Wafer Map — User Guide¶

This guide describes the display and analysis features of the wafer map viewer. It is written for users who may be semiconductor test engineers, device engineers, and yield engineers or anyone else who may use a wafer map application — not for developers integrating the wafermap library.

Some features depend on what data the application has loaded (bin names, test definitions, test limits, reticle geometry). Where this applies it is noted.

Getting oriented. The wafer map is an interactive viewer. A toolbar is always present above the map, at the right of its title row — use it to change what the colours represent, toggle overlays, rotate or flip the wafer, zoom, select dies, and open the summary panel. Hover any die for a tooltip; the panels and tooltips always report the original die grid coordinates, never the on-screen position after a rotate or flip.

1. Reading the map¶

1.1 Die grid and coordinates¶

Each square on the wafer represents one die. The position labels you see — in tooltips, axis ticks, and selection readouts — are die grid coordinates: usually the X and Y step indices from the prober (integers such as −7, 0, 5). They are not millimetre values.

These coordinates are always the original grid values. Rotating or flipping the display does not change the coordinate labels — a die at (3, −2) always reads (3, −2) regardless of how the wafer is oriented on screen.

1.2 Wafer orientation¶

The notch (shown as a V-notch or flat edge) marks the physical reference edge of the wafer as configured. Use the Orientation toolbar controls to rotate or flip the display to match your convention; die coordinates are unaffected. The first rotate or flip also switches on the XY indicator (the +X/+Y arrows in the Overlays menu), so you can see which way the die coordinates now run. It stays on until you switch it off there; Reset orientation leaves it as it is.

An example wafer bin map above and below the same wafer rotated 90°. The notch has moved, but die coordinates — shown in tooltips — remain their original grid/prober values.

1.3 Die appearance¶

A no-data die is not a fail; it simply has no result recorded. Edge-excluded dies are shown dimmed and are not counted in yield calculations. Dies the filter drops are drawn faded and are counted nowhere: not in the legend, the yield, the Summary panel or the findings.

1.4 Legend and colour bar¶

Bin modes show a discrete colour legend: one swatch per bin, with bin number and name if the application has supplied bin names. Passing bins are listed first, then failing bins from the most dies to the fewest — the same order as the Summary panel and reports.

Value mode shows a continuous colour bar: the colour scale runs from the minimum to maximum value, with units when available. The colour bar is informational only — clicking it has no effect. To switch to a pass/fail view, use the Limit pass/fail or Test pass/fail option in the Overlays menu (available when test limits are defined, or a recorded verdict exists, for the active test).

Limit pass/fail mode replaces the colour bar with a small legend: Pass, Fail high, and Fail low swatches (only the categories that apply to the test's limits) with a die count beside each. This judges dies against the test's lower and upper test limits.

Test pass/fail mode replaces the colour bar with a Pass / Fail legend and die counts, coloured by the tester's own recorded verdict for that test — not a limit judgement. A test with no measured value (a functional, go/no-go test) always displays this way; selecting it switches the map into Test pass/fail automatically, since there is nothing to plot on a gradient.

Every map also shows a short title by the colour bar or legend naming what is displayed — the test name (and number, in limit pass/fail mode), the bin type, or the stacked wafer count.

Clicking a bin swatch in the legend filters the display to that bin: every other bin is greyed out. Ctrl / Cmd + click adds or removes further bins, so several can be shown together. Clicking the only bin shown again clears the filter (see Highlight bin).

1.5 Wafers and dies with no position data¶

Not every die necessarily has a reported grid position — some data sources supply bin or test results with no X/Y coordinates at all, for some or every die on a wafer.

A fully positionless wafer never renders as a map or gallery card — showing dies at fabricated positions would risk being misread as real spatial layout. Instead the card shows a compact summary matching the active plot mode: a bin breakdown (coloured the same as a positioned card's own bin legend) for hard/soft-bin modes, or a histogram for value mode — coloured through the same colour scheme, log-scale, and limit/data-range settings the map itself uses, so switching those in the toolbar updates the chart the same way it would a real map. A View die list toggle switches to the full per-die table (position/site, hard bin, soft bin, every test value) with its own CSV export; View chart switches back.

A mixed wafer — some dies positioned, some not — renders its normal map for the positioned dies, plus an expandable "+N dies without position data" footer beneath the card. Expanding it (click the footer or its chevron) shows the same chart/die-list toggle, scoped to just the unpositioned subset.

The toolbar's spatial-only controls (zoom, pan, box select, save image, orientation, overlays, legend position) are hidden on a fully positionless card, since there's no map for them to act on. Plot mode and colour scheme stay available — both still drive what the summary shows.

Findings that depend on physical layout — edge ring, quadrants, sectors, reticle position, cluster and pattern detection — only ever consider positioned dies, so a positionless wafer contributes none of these. Everything else — yield, bin counts, per-test statistics, and the Insights tab's histograms/correlation/scatter — still includes every die, positioned or not.

2. Plot modes¶

The active plot mode determines what the colour of each die represents. Use the Plot mode toolbar dropdown to switch. Available modes depend on what data the application has loaded.

Stacked modes are only available on lot-stack maps (multiple wafers combined into one display). The panel identifies how many wafers were stacked and which aggregation method is active.

Test Value mode — colour bar range and test limits¶

When test limits are defined for the active test, two additional display options become available:

• Colour bar range — Data / Limits: switches the colour bar scale between the actual data extent and the test limits. This affects only the colour bar's range, never how failing dies are shown. In both ranges every die is coloured by the gradient so you can read the value distribution, and dies outside the test limits are marked with a triangle — pointing down (▽) for below the low limit, up (△) for above the high limit — so they stand out without leaving the distribution. The triangle is drawn black or white per die for contrast against its own colour, so it stays visible under any colour scheme, and its shape (not colour) carries the below/above-limit meaning. On dies too small to carry one (under 5 pixels, as on a dense wafer zoomed out) the triangles are left off and the colour bar says ▽△ limit marks: zoom in; zoom in and they appear.
• Limit pass/fail: when active, dies within the test limits are shown in a pass colour and dies outside them are highlighted — blue for below the low limit, red for above the high limit. Both flags apply independently; a die can be flagged on either or both limits. A Pass/Fail legend replaces the colour bar, listing each applicable category with its die count.

Some tests have no measured value at all — a continuity check or any other go/no-go test, where the only result is a recorded pass or fail. Selecting one of these functional tests as the active test switches the map into Test pass/fail automatically: a Pass/Fail legend by die count, coloured by the tester's own recorded verdict rather than a limit judgement (there is no gradient to fall back to). Test pass/fail is also available as an Overlays option on an ordinary parametric test when it carries recorded verdicts, as an alternative to limit judgement.

Values outside a limit set¶

A tester that runs out of range records its rail instead of a reading: a current held at the compliance clamp, a voltage at the supply, an open-circuit value. A test can carry validity limits, the range a real measurement lies in, and a value outside them is not a reading. It is missing for that test on that die: grey on the map, and left out of the statistics and every chart, so one clamp cannot stretch the colour scale or move the mean, σ and Cpk. Which limit set a value must lie inside is the host's choice. The usual one is the validity limits, which only affect tests that have them; a host can instead keep only values inside the specification limits or the test limits, or filter nothing.

Nothing is hidden. Hover a grey die and the tooltip names the excluded value and the limit set. The Summary panel's Test Values table shows the total beside N and an Excl. column, and the same count appears in the histogram and box plot captions, the capability tooltip and caption, the My plots tab's footnote, the report and the CSV exports. The build also lists it among the data warnings. The bin map and each die's pass/fail are the tester's own and do not change, so yield is the same with the filter on or off.

Derived tests (†)¶

A test name marked † in front is a derived test: its value was calculated from other tests on the same die — a difference, a ratio, a margin — rather than measured by the tester. It behaves like any other test: you can map it, chart it and find patterns in it.

The dagger appears wherever the test is named — the map title, tooltips, the plot-mode test list, Insights charts and their test pickers, findings, the Summary panel, the die list and reports — always in front of the name, so in a list the derived tests line up in a column down the left edge. The words † Derived, not measured always appear nearby. Hovering over a derived value, or reading the key line under a table or findings list, shows the expression it was computed from, e.g. t[1020] - t[1010]. The numbers in square brackets are the test numbers it reads. In a CSV export the expression appears in a Derived from column instead.

3. Toolbar controls¶

The toolbar sits above the map, at the right of its title row (each card in a gallery has its own), and is always visible. Controls that are not applicable to the current mode are hidden automatically. (Hovering the map shows die tooltips — a separate thing from the toolbar, which is always present.)

Single map¶

	Control	Description
	Plot mode	Switches the active plot mode (see Plot modes). When multiple tests are available, a test selector appears alongside it.
	Overlays	Check-menu of optional display layers: axis labels, XY axis indicator, ring boundaries, quadrant lines, die coordinate labels, reticle grid (when geometry is configured), Limit pass/fail (Test Value mode with limits), Test pass/fail (Test Value mode, active test is functional or has recorded verdicts), and Mark failing dies (bin modes only). Compact layout and Layout diagnostics sit at the top of the menu (see Compact layout). Clear overlays at the foot turns off everything in the menu at once, and is greyed when nothing is on — so the menu also answers "is anything active?" without you auditing every row.
	Colour scheme	Picks the colours for the map on screen. In Hard Bin and Soft Bin modes it lists bin colours (Default, Colour-blind safe); in every other mode — including the stacked modes, which show values — it lists value colours. The two are separate choices, so changing one never resets the other. Value gradients run dark at the low end and bright at the high end, so on a stacked map the healthy wafer sits back and a defect ring or scratch stands out; Reverse gradient flips that when the low end is the one you care about. When bin definitions carry their own colours, Use colours from bin definitions turns them on or off. (Mark failing dies is in Overlays, beside the pass/fail display it belongs with.)
	Log scale	Test Value and Stacked Test Values modes only. Applies a log₁₀ scale to the colour mapping. Only active when all displayed values are positive. Hidden whenever a pass/fail display is active or the active test is functional.
	Colour bar range	Test Value mode with test limits only. Toggles the colour bar between the limit range (default — the colours mean the same thing on every wafer, so maps are comparable) and the data range (stretches the scale to the values actually present, which shows more contrast but is not comparable between wafers).
	Legend style	Bin modes only. Controls where the bin legend is positioned relative to the map: Default (right), Compact, Left, Top, Bottom, or Floating.
	Orientation	Menu of display transforms: Rotate 90° clockwise (applies cumulatively), Flip horizontal, Flip vertical, and Reset orientation. The first rotate or flip also switches on the XY indicator (see Overlays), which stays on until you switch it off. These change only how the wafer is drawn — die coordinates in tooltips and labels are always the original values, whatever the orientation. Reset is worth knowing about: rotation and mirroring do not combine in the order you applied them, so undoing a few clicks by clicking again does not reliably get you back — Reset does, and is greyed when you are already there.
	Zoom mode	Click and drag to draw a zoom region.
	Zoom in	Zooms in one step.
	Zoom out	Zooms out one step.
	Reset zoom	Returns the map to the default fitted view.
	Pan mode	Click and drag to pan the map.
	Box select	Click and drag to select a rectangular group of dies (see Box select). This is the mode a map opens in.
	Chart	Opens a menu of charts drawn from the selected dies, or from the whole wafer when nothing is selected (see Charting dies and wafers). Only shown when there is something to chart.
	Expand	Opens the map in an enlarged modal overlay. Opening the Summary panel in it widens the modal by the panel's width, so the map keeps its size. A maximise button in the modal grows it to fill the window (or press F). Press Esc or click outside to close. Useful for detailed inspection without changing the main view. Works in the Insights view too, where it opens the whole chart suite in a wide modal — useful because those charts interact and are best read side by side. Each chart can also be expanded on its own, with the expand button beside its ⋯.
	Save image	Downloads the current map view as a PNG. Captures the canvas as displayed, including all active overlays and the legend.
	Data warnings	Appears only when there is something to report about the data behind the map. Click it for the details. A red ⛔ means the map may be positionally wrong — usually that wafer geometry was guessed rather than supplied, so dies may not sit where they appear to. An amber ⚠ means something expected is missing or was skipped, but what is drawn is correct.
	User guide	Opens this guide.

Filter… is the last button in the row, at the right edge, with the Summary button and the Maps | Insights switch before it; it does not move between the map, the gallery and Insights. It narrows the map, the gallery and every Insights tab to the wafers and dies that meet conditions you set (see Filtering wafers and dies).

This guide's own window (and a gallery card detached into its own window, see Section 3 — Gallery below) is a floating window, not a modal — its header shows collapse, maximize, and close buttons, and it can be dragged and resized by its corner grip. Collapse shrinks it to a small title strip without closing it; click the same button (now Show contents) to restore it. While the window is maximized only restore and close are shown.

Highlight bin — in bin modes, click any bin swatch in the legend to highlight that bin and dim all others. Click again to clear. Useful for isolating a specific failure category across the wafer.

Bin 2 (Fail) highlighted — all other bins are dimmed.

Bin colours — a bin's colour tells you whether it passed before you read the legend: passing bins are always shades of green and failing bins never are, whichever bin numbers your test program uses. Each bin number keeps its colour: bin 7 is the same colour in every lot, every gallery and every screenshot, however many dies it holds, so you can learn your program's colours. Hard and soft bins are coloured separately, so hard bin 3 and soft bin 3 are different colours. A colour scheme has a limited number of colours, so bins far enough apart in number repeat one — when two of them are on screen together a Data warnings entry names them, and you can tell them apart with the legend, the tooltip, or by highlighting one bin at a time.

Overlays¶

Use the Overlays menu to toggle optional display layers on and off:

• Axis labels — the die coordinates along the bottom and left edges. They appear when you zoom; this row shows them all the time, or hides them. Clear overlays returns to showing them on zoom. Labels run across the dies and no further, and sit clear of the legend wherever it is placed
• XY axis indicator — arrows showing which way the X and Y die coordinates run
• Ring boundaries — concentric ring divisions that match the spatial analysis zones
• Quadrant lines — divides the wafer into NE, NW, SE and SW quadrants
• Die coordinate labels — draws the (x, y) grid position inside each die (useful at high zoom)
• Reticle grid — stepper field grid (only shown when reticle geometry is configured)
• Limit pass/fail — pass/fail colouring for Test Value mode, judged against the test limits, when the active test defines them
• Test pass/fail — pass/fail colouring for Test Value mode, coloured by the tester's recorded verdict; always on for a functional (no measured value) active test

Ring boundaries, quadrant lines, and XY indicator all active.

Compact layout¶

On a multi-project wafer, each reticle holds only a few of your dies, so the map is mostly empty and the dies you care about are small. When the occupied columns and rows repeat at a regular pitch, the Overlays menu offers Compact layout. Switch it on and the empty rows and columns disappear, so your dies are drawn as one grid at a much larger size, with each group of dies outlined so you can still see the reticles they came from.

The same synthetic wafer in the wafer view and in the compact layout.

• Nothing is left out. Only the arrangement changes. Every die is still drawn and counted, so the legend, yield and statistics are the same as in the wafer view. A die that has no bin (for example one that carries metadata only) keeps its place, shown in the no-data grey.
• Coordinates stay original. Hovering a die and the axis labels give its real die coordinates, not its position in the compact grid.
• Orientation. The notch marker shows the wafer orientation and follows rotation and flips. The XY indicator works as it does on the wafer view, drawn in a margin beside the grid.
• No wafer outline. The wafer circle, ring boundaries, quadrant lines and reticle grid describe the physical wafer, which the compact grid no longer follows, so those rows are greyed out while the layout is on.
• Galleries. All cards share one layout, built from every wafer shown, so wafers can be compared cell for cell. Each card is as tall as its map needs, so more of them fit on screen; a quarter turn swaps a card's width and height.
• Axis labels. With the Axis labels row on, the first column of each group of dies is labelled with its real die coordinate, which marks the reticle boundaries; zoom in far enough and every die is labelled.
• When it is offered. The layout is offered when the occupied columns and rows repeat. If you supply reticleConfig, it is offered when they repeat at the reticle width and height or a multiple of it (a product on every second reticle repeats at twice the width). Random missing dies do not repeat, so a wafer that merely has holes is not offered it. Developers can switch it on regardless with viewOptions: { compact: true }.

Layout diagnostics (also in the Overlays menu) shows what the detector saw as counts and scores only: it holds no die positions, bins, test values or wafer names. If a layout is not recognised the way you expect, copy the text or save it as a file and send it to whoever supports this software.

Keyboard shortcuts¶

Gallery¶

The gallery control bar applies to all cards simultaneously.

	Control	Description
	Plot mode	Switches the plot mode for all cards.
	Overlays	Toggles display layers for all cards simultaneously.
	Colour scheme	Picks bin colours (bin modes) or value colours (other modes) for all cards, including Reverse gradient.
	Orientation	Opens the rotate/flip controls, applied to all cards.
	Columns	Sets the number of columns, or press 1 to 5, or 0 for Auto, while the pointer is over the gallery or focus is in it; it offers no more columns than there are cards. Auto fits as many cards as keep each die at least 6 px across, so a dense wafer gets a wider card and a window that cannot hold another column at that width keeps the columns it has; a number divides the full width between that many columns.
	Select on every wafer	A toggle. Off (the default), each card selects on its own. A badge counts the selected die positions; clicking the button while something is selected clears it. On, a box, a click or a clear on any card applies to every card at the same die positions, so a region is read lot-wide; right-click then opens charts and tables of those dies across all the wafers (see Box select).
	Save image	Downloads the full gallery grid as a single PNG.
	Aggregation method	Stacked modes only. Selects how values from multiple wafers are combined per die position: Mean, Median, Std Dev, Min, Max, or Count.
	Log scale	Test Value and Stacked Test Values modes only. Applies a log₁₀ scale to the colour mapping for all cards.
	Legend style	Legend on each map adds a legend to every card as well as the shared one above the grid (off by default). While it is on, bin modes also let you choose where that legend sits on each card.
	Colour bar range	Test Value mode with test limits only. Toggles all cards between the limit range and the data range. Leave it on the test limits when comparing wafers — the data range rescales per view.
	Data warnings	Appears only when something is worth reporting about the wafers shown. Collected across every wafer and de-duplicated, so a problem affecting all of them is stated once rather than repeated per card.
	User guide	Opens this guide.

Filter… sits last in the row, as on a single map: one filter for the gallery, a single map and Insights (see Filtering wafers and dies).

Right-click a card anywhere outside its map to chart that wafer on its own — a histogram, process capability or a sweep (see Charting dies and wafers).

Click a card's expand button to detach it into its own separate window with the complete single-map toolbar (falls back to a floating window inside the page if separate windows aren't available in your environment). The vacated grid card becomes a placeholder whose own button reattaches it; closing the detached window does the same.

Filtering wafers and dies¶

Filter… keeps only the wafers and dies that meet conditions you set. It is one filter for everything you are looking at: the cards of a gallery, a single map and every Insights tab, so setting it in one view sets it in the others.

It opens a window where you add conditions. Each is on one field and all of them must hold:

• a category field (a split, a lot, a bin, a ring, a pass/fail verdict) keeps the values you tick;
• a number field (a test, a wafer's yield) keeps a range, either end left open, started at the data's own range so adding it changes nothing until you narrow it;
• is not / outside turns a condition round, to keep everything else.

The window says as you edit how many wafers and dies the conditions would keep (Keeps 2 of 3 wafers · 48 of 72 dies). Apply filters; Cancel leaves things as they were.

The Filter window: a condition on the process split, what it would keep (3 of 6 wafers), and below it the saved filters.

What each kind of condition does. A condition on a wafer field (split, lot, yield) keeps whole wafers: the others' cards leave the gallery. A condition on a die field (bin, ring, a test) keeps those dies inside the wafers kept: on every map the dies it drops are drawn dimmed, so the wafer's outline and where they were stay readable. A die with no value for a condition's field meets no condition, negated or not. A wafer condition is judged on the whole wafer, before die conditions narrow it, so yield below 90 % means the wafer's yield.

A wafer filter: only the Hi-dose wafers' cards remain, the line above them states the filter and what it kept, and the Summary panel describes those three wafers.

A die filter (bin 1 in the east quadrants): every wafer stays, and the dies it drops are drawn faded.

Everything is analysed again over what is kept. The legend, the yield, the Summary panel and its button, What stands out, the findings and warnings, each wafer's own summary, the report and the tables all describe the wafers and dies that remain, on a map and in the gallery alike, not the whole lot. Removing the filter, or switching it off, brings back the analysis you started with. In a stacked view, the wafers dropped are not stacked.

Filtering is never hidden. While a filter exists, a line above the view states it, with what it keeps (Filter on: Split is A — 2 of 3 wafers · 48 of 72 dies), Edit… and Remove. Its switch turns the filter off without losing it, and the line stays, saying the filter is off, until you remove it. A filter that names a field these wafers do not have (a test this lot lacks) filters nothing, and the line says so.

The filter is view state, like Group by: it is not part of any plot or plots file, and a host application can remember it for the next lot.

Filter as JSON¶

The Filter window has a Conditions | JSON switch. JSON shows the same filter as text, which you can edit, paste into or copy out of. The text is read as you type: the preview follows it, and Apply is held, with the reason under the text, while it cannot be read. Suggestions list the filter's keys, the lot's own wafer fields and tests, and the values a field takes (Suggest or Ctrl/Cmd+Space opens them by hand), and Ctrl/Cmd+Z steps back through the text. Pasting a saved filters file with several filters offers each to choose from; anything the reader leaves out is named under the text. Switching back to Conditions drops text that could not be read.

Saved filters¶

Conditions you use again can be saved. In the Filter window, type a name and press Save to keep the conditions you have set; saving under a name already used updates that filter. Load puts a saved filter's conditions in the window without applying them, so you see what they keep first, and Delete removes it. A saved filter that names a field this lot lacks says so beside its name, and can still be loaded.

Export… writes all your saved filters to a file, to share or keep a copy, and Import… adds the filters from a file written that way. An import never replaces: a filter whose id is already saved is added as a copy.

The file is JSON, so it can be written or edited by hand, and a host application can generate one:

{
  "format": "wafermap-filters",
  "version": 1,
  "filters": [
    {
      "id": "ne-fails",
      "name": "Bin 2 in the north-east quadrant",
      "conditions": [
        { "field": { "builtin": "hbin" }, "values": ["Bin 2"] },
        { "field": { "builtin": "quadrant" }, "values": ["NE"] }
      ]
    },
    {
      "id": "good-vth",
      "name": "Vth between 0.2 and 0.8 V",
      "conditions": [ { "field": { "test": 1001 }, "min": 0.2, "max": 0.8 } ]
    }
  ]
}

Each filter has an id (any text, unique in the file), a name and its conditions. A condition names its field and either values to keep (a category) or min / max (a number), with "not": true to keep the rest. A field is one of: { "builtin": "hbin" } (also sbin, ring, quadrant, yield and the other built-in die and wafer fields), { "test": 1001 } (a test, by number) or { "meta": "split" } (a wafer attribute, by the name it has in the data). A value is written as the Filter window lists it: a bin as "Bin 2", a ring as "Ring 1", a quadrant as "NE". The reader is forgiving: it keeps every filter it can read and tells you which condition it left out, takes a plain list of filters, takes a lone condition written without its list, and makes up a missing id or name.

4. Interacting with dies¶

4.1 Hover¶

Hovering over a die shows a compact tooltip. It always includes the die grid coordinates (x, y), and a retest count if the die was probed more than once. The rest depends on the active plot mode:

• Bin modes — the die's bin verdict (number and name, if named), plus a note of how many test values are recorded for the die
• Test Value mode — the active test's value in bold, flagged when it is outside the test limits, with the remaining tests summarised as "+N more tests"
• Stacked modes — the single aggregated value or count at that position

4.2 Zoom and pan¶

Scroll to zoom in and out. To pan, hold Space and drag, use the arrow keys, or choose Pan mode in the toolbar so that a plain drag pans. The toolbar also provides dedicated Zoom mode (drag to draw a zoom region), Zoom in, Zoom out, and Reset zoom buttons. Tooltips and die selection remain accurate at all zoom levels.

4.3 Box select¶

Click and drag on the map to draw a selection rectangle — Box select is the mode a map opens in. Click a die to select just that die, and click it again to clear it. The application may display statistics or details for the selected dies. This is useful for comparing a sub-region against the full wafer.

Selected dies keep their full colour and every other die is faded, with an outline round the selection, so it is clear which dies are selected whatever their shape — a block, a ring or scattered dies. A finding highlighted from the Summary panel is shown the same way.

Use Ctrl / Cmd + click to add or remove individual dies. Press Esc, or click outside the wafer, to clear the selection.

In a gallery, each card selects on its own unless you turn on Select on every wafer in the toolbar. Then a selection made on one card is made at the same die positions on every card — draw a box round the edge ring, a scratch zone or a reticle corner once, and see that region on all the wafers. Right-click (or use the Chart menu) and the charts and tables open on those dies across the whole lot: Dies, Test statistics and a Wafers table with one row per wafer, each stating "selected at the same die positions on N wafers".

Pick whole wafers instead with Ctrl / Cmd + click on a card's header: the card is outlined and the toolbar counts "N wafers picked ✕" (click it to clear). Right-click one of the picked cards and the charts and tables open on every die of every picked wafer, with a Wafers table. Dies selected on a map take precedence over a pick.

The button shows how many die positions are selected. Clear it by clicking an empty part of any map, pressing Esc, or clicking the button itself (it clears the selection and stays on); a second click turns the mode off. Turning it off always clears the selection on every card.

A block of dies near the centre is shown pre-selected. Drag on the map to make your own selection, or Ctrl/Cmd + click individual dies.

4.4 Charting dies and wafers¶

Right-click a population to open a menu of charts and tables drawn from just those dies. Picking one opens it in a window over the map. This is how you look at a cluster, a scratch or an edge region on its own, or at one wafer out of a lot.

What you right-click decides the population:

• Selected dies — select some dies, then right-click on the map. Right- clicking a die that is not selected selects that die alone first; right-clicking a selected die, or empty space, keeps the selection. On a single map the menu's first row, Invert selection, selects the dies that were not selected and deselects those that were. Below the charts of the selection, Selection vs the rest of the wafer offers a Value histogram (one colour for each side, overlaid) and a Box plot of the selected dies against the wafer's other dies, each side labelled with its die count. Partial and edge-excluded dies are left out of both, and with fewer than 10 dies on a side the window says to read the spread rather than small differences.
• Dies selected on several wafers of a gallery — by clicking a finding, or by selecting on more than one card. Right-click a selected die on any of those cards and the menu opens on every selected die, with a first row, Only this wafer, that narrows it to the dies selected on the card you clicked (and All selected to go back).
• A bin in the legend — right-click its entry, on a map's legend or the gallery's lot legend. The menu opens on that bin's dies (on this wafer, or on every wafer for the lot legend), and nothing is selected: the legend only filters what is drawn. With several bins filtered in, right-clicking one of them opens on all of them, with an Only bin N row to narrow it (and All filtered bins to go back). On the lot legend the Menu key or Shift + F10 works on a focused entry.
• Plots and sweeps you have saved — they open in their editor over just those dies, so you can change one there; the change is saved to the plot.
• A finding — right-click it in the Summary panel. If it is not already shown, its dies are selected first, then the menu opens on them: on all the wafers of a gallery that the finding selects dies on. A finding that is about whole wafers and selects no dies (a wafer's yield) opens the menu on those wafers.
• A whole wafer — right-click empty map space with nothing selected, a gallery card anywhere outside its map (its header, the space around the map), or one wafer's bar, box or point in an Insights chart (Yield by wafer, Test value distribution, Wafer-to-wafer trend — their tooltips say "right-click to chart this wafer").
• Dies dragged out of the Correlation scatter — across any number of wafers. The menu then also offers a Wafers table, one row per wafer the dies fall on.

On the map you can also press the Menu key or Shift + F10, or use the Chart toolbar button, which charts the selection when there is one and the wafer otherwise.

The menu offers:

• Value histogram — one test's distribution, opening on the test the map is showing. Edit as plot in its ⋯ menu closes it and opens the same test as a draft plot over the same dies, to change its axes, colour or limits and keep it if you want.
• Process capability — every test normalised to its spec limits, or its test limits when it has none, with its Ppk. Tests beyond the card's width scroll sideways, with the scrollbar under their names or a sideways wheel. Below 30 dies the chart says each Ppk is a rough estimate: a Ppk from a handful of dies can be far from the process's real capability.
• Plots you have saved, sweeps included (see Plots and Sweep cards), and New plot… and New sweep….
• Dies and Test statistics — the same tables as Insights' Data tab (see Data tables), over just these dies: every die as a row, or the Test Values and Functional Tests tables. Each has Export CSV and Copy, and a saved file carries the population ("selected on W03") in its name and in a Wafer column, so it cannot be mistaken for the whole lot.

What every chart opened this way does:

• It states its population — how many dies and which wafer they came from — in its title and above the plot, so it cannot be mistaken for a chart of the whole lot. Partial and edge-excluded dies are left out, as they are everywhere else, and the chart says how many that was.
• It is a snapshot: changing the selection afterwards does not change a chart that is already open.
• A chart that cannot be drawn from the population stays in the menu, greyed, with the reason — for instance when none of the dies has values for its tests, or when the map is a lot stack, whose dies are averages across wafers rather than measured dies.

A map with only bin data has no charts to draw, but its menu still opens with Dies: the selected dies, or the whole wafer, as a table.

5. Findings panel¶

When spatial analysis has been run, the Findings section of the Summary panel lists statistically detected patterns on the wafer. Open the panel with the toolbar's Summary button.

Each finding shows:

• Severity — Unusual, Notable, or Minor (ordered most to least significant)
• Description — plain-language summary of what was detected and where
• Click to highlight — clicking a finding fades the rest of the wafer and outlines the affected dies. A finding about a bin also filters the legend to that bin, so the legend always names what the finding describes; a finding about yield or a test clears the filter. Changing the selection or the legend filter yourself releases the finding.

Severity reflects how strong the statistical evidence is. Unusual findings have both a very low adjusted p-value and a large effect size — they are reliably important. Notable findings are statistically significant with a meaningful effect but not as extreme. Minor findings pass the significance threshold at lower strength and are worth reviewing but may reflect smaller or noisier patterns.

Test-value findings (a region reading high or low on a test) take the longest to compute, so they are included automatically only when the data is small enough to take about a second. On more data the panel says they are not included, with the number of tests and dies and roughly how long they would take, and an Analyse button that adds them. For a lot, one click analyses every wafer. Findings about tests failing their limits are included either way.

Findings panel open showing a detected edge-ring pattern. Click any row to highlight the affected dies on the map.

A failure cluster finding: the rest of the wafer is faded, so the affected dies stand out, with an outline round them.

Clusters are looked for in the failing dies together and in each failing hard bin on its own, so a spot of one bin next to other failures (a spot touching a failing edge, say) is still reported as that bin's cluster.

Finding types¶

Which finding types are active depends on the application's analysis configuration.

When a spatial pattern classification is detected alongside supporting regional findings (ring, quadrant, sector), the panel groups them together — the pattern classification is the primary finding and the regional findings provide supporting detail. A cluster of one bin does the same for that bin: the rise in the bin in the quadrant, sector or ring around it is listed under the cluster, which says where the failures are. A region row that is more severe than the cluster is listed on its own.

6. Summary panel¶

The Summary panel docks next to the map and gives you the wafer or lot at a glance, with findings sitting directly under the headline numbers so a clicked finding can highlight the affected dies right there. The Summary button in the row above the map, just before Map | Insights, opens and closes it, and shows as pressed while it is open. The ▸ button at the end of the panel's title closes it too, and stays in view as the panel scrolls. While a Filter… is on, the panel describes the wafers and dies the filter keeps, analysed again over them, not the whole lot (see Filtering wafers and dies).

The Summary panel open alongside a single wafer. Click any finding to highlight the affected dies.

Sections, top to bottom:

• Summary — die and wafer counts, yield, and the population the rest of the panel is computed over. For several wafers the heading names them: Lot LOT123 · 13 wafers when every wafer comes from that one lot, otherwise 26 wafers from 2 lots (or just 13 wafers when the data records no lot). The panel only says "lot" when it really is one lot. "Mean per-wafer yield" is an unweighted mean of each wafer's own yield; it is deliberately not the same statistic as the bin breakdown's pass-bin share, which weights every die equally. The two agree only when die counts are even across the wafers.
• What stands out — the headline (the yield, and for a lot its spread across wafers), then the items ranked by how much of the lot each affects: the largest in a tinted box with its full sentence, the others as a compact list (impact, name, and its size: a share of the dies, a shift in standard deviations, or a Ppk). An item is a loss or a signal. A loss costs dies: a region's shortfall with the failing bins that make it up, a wafer well below the lot, or a test outside its limits (or a functional test failing). A signal costs no dies of its own but moves the values, which is often all a characterisation lot has to show: a test whose mean shifts in a region by at least half a standard deviation, a trend across the wafers in the order given (said as "input order": the order is only a physical one if you know it is), a test whose Ppk against its limits is under 1.0, a test outside its limits on dies binned as passing, or, in a lot, a test that differs between the values of a wafer field Group by offers (a process split, a lot, a test program). A group is judged with each wafer as one value, the way Insights compares groups, adjusted over every field and test compared, and its size is the shift in the test's within-wafer standard deviations, the scale a region's shift is measured on. A field is one item however many tests it moves ("34 tests differ by Split. Largest: …; next: …"), with the tests that moved most as its chips, so a corner lot does not fill the list with one split. The same comparison is made on the wafers' yield and on the pass rate of the lot's weakest ring ("Yield is 4.7 points lower on probe card PC-B than on PC-A"); each set of groups is read after allowing for the strongest other one, and says so, and fields that group the wafers identically are one item that names them together. Groups that differ on two or more measures with no overlap but that the sample is too small to confirm are listed apart, under Possible, not confirmed. Its chart is a grid of the field's values by those tests, in z-scores (and a card that shows a bar per wafer draws the wafers under their group, in the group's colour, when Group by is on): blue where a group sits below the lot's average on a test, red where it sits above, so a process corner reads as a pattern of cells. Both are ranked on one scale, the share of the dies an item affects: a loss by the dies it cost; a shift by its dies, counted in full from two standard deviations and by the square of its size below that (a one-sigma shift counts a quarter of its dies); a trend by how far it moved; a group's shift by the dies of the wafers that moved, weighted as a shift is; a Ppk under 1.0 by the share of dies its spread would put outside its limits. Anything affecting less than 1% of the dies is left to the findings list. A line says what was compared. Its three dots and word (High, Moderate or Low impact) say how much of the lot an item affects: the higher of its share of all the dies (2% moderate, 4% high) and, for a loss, its share of the dies that fail (15% moderate, 40% high), so a small area that is most of a good lot's loss still reads as high. A test is costed by the dies it fails that also fail yield: dies outside its limits but binned as passing cost nothing. A test whose failures lie mostly in a region, or whose values shift on a region's dies, is said as part of that region's item rather than listed again beside it, and one test is one item. Every item is a row, and every row shows its dies on the map — a wafer icon at its end marks it. Click a region or wafer to highlight it, or a test to show the test's values with the dies it fails selected; click again to clear. Under the top item's sentence, chips name its parts (its failing bins, with their legend colours, the pattern, and tests failing or shifted on the same dies), and each chip shows that part alone. With Insights on, each of the top items also has a Chart link, which opens Insights at its card under Worth a look on the Overview, or, for an item about the yield of groups of wafers, at What divides yield. A wafer or lot with nothing affecting a yield point's worth of dies says so: "Nothing stands out". The reports open with the same section, and Full report at the foot of it opens the report; in a report opened from the panel, clicking a finding's row shows it on the map.
• Findings — detected anomalies, most severe first, each group marked with three dots filled by severity (Unusual, Notable, Minor). The header says how many are unusual or notable, even when the section is collapsed, and the severity chips narrow the list. Click a finding to show it on the map. A spatial pattern has a Show N supporting findings line under it that lists the regional findings it explains.
• Bin breakdown — bars as a share of dies, pass bins first and then failing bins by descending count, so the dominant failure mode is at the top. It follows the map's plot mode: a soft-bin map gets a soft-bin breakdown. When a wafer carries both bin types, a Hard / Soft selector in the section header overrides that.
• Region yield — ring yield by default, with a Ring / Quadrant selector (and Reticle cell, the yield of each position within the stepper field, when a reticle is set). Each row prints its difference from the wafer's yield in points and is tinted when it is 1, 2 or 4 or more points below it (nothing above is tinted). Ring is the default because edge roll-off is the pattern that dominates real wafer maps; a genuinely asymmetric quadrant is reported as a finding above, with a significance test behind it.
• Wafer yield (several wafers only) — one bar per wafer with the median marked, each row's difference from the median in points, and a tint when it is 1, 2 or 4 or more points below it. Wafers are in slot order by default, since that is what makes a slot-correlated pattern visible; a Slot / Yield selector re-sorts. Wafers well below or above the rest are labelled "low outlier" or "high outlier" (3 or more wafers, and at least 3 points from the median — the same rule as the outlier-wafer findings).
• Test values — per test: mean, Ppk, and limit yield. Ppk (not Cpk) because it measures against the overall spread, including wafer-to-wafer variation, which is what the dies actually ship against. The full descriptive statistics — min, quartiles, median, max, σ, both test limits, and all four capability indices — are in the CSV export, and the wafer summary report shows N, min, quartiles, median, mean, max, σ and limit yield (a lot's report has N, min, mean, max, σ and limit yield: quartiles cannot be combined from each wafer's own). Both reports also carry the full Cp, Cpk, Pp and Ppk table.
• Functional tests — pass/fail counts and pass rate per functional test.

Every section header can be collapsed, and stays collapsed as the panel re-renders.

A Summary report button (when present) opens a printable full-detail report — yield, bin breakdown, ring and quadrant statistics, the full per-test table with Cp/Cpk/Pp/Ppk, and the findings list — and can be saved as a PDF from your browser's print dialog. It opens with What stands out: the yield, up to three items ranked by how much of the lot each affects (losses and signals alike, as the panel ranks them), and a line saying what was compared. An item is listed only when it affects at least one per cent of the dies analysed; a lot with none says so. Each item links to its evidence: a finding to its row in the Findings table below, a test to the Test Values or Functional Tests section. Severity is drawn as three dots and a word, so it reads in a black-and-white print; yields carry a bar, and a wafer or region is tinted only when it is at least a point below the rest.

A Data tables button beside it opens the same tables as Insights' Data tab (Data tables) over this wafer, or over the whole lot in a gallery: Statistics, every die, and, for a lot, one row per wafer — each with Export CSV and Copy. It does not need Insights to be switched on.

Why some findings name two bins¶

A finding may read "hard bin and soft bin 3 (Fail) (same dies)". That is one group of dies counted in two bin spaces, not two separate groups added together. (When the two bins carry different names, both are spelled out in full.) Hard and soft bins are independent numbering systems, so this wording appears only when the two happen to cover exactly the same dies — reporting it twice would look like two independent problems.

For the same reason, when a single pass bin is configured you will see the yield finding for a region but not a separate finding for the pass bin itself: "pass-bin occurrence is 22 points lower" and "yield is 22 points lower" are the same sentence.

Nothing is discarded — an application reading the findings programmatically still receives every one of them.

For lot-level views (gallery), the panel shows lot-level findings, and the Wafer Yield section badges each wafer with its own findings count — click a row to open that wafer on its bins, as a yield opens everywhere. There is no separate wafers tab: every wafer appears in that one list, including the ones with no findings, which is what lets you see a low-yielding wafer that nothing flagged.

7. Lot-stack maps¶

A lot-stack map combines multiple individual wafer results into a single composite view. The display clearly identifies this: the number of wafers included and the aggregation method in use are shown in the panel header (for example, "3 wafers · mean") so you know you are not viewing a single wafer's data.

Individual die coordinates are preserved. For each die grid position, results from all wafers are aggregated into a single value or bin count according to the selected aggregation method.

Stacked plot modes¶

Switching to a stacked mode changes what each die's colour represents:

Aggregation method¶

In Stacked Test Values mode, use the Aggregation method toolbar button (Σ) to choose how values from each wafer are combined at each die position:

Gallery stacked modes¶

In a gallery, stacked modes are always available in the control bar. Switching to a stacked mode shows one card per bin category or per test parameter, aggregated across all wafers in the gallery.

Stacked Hard Bins mode: each die position is coloured by how many wafers had bin 1 (Pass) or bin 2 (Fail) at that location.

8. Insights tab¶

The Insights side of the Maps | Insights switch (at the far right of the toolbar row, single map or gallery) swaps the wafer view for a chart suite computed from the same die data — without leaving the toolbar. Choose Maps to return to the map; the switch stays where it is. While Insights is open the buttons that act on the map, the Summary button among them, are hidden, since there is no map for the Summary panel to highlight; Maps brings them back.

The Chart reference describes every chart, plot and table: what it shows, how to read it, when to reach for it and where to find it. This section covers how the tab works.

The Columns control beside Group by and Show, or the keys 1 to 5 (and 0 for Auto), sets how many columns of cards every section of the tab uses. Auto fits as many as the width holds at 420 px or more; a number is the most columns a section uses, dividing the width that many ways, and the choice stays while you move between the tabs. A tab uses the smaller of that number and the cards in its fullest section, so a tab of two cards is two columns even with 4 chosen.

Any chart mark that stands for one wafer — a bar in Yield by wafer, a box in Test value distribution, a point in Wafer-to-wafer trend — can be right-clicked to chart that wafer on its own (see Charting dies and wafers); its tooltip says so.

Insights is organised into three chart sub-tabs, a Data sub-tab holding the same numbers as tables (see Data tables below), and a My plots sub-tab for your own charts and sweeps (see Plots and Sweep cards below):

• Overview — how the lot yielded: headline tiles stating the population, Yield by wafer, Bin pareto, Pass rate by test, Ring yield and Quadrant yield, and with a reticle set Reticle cell yield. With four or more wafers and wafer fields to compare, What divides yield ranks the fields by how clearly their groups differ; Group by on its row groups the whole tab by that field. With Group by on, the yield bar and the pareto draw a bar per group (click a group's bar to see its wafers, Back to return). The pass-rate card offers up to three ways of judging, whichever the data supports: Test limits, Tester flag and Functional; when the first two disagree it says how many dies they judged differently. The card is the size of the others whatever the number of tests: more than fit scroll. With nothing to judge it is left out.
• Worth a look, at the top of the Overview — the same items as What stands out beside the map, in the same order and words, each drawn as the chart that shows it: a region's loss as yield by that kind of region (each bar with its value, and the median of the bars as a line, as on the yield card), a fail bin as the bins' shares by it, a shifted test as a box of the test by it, a failing or too-wide test as its histogram (a pass/fail test as its pass rate), a trend as a control chart, an outlier wafer as yield by wafer. An item about the yield of groups of wafers (a probe card, a recipe) is a row of What divides yield just below, which reads the same fields the same way, so it has no card here. Each card says why it is shown in one sentence (hover it for the whole sentence What stands out gives, with the bins and tests it names), and Show on map goes back to the map showing it as its row does, with the Summary panel open on that row. Smaller items are listed under What stands out only. Under a filter it shows the items What stands out lists for what the filter keeps; while Show narrows to one group it says so instead of drawing cards.
• Distributions — how values spread: Process capability, the test-value box plot (one box per wafer, or with Group by on one per group: click a group's box to see its wafers, Back to return), the value histogram (all wafers, one wafer, or one series per group) and the wafer-to-wafer trend. The trend is always in slot order, deliberately: drift across a cassette only reads in the physical sequence, so there is no sort control to destroy it, and it needs at least two wafers; drag across it to pick a run of wafers. The box plot, histogram and trend are plots: each has the test picker and the axis settings (Axis includes limits, Clip outliers, and which limits to draw), and a change in one reaches the others. Clicking a capability box drives them onto that same test. In a lot with hundreds of tests, click the ☆ at the end of a row in any test picker (or press P on the focused row) to pin that test: pinned tests are listed first, under a "Pinned" heading, in every picker.

With Group by on (a split, a lot), the cards that compare the groups come first, under Comparing …: the Comparison by table and the difference map (pick the two groups and a test or Yield; it follows the test the other panels show until you pick one), then the cards for each test under Test by test. In the table, choose a Control and each other group's cell adds its difference from it; click a test to show it in the panels below; Comparison CSV saves the table. Narrowing Show to one group leaves nothing to compare, so the table is not shown. Comparing groups of wafers explains each figure.

The box plot, histogram, trend and the Correlation tab's scatter draw the test's limits as dashed lines. A test can have two kinds: test limits (labelled Lo limit / Hi limit, short dashes) — what each die was judged pass/fail by — and spec limits (labelled LSL / USL, long dashes) — the process specification that capability is measured against. When a test has both, a Limits choice appears — Test + spec (the default), Test limits, Spec limits or None — and it applies to all four charts at once. A test that lacks the kind you chose shows the kind it has, so it never looks as if it had no limits; a limit that falls outside the plotted range is marked at the edge of the chart with an arrow.

The box plot, histogram and trend share one row of axis controls, kept in sync across all three so switching between them never re-reads the same data on a different scale: Axis includes limits widens the axis to cover every limit shown even where the data sits well clear of them (on by default only when doing so still leaves the data at least a third of the axis — otherwise a generous limit window would squash a perfectly capable distribution into a sliver), and Clip outliers narrows the axis to a robust range (the Tukey fence, 1.5× IQR beyond Q1/Q3) so a handful of extreme values don't stretch the axis until the rest of the distribution reads as a flat line. Clipping affects the AXIS only — every reported statistic (mean, σ, Cpk, the box's own five-number summary) is computed from every value; nothing is excluded from the numbers, and the panel states how many points sit outside the visible range when some do. The box plot additionally has its own Log scale toggle, independent of the shared pair above (available once every plotted value is positive) — the histogram and trend charts do not offer one. - Correlation — the test correlation matrix and a die-level scatter, both stating the sample size the coefficients are computed over. Max tests (5 to 100, 20 by default) sets how many tests the matrix shows, and Correlation CSV saves it as a table. Clicking a matrix cell drives the scatter onto that pair. Hover a point to see which wafer and die it is; click it to open that wafer on the X test. Drag a rectangle over the plot to select the dies inside it — across as many wafers as they fall on — and a menu opens with charts and tables of just those dies (see Charting dies and wafers). The selection stays ringed until you change the X or Y test or click empty space.

In a gallery with more than one wafer, a Group by control appears whenever a wafer attribute (lot, product, test program, temperature, split, or a custom field) actually varies across the loaded wafers — grouping pools or restricts each panel differently depending on what makes sense for that chart type. Group by separates whole wafers; to separate the dies within each wafer, a plot's own Compare by choices (see Plots) take die regions, bins, sites and pass/fail tests as well. The Filter… button, at the right of the toolbar row, narrows every Insights tab at once, the maps and the gallery with them (see Filtering wafers and dies). The line that states the filter sits above the tabs, with what it keeps (2 of 3 wafers · 48 of 72 dies), Edit… and Remove. Group by and Show then work on what the filter kept.

Histogram, correlation, and scatter also offer a Wafer picker to narrow from "all wafers pooled" down to one wafer at a time. Clicking a yield bar or box-plot row for one wafer opens that wafer's own map — a box-plot click opens directly on the test you were looking at.

Insights tab open on a single wafer, Overview sub-tab: yield and bin pareto and ring/quadrant yield — all computed from this wafer's own dies. The Distributions sub-tab has process capability, a test-value box plot and histogram; Correlation has a correlation matrix with scatter plot. In a gallery, Overview also gains a "Group by" control and per-wafer yield/box-plot rows.

Gallery Insights, Overview sub-tab: lot-wide yield by wafer (click a bar to open that wafer's map), hard/soft bin pareto, and ring and quadrant yield. The Maps | Insights switch at the top left (it does not move between the two views) returns to the card grid.

Distributions sub-tab: process capability (coloured by Ppk — green capable, orange marginal, red poor; tests without limits are muted and dashed), a per-wafer test-value box plot, and a value histogram. Clicking a capability box drives the box plot and histogram onto that test.

Correlation sub-tab: test-to-test Pearson correlation matrix (orange = positive, blue = negative; intensity = strength) and a die-level scatter plot coloured by hard bin. Clicking a matrix cell drives the scatter plot onto that pair.

What a click does¶

Every mark on an Insights chart stands for something, and a click opens it. A right-click on any mark opens the menu on what it counts, so on a mark that stands for dies a click and a right-click do the same; on a wafer's own mark, a click opens the wafer's map and a right-click its menu. The tooltip says which. A wafer opens on what was clicked, wherever it was clicked from: a test (a chart of the test, a test's column in a table) opens it on that test, a yield or bin (a yield bar, the Yield % column) on its bins, and anything else in the mode the maps are in.

The menu offers the charts and tables for just those dies (and your own plots), and its heading names them, for example "412 dies in hard bin 3, across 13 wafers". A mark that has no dies behind it does nothing. Right-click on any wafer's bar, box or point opens the same menu for the whole wafer. The exceptions only move within Insights and have no menu: a test in the process-capability chart and a cell of the correlation matrix.

Data tables¶

Data sub-tab, Statistics view: the Test Values and Functional Tests tables for the lot, each with its own CSV button; Dies and Wafers are one click away.

The Data sub-tab shows the wafers Insights is scoped to as tables, one at a time, chosen with the Statistics | Wafers | Dies control. It follows the same "Group by" and "Show" scope as the charts.

• Statistics — the Test Values table (count, min, quartiles, mean, σ, Ppk, limits and the limit yield for each test, with its N stated) and the Functional Tests pass-rate table. With "Group by" active there is one set per group, below one table that compares every test across the groups (a test's groups side by side, with its own CSV).
• Dies — one row per die: wafer, X, Y, ring, quadrant (and reticle cell and shot when a reticle is set), site, bins, a column for each test, and the wafer's metadata. Only the rows in view are drawn, so a lot of hundreds of thousands of dies scrolls smoothly. Click a column heading to sort by it (again to reverse; a die with no value sorts last).
• Wafers — one row per wafer: the wafer, the field Group by divides it by, the die counts, the yield (the same figure as the yield chart), the mean of each test (the first 50), then the other attributes that differ between wafers. Attributes the same on every wafer (lot, part type, tester…) are named in a line above the table and kept in the export.

Shading. Where a table compares values, a pale colour sits behind the number and the number is always printed: a group's or wafer's mean is blue below the average of its peers and warm above (the same two hues as the correlation matrix), a yield (or a share inside limits) is tinted red only when it falls a point or more short of its reference, as in the summary report: a wafer short of the lot's mean per-wafer yield, a group short of the best group's share inside limits, and nothing at all on a lot with nothing wrong. A Ppk is capable, marginal or poor, and in the Dies table a reading outside its test's limits is tinted warm. The Comparison by table in Distributions shades each group's mean among the groups the same way.

In the Dies table, click a row to show that die on its map: the table steps aside (or Insights closes) and the die is ringed, on its own card in a gallery. The ring goes with the next click on the map, or Esc.

Export CSV saves the table as shown, in the order shown. Numbers are written in full, as plain numbers, not as the formatted text on screen (0.5123457, not 512 mV), with the unit in the column heading. Copy puts the table on the clipboard as tab-separated text, ready to paste into a spreadsheet; it is offered for tables up to about 200,000 cells. For the Dies table, Export format (Wide | Long) chooses the layout of the file: wide has a column per test, long has a row per die per test (a die with no result for a test has no row), the shape statistics tools prefer. The note under the buttons says how many rows the file will have, and warns when that is more than a spreadsheet can open (1,048,576).

Plots¶

The My plots sub-tab is where you build your own charts. + New plot starts one, already filled in with the first two tests of the lot, and opens it beside a large copy of the chart. Edit as plot, in the ⋯ menu of an Insights chart, opens that chart as a plot of your own (see A chart's buttons below). The My plots | Examples switch above the list shows one or the other, never both: Examples shows plots to start from, or to see what each chart and setting is for, each marked as an example; none is saved until you add it. They are chosen from the lot in view: the lead test is the one that varies most from wafer to wafer (for a single wafer, the one that follows distance from the centre most), its partner the test that tracks it best, and the split is the lot field that best explains the lead test. Each other lot field that divides the wafers' yield (a probe card, a tester) gets a bar of yield by its groups too; fields that group the wafers identically, such as a tester and the handler docked to it, each get one, so you can follow either. The examples are a scatter of the lead test against its partner with a fitted line, a histogram with its normal curve, a box of the lead test by wafer, a bar of yield by the split, a line over wafer order with a ±1 SD band, and a probability plot. The examples also show one of each newer feature the lot can show: the most telling tests as one set rescaled to z-scores (a box per test, and a heatmap of tests by wafer), the lead test by split and by wafer nested in split, panels by ring, the lead test by ranges of its partner and of the distance from the centre, a pareto of the hard bins and each wafer's bin mix, the lead test by PAT verdict, an individuals control chart (five or more wafers), a capability goal plot and a pass-rate bar when tests have limits, and a sweep of the first tests in test order. Each example has Add to my plots, and Add all adds the ones you do not have yet and shows My plots, with an Undo; edit a plot to see how it is set. Every change you make there is drawn at once; there is no Apply button, and the plot is kept as you go.

The My plots tab after the examples are added: each card is a plot, with Edit, Duplicate and Delete.

Edit opens the plot large, beside its settings. Every change is drawn at once.

In Setup, choose the chart type, then the field for each role:

• X axis and Y axis (a histogram has one Values field; a box or bar has Categories instead of a numeric X). Any measured test, a die's X or Y position, a wafer's yield or die count, or a lot field with numbers in it (slot, temperature). The list is grouped Tests · Die · Wafer and filters by name or test number, so with a few hundred tests you type 1050 or vth rather than scrolling. A field the chart type cannot use is not offered; switching type keeps your fields where they still apply and brings back the ones that did not when you switch back.
• Compare by. Two controls pick what a chart compares: Colour and, on a bar or box, Categories. Either takes a wafer attribute (lot, split, temperature: it separates whole wafers) or a die region or field (ring, quadrant, reticle cell, reticle shot, bin, site: it separates the dies within each wafer). Group by, above the charts, is the other way to separate wafers: it pools whole wafers by a wafer attribute for every chart on the tab, and a plot's Colour can follow it. A die region is never a Group by choice, because a wafer is in one lot but a die is in a ring and a reticle cell at once. A pass/fail test (a functional test, or a derived one that gives a verdict) is a category too, listed under Verdicts as Scan Chain (pass/fail): it separates the dies that passed from those that failed, for Colour and Categories and for splitting yield. It has no value, so it is never an axis. A die with no recorded outcome is in neither group.
• Colour: a category such as wafer, split, hard bin, site, ring or, with a reticle set, reticle cell or reticle shot, or, on a scatter, a measured value or wafer figure drawn on a gradient (a colour bar under the title states its range). Follow Group by (the default) takes the colours from the Group by control above, so changing it recolours every plot that follows it; None draws one colour.
• Several tests (box, bar, line and histogram): choose Several tests… at the top of the values list, then type the tests in the Tests box beneath it, as numbers or ranges (1050, 1060..1080, a range taking the tests declared between its ends). A box or bar puts one box per test along X; with wafer, split or ring as the categories instead, the colour tells the tests apart. This is how the Process capability chart's view, every test side by side, is made for your own choice of tests, by split or by wafer. Tests in different units cannot share an axis as measured, so the plot asks you to rescale them. On a bar or line, Read as → Pass rates draws each test's pass rate instead of its values: the Overview's pass-rate card is this plot, horizontal and smallest first.
• Judged by (a pass rate, one test's or several): Test limits judges a parametric test's value against its limits, Tester flag reads the pass or fail the tester recorded. A functional test is judged by its verdict either way.
• Probability plots each group's values in order against the share of the group at or below them, at (rank − ½) / n, so 50% is the median. On the default normal-probability axis a normal distribution is a straight line and the tails are spread out: a shifted split is a curve moved sideways, a wider one a flatter curve, and outliers or a second population show as a bend or a kink at the ends. Choose Linear under Customise › Share axis for a plain 0 to 100% axis. Limits are drawn on the values axis. Hover a point for its die, its rank in its group and where that rank is plotted; click it to open its wafer; drag a box to take every die inside it to the drilldown menu (a curve of many dies draws a sample of its middle, but all of both tails, and a drag still takes every die).
• Heatmap: a grid with one row per value of one field and one column per value of another, each cell coloured by the value combined over its dies (with no values, a count of them): hard bins by wafer, rings by split, and with several tests as the values Test as the rows, so tests by wafer, rescaled to compare. The colour bar above it spans the cells; a cell shows its value when it has room, and a click charts or tabulates its dies. It is a bar chart of the two fields nested, laid out as a grid, so the same rules apply: pooled yield, pass rates, Filter.
• Control charts each wafer's value (a test's wafer mean, or as Combine values by says; a yield or a pass rate as the wafer's own) in wafer order, or along another wafer field with a number per wafer such as slot or date. It is an individuals chart: the centre is the mean, the control limits ±3σ with σ taken from the average moving range (÷ 1.128), which measures wafer-to-wafer noise so a slow drift does not widen the limits it is judged against. Faint lines mark ±1σ and ±2σ. A point that breaks a Western Electric rule is drawn in red with the rule's number: 1 beyond 3σ, 2 two of three beyond 2σ on one side, 3 four of five beyond 1σ on one side, 4 eight in a row on one side of the centre; the line under the chart names the rules broken. Limits from the first sets the centre and σ from the first wafers only (a baseline), so later wafers are judged against them; a dotted line marks where it ends. Hover a point for its wafer and how many σ it is from the centre; click it to open the wafer.
• Goal (capability goal plot) places each test that has both limits by its mean's offset from the centre of its limit window (left and right) and its standard deviation (up), both as shares of the window, using spec limits where a test has both kinds. Every point on a triangle has the same Ppk: the dashed one is Ppk 1, the solid one 1.33, so a test outside the solid triangle is short of 1.33, and whether it sits to the side (off centre) or high (too wide) says what would fix it. Tests short of 1.33 are named. Colour by a split to see each test move between splits; Tests narrows it to a set.
• Fit line (scatter): one line per colour, Least squares (Y on X, the usual regression) or Orthogonal (the line closest to the points at right angles, for two measurements of the same quantity, such as one test on two testers, where least squares would tilt the line towards flat). Each line's equation, R² and number of points are written above the chart. y = x line draws the line of agreement, when X and Y are in the same unit.
• Bars as (histogram): Counts, or Shares: each colour's bars as a percentage of its own values, so a split of three wafers and one of thirty can be compared by shape. Normal curve overlays each colour's normal distribution at its mean and standard deviation, scaled to its bars (not on a log axis).
• What-if limits (histogram or probability plot of one test, not rescaled): a proposed low and high limit in the test's own unit, drawn as their own lines (What-if lo / What-if hi). The line under the plot gives the test's pass rate over the dies it judges by its current test limits and with the proposed ones, and how many dies change. It is the test's pass rate, not the yield: a die's bin depends on every test.
• Pass rates, listed under Pass rates in the values list: a test's pass rate, each die 100 when it passes and 0 when it fails, judged as the Overview's test pass-rate card judges (a parametric test by its test limits, a functional one by its recorded outcome). A bar or line of it is the share of judged dies that pass, pooled over the dies in each bar; with one mark per wafer it is each wafer's own rate, which a box or scatter can then show. A single die's 0 or 100 drawn as a point says nothing, so a box, scatter or histogram of it per die is refused.
• In ranges: on the categories of a box or bar, the colour, a heatmap's rows and the panels, a measured test, Radial position, Die X or Y and Yield are listed as "… in ranges". The values are cut into 2 to 20 ranges (5 to begin with), either holding about the same number of values (Equal counts: quartiles, deciles) or spanning the same width (Equal widths), and each range is a category named by its limits in the field's unit (7.16 pA to 19.4 pA), lowest first. Vth by Idsat decile is a box with Idsat in ranges on the X axis. The edges are set over every die plotted, so panels and colours share them.
• Filter: keep only the dies that meet every condition you add. + Add a condition asks for a field; a category (bin, ring, site, split, a pass/fail test) then lists its values to tick, and a number (a measured test, Radial position, Die X or Y, Yield) a range, either end left open, that starts at the range the data spans. is not or outside keeps everything else instead. A condition on a wafer field (split, lot, yield) keeps whole wafers and is judged on the whole wafer, so Yield at most 70% means the wafer's own yield; the die conditions then keep dies inside those wafers. A die with no value for a condition's field (no bin, no reading) meets no condition. The filter applies before everything else, panels included. Its line on the card states each condition, and its switch turns the filter off and back on without losing it; the line under the plot says what was kept while it is on. PAT limits stay the whole wafer's.
• Band (line): round a line of means or medians, ±1 SD (the standard deviation of the values combined at each X), 95% CI (the mean's confidence interval, with Student's t for the number of dies or wafers there), Quartiles or Range, as a pale band in each line's colour. A pass rate has no spread of values to show, so it takes no band.
• Radial position, a die field: a die's distance from the wafer centre as a percentage of the radius, 0 at the centre and 100 at the edge. The rings are bands of it (with four rings, ring 4 is 75–100%), so a scatter of a test against radial position shows an edge roll-off the ring chart only buckets.
• Order and Show the first (box and bar, one category field): Largest first sorts the categories by value (a box by its median), Smallest first the other way (the worst pass rate at the top); with colours, a category is ordered by all its colours together. Show the first keeps that many and counts the rest in a note.
• Colours (bar): Share of colour keeps the colours side by side, each as its share of its own total, so groups of different sizes compare; Stacked puts a category's colours one on another, 100% shows each as its share of the category. Shares and stacking add values, so they need a count or a sum; a mean or a yield is refused with the reason.
• Cumulative line (a bar sorted largest first, of counts or sums): each category's running share of the whole, on its own 0–100% axis at the right. A bar of hard bins with it is a bin pareto; with Show the first, the line still counts every category.
• Axis includes limits and Clip outliers (Customise, per value axis): the first widens the axis to every limit drawn on it (or, unticked, never), where by default it does so only when the data keeps a third of the axis; the second narrows the axis to a robust range (the quartiles ± 1.5 × their spread) so a few extreme values do not flatten the rest. Clipping changes only the axis: the footnote counts the values outside, and every statistic still uses them.
• Orientation (box, bar): Horizontal puts the categories down the left and the values along the bottom, for many categories or long names (tests, wafers); the card grows so each category keeps room.
• Values on the bars (bar): each bar's value printed at its end. Shown only when every value fits its bar's width, so a crowded chart never labels some bars and not others.
• PAT (part average testing), a die field: each die's verdict under dynamic PAT, per wafer. For every test with both a low and a high limit, the wafer's PAT limits are the median ± 6 robust standard deviations ((Q3 − Q1) / 1.35) of its passing dies' values, never wider than the test's own limits (spec limits where it has both). A die is Within PAT limits, a PAT outlier (it passes every limit but is outside PAT limits on at least one test: the dies PAT would reject), or Fails (bin or limits) (already rejected, so not counted as a PAT outlier). A test with fewer than 30 passing values on a wafer, or no spread, does not judge that wafer's dies. Use it as the colour, the categories (a bar counts outliers per wafer) or the panels. A plot of one wafer, a panel per wafer included, also draws that wafer's PAT limits as dotted lines labelled PAT lo / PAT hi; a plot of several wafers has several sets and draws none. The limits are always the whole wafer's, even in a panel of one ring. Part average testing (PAT) explains how the limits are set and what this does and does not cover. The Overview shows the same verdicts as a Part average testing (PAT) card: for each wafer, the dies within the PAT limits, the outliers and the rate, the dies that fail anyway, and which tests the outliers are outside on. The Dies table (Data tab) has a PAT column, and PAT tests where there are outliers, in the CSV too; a selection that is only part of a wafer has neither, since the limits belong to the whole wafer. Map the verdicts on the card goes back to the maps coloured by PAT verdict, which is also the PAT verdict entry of the Plot mode menu on a map and on a gallery (offered when a test has both limits and a wafer has enough dies to set PAT limits from): within the limits in blue, outlier in amber, fails anyway in red, so the legend and value toggles show where the outliers are on each wafer. The dies keep their bins and measurements, so the tooltip shows them beside the verdict. With a filter on, the verdicts are judged over the dies the filter keeps, as on the card.
• Panels: one small plot per value of a category field, laid out like the gallery's cards, on shared axes and colours with one legend and one statement above them. A wafer field (split, lot) gives each panel its whole wafers; a die field (ring, site, bin) the dies with that value from every wafer; Test, with several tests as the values, one test a panel. So "one scatter per split" or "a histogram per ring" is one plot. Rescaled values are measured over the whole population, so a panel is not re-centred on itself and panels compare. Each panel has its own expand button, and the plot's Save as PNG saves the whole grid. At most 24 panels; a field on the X axis or the colour cannot also make the panels. The Columns control above the panels, or the keys 1 to 5 (and 0 for Auto) while the pointer is over the plot, sets how many panels sit side by side; Auto fits as many as the width holds at 260 px or more.
• Wafer-level comparison (box and bar of one measured test): when the categories or the colour are a wafer field such as split or lot, the line under the plot adds the comparison of those groups with each wafer as one value, as the Distributions table does: the wafers in each group, Welch p and rank p.
• Nested categories (box and bar): + Add a level under Categories divides each category by another field, so Test then Ring gives one box per ring within each test, and the colour is still free for split. The first level is the outer groups, labelled in a row beneath the axis with a line between groups; the last level's values are at the ticks. Move a level out or in with its ↑ and ↓ buttons, or remove it with ×. Up to three levels.
• Rescale values (box, bar, line and histogram of measured tests), each test on its own: To limits puts the low limit at 0 and the high one at 1 (spec limits where a test has both, its test limits otherwise, as in Process capability), so a value near 0 or 1 is near a limit whatever the test measures; z-score counts standard deviations from the mean of the values in view; Robust z does the same from the median, with the spread taken from the median absolute deviation, so a few outliers do not set the scale. A test that cannot be rescaled (no limits, or every value the same) is left out and named under the plot. The footnote says how the values were rescaled.
• Combine values by (bar and line, and any plot with one mark per wafer): mean, median, minimum, maximum, sum or count. Pooled yield is offered for Yield: it is the passing dies over the judged dies of all the wafers in a bar, not an average of their percentages (the plain mean of Yield is the mean of per-wafer yields, and is labelled so).
• One mark per: Automatic uses the finest level the fields allow. Wafer makes every mark a wafer, combining its dies' values as chosen above, which is how "mean Idsat per wafer against slot" is drawn. Yield can be split by a per-die field (ring, quadrant, bin, die position) in a bar (pooled yield per ring) or a line, which take the pass verdict of each die; a box, scatter or histogram of yield cannot, and says why.

When fields do not go together, the list says so before you choose: a field that would leave a plot that cannot be drawn is dimmed in the X, Y and Colour lists with the reason after it, and does nothing when picked. A plot that cannot be drawn says in words what is wrong and what to use instead.

The JSON tab shows the plot as text, in the format of a saved plots file, for when a plot is easier to copy, share or paste in than to rebuild with the controls. Edit the text and choose Apply; the plot changes only then, and the tab says beforehand if a setting would be left out or the text is not a plot. Suggestions appear as you type inside a key or a string and after a comma, brace, colon or bracket, or on demand from the Suggest button or Ctrl/Cmd+Space: the plot's settings, the words a setting can take, your lot's test numbers and wafer fields. Use Up and Down, then Tab or Enter to accept, or Esc to close the list. A list that opened by itself leaves Enter as a new line until you type a letter or use the arrow keys; Tab always accepts. Discard edits drops text not yet applied; Ctrl/Cmd+Z (and Shift+Ctrl/Cmd+Z) step back and forward through the text you have typed, accepted or discarded and then through what you have applied, and the window's Reset and Cancel go back to the plot as it was when the window opened. Copy puts the text on the clipboard. The id is not part of the text, and a plots file can be pasted too, and when it holds several plots the tab lists them to choose one.

A heatmap, or a scatter coloured by a number, also has Colours in Customise: blue to orange around zero (the automatic choice for values rescaled to a z-score), orange-red to blue for a yield or pass rate (bad to good), or any of the map's gradients (Viridis, Cividis, Plasma, Traffic and the rest), with Reverse colours to flip the direction. Blue to orange, orange-red to blue and Cividis stay distinguishable for the common kinds of colour blindness; Traffic and Jet do not, so they are offered by choice and never the automatic one. A gradient is stretched over the plot's own lowest and highest values, so it says "relative to this plot", not "above a target".

In Customise, the title and each axis title are written for you until you type your own (the box shows what it would say, and clearing it brings that back). If you type a title and later change the fields, the plot says when the title still names a test it no longer shows, or names only one of the two it does, with a button to go back to the automatic title; a title that names none of the fields ("Process check") is yours and is left alone. Each axis can have a minimum, a maximum, a log scale (offered whatever the values, but it stays linear, and says so, if any value is zero or negative) and reverse direction. A histogram also has a number of bins. A category axis has a title only.

A value axis can also carry reference lines and bands of your own: a target, a guard band, a customer's limit. Write them in Reference lines in Customise as a value for a line or lo..hi for a shaded band, each with an optional name, separated by commas (0.45 Target, 0.3..0.5 Window), in the axis's own unit (rescaled values for a rescaled plot). median or mean draws a line at that statistic of what the plot shows along the axis (the bars' heights, the points, the values in the boxes), worked out over the whole plot so every panel shares it: median, 0.45 Target. They are drawn dash-dot, labelled with their name and value, and widen the axis in the way test limits do.

When a plotted test has limits they are drawn as dashed lines on the axis that measures it (a histogram's values, a scatter's X or Y, a box, bar or line's values when they are measured values or their mean, median, minimum or maximum): short dashes for the test's own limits, long for its specification limits, dots for a single wafer's PAT limits, each labelled with its value. The axis widens to include them when that leaves the data at least a third of it, as in Insights, and a limit off the edge is marked there instead of vanishing. Choose which are drawn under Limits in Customise. Where the wafers in view disagree about a test's limits none are drawn, as in Insights: the reconciled test list drops conflicting limits.

Under every plot is a line stating what it shows: the wafers and dies, how values were combined ("median of Vth per wafer"), and how many dies were left out for having no value ("12 dies without Idsat not plotted"). Like every Insights chart, a plot reads only the dies a yield judges: partial dies and dies inside the edge exclusion are left out, and the line counts them ("12 edge-excluded dies left out"). A plot never drops a die without saying so. A plot coloured by hard or soft bin uses the map's bin colours.

Hover a point, bar, box or line to see its values and how many dies or wafers it stands for. Click a point (or a wafer's bar or box) to open that wafer, on the plot's own test, so its map shows the values the plot is about and not the bins (a plot of yield or bins with no test in it, like yield against wafer order, opens the bins, as a yield bar does everywhere). Click any other bar, box, histogram bar or line point to open the right-click menu on the dies it counts, and drag a rectangle over a scatter to select the dies (or wafers) inside it. The selection opens the same right-click menu as on a map, so a few dies from a plot can be charted and tabulated in turn.

Keeping your plots. A plot is a recipe, not a picture: it says which fields go where, and nothing about which wafers. It is therefore kept by your application between sessions and drawn again on the next lot. If you open a lot that lacks a test a plot needs, the plot stays in the list, dimmed, with the reason ("Needs test 3001 (Leak), which is not in these dies"); it is not removed. Export plots… writes them all to a file, and Import plots… adds the plots of such a file to yours, as copies where a plot with the same identity already exists, never replacing one. Each card has Edit, Duplicate and Delete; a deleted plot can be restored with Undo for ten seconds. Delete all plots… removes every plot, sweeps included, after asking you to confirm (the question says how many, and that Export plots… first keeps a copy); Undo restores the whole list for ten seconds. It is dimmed while there are no plots.

Three different ways to divide the data, kept apart. A plot divides data in three places, and they do different jobs, so neighbouring plots can ask different questions:

Plots in the right-click menu. Every saved plot is also a row in the menu that opens on a selection or a wafer (Plots section), drawn over just those dies and captioned with them.

New plot… and New sweep… in the same menu start a draft on the selection; a draft is kept only if you press Add to my plots.

Every change is kept as you make it. If you make a mistake, Reset (under the settings) puts the plot back as it was when you opened the editor, and Cancel does the same and closes it. Closing the window with its own button keeps what you changed.

Sweep cards¶

A sweep is a card on the My plots sub-tab, beside your other plots. + New sweep starts one on the first tests of the lot, and Edit opens it beside a live copy of the curve. A sweep is kept, duplicated, deleted, exported and imported like any plot, and is offered in the right-click menu's Plots section. It exists for a test program that measures the same quantity at a series of drive levels and records each level as its own test number — often one block sweeping up and another sweeping down. Read one test at a time that is a row of unrelated distributions; read as a sweep it is a pair of response curves, and the card measures the pair.

Each line is the population median, with a shaded p10–p90 band — not one trace per die. A lot is thousands of dies, so per-die traces would be a solid block of ink. The band is the spread across the dies currently in scope, so a band that widens at one end of the sweep is telling you the population disagrees most at that level.

The dies behind it are whatever Insights is scoped to. A sweep names tests, not dies — so the group or wafer selected in the panels above it decides the population, and the card's die count states what that came to. To sweep a hand-picked set of dies instead, select them on a map and right-click (see Charting dies and wafers).

Two measurements are drawn on the plot and restated underneath:

• Crossing — where the first two curves meet. If they cross more than once the card says so rather than reporting the first as though it were the only one.
• Separation — the horizontal distance between the two curves at the levels the application asked about. With one curve rising and one falling the pair traces a V, and this is the width of that V. A level that one of the curves never reaches reads not measurable, naming the curve — never 0, which would read as "they meet here".

Only the first two curves are measured. Any further ones are drawn for context.

Editing a sweep. The editor has a title, one block per series, and the axis and measurement settings. For each series, type its tests in sweep order: test numbers and ranges, such as 1010, 1011, 1020..1030 (a range means the tests declared between its ends). Then choose where each test's X comes from: Test order (the position in the run, labelled by test), Values (one number per test, in the same order) or From test name (a pattern such as LEVEL_{x}, for programs that record the swept value only in the test text). The editor lists the first few test names with what your pattern reads from each as you type, and Pattern examples shows worked ones: @ {x} reads 0.55 from Fmax @ 0.55 V, and -{x} reads 5 from 1234-5). Text that is not a test number or a range is flagged under the box, naming the word, and is not applied, so the curve beside it never shows something you did not mean. + Add series adds a curve and Remove drops one; the crossing and widths appear once there are two.

A sweep remembers what its tests were called. The editor records each test's name alongside its number. Open the sweep on a lot where a number now carries a different name (a test program that reuses numbers for other measurements) and that test is not drawn: the card names it, and the crossing and widths are not measured, rather than presenting another quantity as the original curve. Renaming a test in the selector has the same effect. A sweep written by hand, or imported from a file without names, is not checked.

The x axis is the sweep, not the lot — that is what distinguishes this card from the wafer-to-wafer trend, which walks one test across wafers. If the application supplied the real swept quantity, the axis is in those units (dBm, volts, °C) and the crossing is reported in them. If it did not, the axis is the position in the sequence, labelled by test, because test numbers are identifiers — treating them as a scale would invent spacing the data never claimed.

Some sweeps step by multiples — thresholds of 1k, 2k, 5k … 1M — and are drawn on a logarithmic x axis, so each step gets the same room. There the crossing is found along the log axis, and a width is given as a ratio between the two curves, with the level each reaches: ×2.49 (15.9 kΩ → 39.7 kΩ). On a log axis the same shift is the same multiple anywhere along it, which a difference in Ω would hide.

The card footer reports anything that limits what you are seeing: tests in the sweep that have no definition, functional (pass/fail) tests inside a sweep, which have no value to plot, and series whose tests carry different units.

If the swept values no longer line up with the tests — usually because a test in the sweep is missing from this data — the footer lists the tests it did find, and the card is drawn in test order with the crossing and widths not measured. Pairing the remaining values with the remaining tests would put every later point at the wrong level, which gives a plausible curve and a wrong crossing.

A chart's buttons¶

Every chart card has two buttons at the right of its title. Expand (the toolbar's expand icon) opens the chart on its own in a large window; a chart already in its window does not show it. The ⋯ button opens a menu of what else the card offers about itself:

• Save as PNG saves the chart (see Exporting a chart below). A table has no image to save and offers its CSV instead.
• Edit as plot, on the Insights charts drawn as plots (yield, bin pareto, pass rate, box plot, histogram, trend, scatter, and the value histogram of a selection), opens the chart as a draft plot of your own, drawn over the same wafers: the test and Group by it shows now are filled in as its own fields, so it draws the same thing on the next lot and follows nothing. It is kept only when you press Add to my plots. Greyed, with the reason, when the chart has nothing to draw yet (no test chosen).
• About this chart (About this table on a table) shows one line under the title saying what it shows and how to read it; choose it again to hide the line.

Use the arrow keys to move through the menu, Enter to choose, Esc to close it.

How to read this. A chart whose statistics this guide explains ends its note with How to read this and the section's title: PAT, What divides yield, the Comparison by table, process capability, the difference map, and the probability, control, goal and heatmap plots. A plot also names a section while it uses what that section explains: a fit line or the y = x line, a band, rescaled values, ranges of a number, panels, what-if limits, or groups compared wafer by wafer. A plain chart names none, as its About this chart says all there is. Each title is a link: it opens this guide at that section, beside the app, so you can read it with the chart in view. If the guide is already open, it moves to the section and comes to the front rather than opening again. A saved image or a printed page names the section instead.

Exporting a chart¶

Save as PNG in a chart's ⋯ menu saves the current view as a PNG at the displayed resolution. To get a clean full-resolution render, press Expand first to open it in the large window, then choose Save as PNG.

The saved image is the chart as shown, under a header that says what it is: its title, what its controls are set to (Test: vth_n_mV · 1001, Wafer: W03, Sort: yield), the lines the chart states about itself (the population and N, how values were combined, what the box and whiskers mean, how the bars are normalised), and its colour key (the legend, or the colour bar's name and range). The instructions that only make sense on screen ("click a point to open this wafer") are left out. A chart on the My plots tab carries the line stating its wafers and dies in the same place. The image takes the card's own background, so a dark theme saves a dark chart. The file name identifies the lot, wafers and chart (see below).

Printing. The dropdowns, toggles and buttons (Expand, the ⋯ menu button and, on the My plots tab, New plot, Examples, Import, Export, Delete all, Edit, Duplicate and Delete) are left off the printed page. In their place each chart prints a small block under its title giving the same settings and lines as the saved image, so a printed histogram says which test it is of.

Names of saved files¶

Every image and CSV you save is named for what it shows: the lot, then the wafer (or how many wafers), then the content. For example, LOT123_W05_hard-bin.png, LOT123_W05_die-list.csv or LOT123_25-wafers_yield-by-wafer.png. Saving the same export from different wafers gives different names, so files don't overwrite each other or need renaming. A part is left out when the data doesn't have it, for example a wafer with no ID. Your application may choose its own name for map images instead.

9. Reticle overlay¶

When reticle (stepper field) geometry is configured, the Reticle grid overlay draws the stepper field boundaries on the wafer. Each rectangle represents one exposure field from the lithography stepper — the group of dies exposed in a single step. This lets you correlate failure patterns with specific reticle positions — useful for identifying stepper field signatures, alignment drift, or mask defects.

Reticle-position findings in the Summary panel's Findings section highlight the specific field positions that show elevated failure rates. Hovering any die also shows its Reticle (column, row) position — its location within its own stepper field — just below the die's coordinate in the tooltip.

10. The statistics, explained¶

This section is for reading the statistics in Insights and the My plots tab for the first time. Each entry says what question the figure answers, how it is worked out in plain terms, how to read it, and what it cannot tell you. None of these figures is a verdict: they say how surprising a difference is, and the decision stays with you.

10.1 Why wafers are counted, not dies¶

Dies on one wafer went through the same processing, so they are not independent samples of anything. If a split of three wafers is compared die by die, thousands of dies make a tiny offset between two wafers look certain. Wherever the question is "do these groups differ?" (the Comparison by table, What divides yield, the note under a box or bar divided by a wafer field), each wafer is one value, the mean of its dies, and n is the number of wafers. With two or three wafers in a group many real differences come out inconclusive. That is the true state of the evidence, not a fault; more wafers are what settle it.

10.2 Comparing groups of wafers¶

• Welch p answers: if the groups had the same true mean, how often would wafer means differ at least this much by chance? It is Welch's t-test for two groups and Welch's ANOVA for more. It does not assume the groups have equal spreads, which is safer for splits. A small p (conventionally below 0.05) says the difference is unlikely to be chance; it does not say the difference matters.
• Rank p asks the same on the ranks of the wafer values (Kruskal–Wallis, with a correction for tied values), so one odd wafer cannot drive it. With fewer than five wafers in a group it is approximate, and is marked so.
• q corrects p for looking at many tests at once. Check 100 tests at 5% and about five will "differ" by chance alone. q is Welch p adjusted by the Benjamini–Hochberg method so that, among the rows with q under 0.05, about 5% are expected to be false alarms. The table is sorted by q. Read q, not p, when scanning many tests.
• Against a control: choose the process of record and each other group's cell adds its difference from it, with a p adjusted by Holm's method for the number of groups compared, so that comparing several groups with one control does not multiply the chance of a false alarm.
• Largest shift is how far apart the group means are, in wafer-to-wafer standard deviations of the test and, where it has both limits, as a share of the limit window. This is the size of the effect. A tiny p with a tiny shift is a real but unimportant difference; a large shift with a poor p needs more wafers.
• Cautions. A group with fewer than two wafers is left out and named. The tests assume the wafer means are roughly symmetric; the rank p is the check. Nothing here says why groups differ.

10.3 What divides yield¶

The What divides yield card makes the comparison Comparing groups of wafers describes for every wafer field at once, for yield and for each of the five most common fail bins (a bin's value is its share of a wafer's judged dies). It has one row per field, showing its yield, or a fail bin when that differs more clearly. The Effect column draws the worst and the best group as two bars on a scale of 0 to 100%, each with its mean, with a sentence giving the gap in percentage points; hover a bar for the group's median, spread and wafer count. Confidence is Clear (q under 0.05), Possible (under 0.2) or Not clear, so the fields at the top separate the wafers more clearly than wafer-to-wafer noise would; hover it for q, Welch's p and the rank p. A button on a field opens its other measures. The Group by button regroups every chart by that field. Each field is read after allowing for the strongest other field that explains the same measure (and does not contain it): a probe card on wafers that also differ by recipe is ranked by what the recipe leaves, and its row says "allowing for" that field, with the degrees of freedom the removed means used charged to the p. A field is compared when at least two of its groups have two wafers or more and it has no more groups than half the wafers (two, with fewer than six); What stands out and the example plots compare the same fields. Two limits are stated on the card itself: a field that gives every wafer its own value cannot be tested, and fields that change together (a lot and its start date, or a tester and the handler docked to it) cannot be told apart by the wafers; the card names each set of them, and names fields that nearly match (most wafers in matching groups) with how many wafers agree. A low q points to where to look; it is not a cause. Tool, chamber and probe-card fields appear only when the host supplies them as wafer attributes.

10.4 The cumulative probability plot¶

Each group's values are sorted, and the k-th smallest of n is drawn at height (k + ½) / n (Hazen's plotting position), so the curve is the cumulative share of dies at or below each value. On the default normal axis the vertical scale is stretched by the normal distribution, so values that follow a bell curve fall on a straight line: the middle of the line is the median, its steepness the spread. Read it by comparison: a curve shifted sideways is a shift in the mean, a flatter one a wider spread, a bend or a step a second population, and points off the ends of the line outliers or heavy tails. It needs no choice of bin width, and many groups overlay legibly. Choose linear for the plain 0–100% cumulative share. Large groups are drawn from a sample of 1,500 points with both tails kept in full.

10.5 Fit lines, R² and y = x¶

• Least squares is the usual regression of Y on X: the line that minimises the vertical distance to the points. Use it when X is the cause or the controlled quantity.
• Orthogonal (Deming regression with equal errors) minimises the distance at right angles to the line. Use it when X and Y are two measurements of the same quantity (one test on two testers, two insertions), where least squares tilts the line towards flat.
• R² is the square of the correlation of the points: the share of Y's variation a straight line in X accounts for. It says nothing about whether a line is the right shape; look at the points. y = x draws the line of perfect agreement, offered only when X and Y share a unit.

10.6 Bands on a line¶

A band round a line of means shows the spread of the values combined at each X: ±1 SD (mean plus and minus one standard deviation: about two thirds of normally distributed values), 95% CI (the range that would contain the true mean 95% of the time, from Student's t, so it narrows with more dies and says nothing about how spread out the dies are), Quartiles (the middle half of the values, Q1 to Q3) or Range (smallest to largest). Use SD or quartiles to see how much the values vary, CI to see how well the mean is pinned down.

10.7 Control charts¶

The Control chart is an individuals (I) chart: one value per wafer in order, with the centre at the mean and control limits at ±3σ. σ is estimated from the average moving range (the difference between each value and the one before) divided by 1.128, so it measures wafer-to-wafer noise and a slow drift does not widen the limits it is judged against. The control limits are worked out from the wafers on the chart: they say what this run normally does, not what any limit your team sets for the test allows, and so answer a different question. Points are flagged by four Western Electric rules, each at the point that completes it: 1 one point beyond 3σ, 2 two of three beyond 2σ on the same side, 3 four of five beyond 1σ on the same side, 4 eight in a row on one side of the centre. A flag means the point is unlikely to be chance if the process were stable; with many points a few false flags are expected. Limits from the first sets the centre and σ from a baseline of earlier wafers and judges the later ones against it. It needs at least three points.

10.8 Capability, Ppk and the goal plot¶

Ppk compares a test's spread with its limit window, penalising an off-centre mean: with mean μ, standard deviation σ and limits LSL and USL, Ppk = min(USL − μ, μ − LSL) / 3σ. A Ppk of 1 means the nearest limit is three σ from the mean; 1.33 is a common target, though yours may differ. The σ here is the plain standard deviation over every die (Cp and Cpk, in the tables, use the pooled within-wafer σ). The goal plot puts each test at (offset of its mean from the centre of the window, σ), both as shares of the window, so points on one triangle share one Ppk. A test short of the target is either to one side (off centre: shift the process) or high (too wide: reduce the variation). Spec limits are used where a test has both kinds, otherwise the test limits, and the chart says which. Below 30 dies a Ppk is a rough estimate. Capability indices assume roughly normal values; check the distribution before quoting them.

10.9 Rescaling values¶

Tests in different units can share one axis once rescaled, each on its own: To limits (0 at the low limit, 1 at the high one), z-score (standard deviations from the mean of the values in view), or Robust z (the same from the median, with spread taken from the median absolute deviation × 1.4826, so outliers do not set the scale). A z-score is relative to the population in view, so it changes when you narrow the wafers; rescaling to limits does not.

10.10 Ranges, panels and what-if limits¶

• In ranges cuts a measured value into 2–20 ranges. Equal counts gives each range about the same number of values (quartiles, deciles), so every box has a similar n; Equal widths gives each the same span of values. The edges are set over every die plotted, so panels and colours share them.
• Panels repeat the plot per value of a field on shared axes and colours, so panels can be compared by eye.
• What-if limits (histogram and probability plot of one test) draw proposed limits and state the test's pass rate now and under them, and how many dies change. It is the pass rate of that test over the dies it judges, not the yield, which depends on every test and bin. Treat it as a screening aid for a limit discussion.

10.11 Heatmaps and the difference map¶

A heatmap colours each row-and-column cell by the value combined over its dies (a count when there is no value), on a colour scale that spans the cells. Its colour is the same combination a bar would show, so the same rules apply. The difference map (Distributions, with Group by on) shows, at each die position, one group's mean minus another's: coloured on the value gradient, symmetric about zero so that the middle is no difference, drawn on the first wafer's geometry. A position is shown only where both groups have a value. A pattern on it (an edge ring, a centre spot) says where on the wafer the two processes differ, not whether the difference is significant.

10.12 Part average testing (PAT)¶

What it is for. A die can pass every test limit and still sit far from its neighbours' distribution on a test. Such a die is statistically unusual and, in automotive and other high-reliability work, is the kind of part PAT screens out.

How the limits are set here. Per wafer, and for every parametric test with both a low and a high limit:

1. Take the values of the dies that pass their bin and lie inside the test's limits (spec limits where the test has both kinds).
2. The robust mean is their median; the robust σ is (Q3 − Q1) ÷ 1.35, the interquartile range scaled to a standard deviation. Both resist outliers, so an outlier cannot widen the limits that are meant to catch it.
3. The PAT limits are the median ± 6 robust σ, never wider than the specification limits (the test's limits where it has no specification limits): AEC-Q001 requires that PAT limits do not exceed the device specification limits.
4. A wafer needs 30 or more such values on a test, and some spread, before it has PAT limits for it.

Verdicts. Each die is Within PAT limits, a PAT outlier (it passes its bin and every test limit but is outside the PAT limits of at least one test), or Fails (bin or limits) (already rejected, so not an outlier). The Overview card counts them per wafer and names the tests the outliers are outside on; the Dies table (and its CSV) has a PAT column and, where there are outliers, PAT tests; the plot builder takes PAT as a colour, category or panel field and draws a wafer's PAT limits as dotted lines.

A wafer in the PAT verdict plot mode: the outliers (amber) gather in the middle, which the bin and test limits alone do not show.

What this is and is not. The limits follow the robust-statistics method of AEC-Q001 Rev D (section 2.4: the median as the robust mean, (Q3 − Q1) ÷ 1.35 as the robust sigma; section 3.1.2: dynamic limits of ± 6 robust sigma from the lot's or wafer's own passing parts), here per wafer. That guideline applies PAT while parts are tested: parts that pass the static limits are held until the dynamic limits are known, and those outside them are rejected. Here the data is already collected, so this shows which dies PAT would have flagged; it is not the screening itself, which belongs in the test programme under a documented procedure. The guideline says no claim of conformance may be made unless all its requirements are met, so nothing here claims it. The differences:

• No static limits. The guideline bases dynamic limits on static ones, set from at least six lots with at least 30 parts each (at least 5 die in different areas of each wafer, 30 per lot) and reviewed during the first six months of production and twice a year after. Here the dies that pass their bin and the test limits stand in for the parts that passed the static limits.
• Only a warning for values that are not roughly normal. The guideline accepts other methods for data that clearly deviates from normal, with statistical justification. Here every test gets the same symmetric limits, so a lopsided test has limits that are too tight on one side and too loose on the other. The Overview card lists the tests whose middle half of values is lopsided (a quartile skewness of 0.2 or more either way, (Q3 + Q1 − 2 × median) ÷ (Q3 − Q1)) and on how many wafers; it does not change their limits, and it sees the body of the distribution, not a few heavy tails or two humps.
• Every test with both limits is judged. The guideline considers PAT for tests whose distribution is suitable for outlier analysis, and lists tests that shall use PAT limits at test (pin leakage, standby supply current, IDDQ, output levels, propagation delays and others, in its Appendix 2). Nothing here checks a test's distribution or matches tests to that list.
• A floor of 30 passing values per wafer. That floor is this library's own choice. The guideline's 1.35 is inexact for under 20 values, and sets sample sizes only for static limits.
• Quartiles are found by interpolation between the ranked values, which the guideline does not specify.

Neighbourhood screens (good die in a bad neighbourhood) are not part of AEC-Q001 and are not in this card. Treat the counts as a way to find out where PAT would matter and which tests drive it.

10.13 Example plots¶

Examples chooses its tests from the data in view rather than the order they are listed in. The lead test is the one with the largest share of its variation lying between wafers (for a single wafer, the one most tied to distance from the centre); its partner is the test that follows it most closely over a sample of dies (correlation above 0.3); the split is the lot field whose groups explain most of the lead test's variation between wafers, when they explain more than 5% of it. It reads at most 4,000 dies and 60 tests, so it is quick on a large lot. These are starting points for exploration, chosen by simple rules, and are not findings.