Flash Measurement System for Semiconductor Wafer Inspection

A wafer presents an awkward combination of requirements to an optical measuring instrument. It carries hundreds of identical sites that each need the same set of dimensions checked, it cannot be clamped or touched on the device side, and the features that matter most - die outline, dicing street width, bump and pillar diameters, array pitch, coating opening - sit in a plane that is flat only as long as nothing presses on it. A flash measurement system addresses that combination by measuring the whole field in a single exposure: the stage does not travel during the measurement, so there is no length reference to drift and no clamping force to distort the part. This page is about the numbers that decide whether that approach fits a given wafer-level job, and about the points where it stops being the right instrument.

Why wafer-level dimension checks suit a single-field instrument

Wafer dimensions are mostly local and repeated rather than long and continuous. Die size, street width, the diameter of a bump, the pitch between two pillars in an array: these are features measured inside a window of a few millimetres, and the same window repeats across the wafer. A stage-based measuring machine samples such a window feature by feature and moves between them, which is accurate but spends its time on motion and on re-datuming each region. A single-field instrument inverts the logic: it calibrates the field once, images the whole part in one exposure, and lets the software find every programmed feature in that image. On a part where the same twelve dimensions are checked at two hundred sites, that inversion is what turns an inspection step that took minutes into one that takes seconds per site, and it removes the motion axes whose wear and backlash would otherwise have to be characterised.

The second reason is mechanical. A thinned wafer, and a wafer held on tape or in a carrier, does not tolerate a contact probe or a clamp that loads the device side. A flash measurement system places no load on the part beyond its own weight, images it from above, and measures whatever the field can see. The published stage load for the IMAGE 3 series is 5 kg, which is far above what any wafer or carrier weighs, so the constraint on a wafer job is never payload. It is field size, calibration and the smallest feature the optics can resolve.

What a flash system measures on a wafer - and what it does not

The measurable list is the list of 2D features that fit inside one calibrated field and are larger than the optical resolution: die outline and die-to-die spacing, dicing street width, passivation or coating openings, bump and pillar diameters, array pitch, the offset of a structure from a die edge, and the kerf left after singulation. These are edge-based measurements on a high-contrast image, and they are what the instrument is built for.

The list it does not cover is just as important to write down before a quotation. Standard configurations measure 2D dimensions; height, step and film thickness are not part of the 2D measurement chain and require the optional 8055 point spectral confocal sensor, which is standard on IMAGE 3 MAX and optional on MAX2. Critical dimension work at the sub-micron and nanometre scale is outside the instrument's resolution class altogether - the smallest display unit of 0.1 µm is a display resolution, not a statement that a 0.1 µm line width can be measured with a usable uncertainty. Anything that needs a cross-section, a film stack or a sidewall angle is a different instrument category, and pretending otherwise is how a wafer inspection station ends up being bought for a job it cannot hold.

The field, the wafer and the arithmetic that decides fit

A flash measurement system works inside a single field, so the first check on any wafer job is whether the part fits that field. The IMAGE 3 series reaches a 300 × 200 × 75 mm field on IMAGE 3 PRO and PRO-H. A 200 mm wafer has a 200 mm diameter, which is exactly the short axis of that rectangle: it can be inside the field, but with no margin for a carrier edge, an alignment feature or edge exclusion. A 300 mm wafer does not fit at all - its diagonal is 300 × √2 ≈ 424 mm (derived), well beyond the 300 mm long axis, so no rotation brings it inside one field. The practical consequence is that the 300 × 200 mm class of flash system is a 200 mm-and-below instrument, and a 300 mm wafer job has to be solved with either a larger field, a stitched measurement on a travelling stage, or a per-site sampling plan that images only the regions that matter.

Field size also fixes the pixel budget. Because the whole field is captured at one magnification, the smallest reliably measurable feature is set by the field calibration and the sensor, not by a zoom setting - which is precisely the trade that is being made when a single-field instrument is chosen over a machine that can magnify one local region. On a wafer this shows up in a specific way: a full-field view of a 200 mm wafer is the right view for die placement and street width, and it is the wrong view for a 50 µm bump. The usual solution is two recipes with two magnifications rather than one compromise, which is why the optics in a flash system are specified as dual-magnification telecentric rather than as a single fixed lens.

Accuracy over a full field: what the 0.02L term does to a wafer recipe

Published accuracy for the IMAGE 3 PRO and PRO-H is ±2 µm in precision mode without a connected sensor and ±(2 + 0.02L) µm with one, where L is the measured length; wide-field mode is ±4 µm and ±(4 + 0.02L) µm on the same basis. The second term is the one that matters on a wafer, because wafer features are often far apart. Evaluated at three lengths that occur in practice, the length-dependent allowance works out as follows (all derived from the published formula):

Read that table against the tolerance on the drawing, not against the repeatability figure. A ±0.5 µm repeatability on a base model, or ±0.25 µm on MAX and MAX2, describes how tightly the same feature repeats when it is measured again - it says nothing about the error on a 200 mm dimension, which is dominated by the length term. A recipe that measures a whole-wafer distance and quotes the repeatability as its uncertainty is overstating the instrument by a factor of ten. The honest method is to keep long dimensions to a coarse tolerance, measure tight-tolerance features inside a small window where the length term is negligible, and state both numbers on the inspection report.

Optics and lighting on reflective and transparent wafer surfaces

Wafer surfaces are either mirrors or windows, and both defeat a naive lighting setup. A metal layer, a solder bump or a polished backside behaves as a curved mirror: ring light alone produces a bright specular band that reads as a displaced edge, and the software then measures the highlight rather than the feature. Coaxial illumination, delivered along the optical axis, keeps the specular return centred and makes the edge the brightest transition in the image. A transparent film or a bonded layer is the opposite problem: what has to be measured is an edge that is only visible in transmission, so backlight is the correct choice and reflected light shows the top surface instead of the feature of interest.

This is why the flash measurement system is specified with multi-zone lighting - ring, backlight and coaxial - switched programmatically per feature rather than fixed for the job. The optics behind that lighting are dual-magnification telecentric, which matters for a wafer for a reason that is easy to miss: telecentric imaging holds magnification constant when the object moves slightly in and out of focus, so a wafer that is not perfectly flat, or a carrier that sits a fraction of a millimetre high, does not change the measured size. On a wafer held by tape or vacuum on a carrier, that insensitivity is doing real work in every measurement.

Repeatability, throughput and the pass/fail decision

A wafer inspection station is judged on two numbers that pull in opposite directions: how many sites per hour it can clear and how confidently each site can be passed. The flash approach helps both by making the measurement program a fixed artefact. Features are taught once, stored as a program, and re-run without an operator aligning anything - the part can sit anywhere in the field at any rotation and still be measured, because the pixel-to-millimetre coefficient belongs to the calibrated field rather than to the part position. The camera is specified as a 12MP or 20MP monochrome sensor, and the software is the AI-Image measurement system with one-key measurement and one-key report output, so the pass/fail decision is produced as data rather than transcribed by hand.

What has to be managed is drift in the calibration itself. Because the field calibration is the length reference, the calibration artefact and its re-check interval are the single most important maintenance item on the station - more important than any lens cleaning routine. A practical schedule re-checks the field calibration at the start of each shift and after any change of magnification recipe, and records the residual so that a slow drift shows up as a trend instead of as a batch of parts that fail on one side of the wafer only.

Where a flash system stops and another head starts

Three boundaries are worth stating plainly, because each one is a place where a wafer project can be sold the wrong instrument. First, sub-micron critical dimension and sidewall work is outside this class: the smallest display unit is 0.1 µm, and the accuracy budget on a full field is measured in micrometres, so a CD-SEM class question needs a CD-SEM class answer. Second, high-magnification defect review on a single die - edge chipping, residue, a crack in a street - wants pixels concentrated on one small region, which is the job of a telecentric lens inspection station sized by magnification or of a motorized zoom head with an APO objective, not of a whole-field capture. Third, continuous in-line gauging of a moving product - wire, tube, extruded profile - is a different measurement geometry entirely and belongs to an inline dimensional measurement head.

The constructive version of those boundaries is a two-head arrangement: a flash measurement system for the dimensional set that repeats at every site, and a magnified head for the small number of features that need resolution rather than field. Both read into the same QC record, and neither is asked to do the other's job.

Integration checklist for a wafer-level job

• Wafer size and carrier outline checked against the 300 × 200 × 75 mm field first; a 300 mm wafer fails this check before any optics are discussed.
• The dimension list split into two groups: features inside a small window, and dimensions that span the wafer. The second group carries the 0.02L allowance and must be quoted with it.
• Smallest feature of interest compared with the field calibration and sensor, not with the display unit.
• Height, step or film thickness identified up front if required, so the 8055 point spectral confocal option is priced with the station rather than after it.
• Lighting per feature decided with real samples: coaxial for metal and bumps, backlight for transparent films and through-features, ring for general edge contrast.
• Calibration artefact, re-check interval and residual log defined as part of acceptance, before the station is released to production.

Frequently asked questions

Can a flash measurement system measure a 300 mm wafer in one field?

No. The IMAGE 3 series reaches a 300 × 200 × 75 mm field on IMAGE 3 PRO and PRO-H, and a 300 mm wafer has a diagonal of about 424 mm, which exceeds the long axis as well as the short one, so no rotation fits it inside a single field. A 200 mm wafer is exactly the short axis of that field, which fits only with no margin for a carrier edge or alignment feature. A 300 mm job needs a larger field, a stitched measurement on a travelling stage, or a per-site sampling plan.

What accuracy should I budget for a dimension that spans 150 mm?

Use the length term, not the repeatability. With the published ±(2 + 0.02L) µm figure in precision mode with a connected sensor, a 150 mm dimension carries an allowance of 3 µm from the length term and about ±5 µm in total, both derived from the formula. Keep tight-tolerance features inside a small window where the term is negligible, and quote the two classes of dimension separately on the report.

Can it measure bump height or film thickness as standard?

Not as standard. The normal configuration measures 2D dimensions from one exposure. Height, step and film thickness need the 8055 point spectral confocal sensor, which is standard on IMAGE 3 MAX and optional on MAX2. If a height measurement is part of the requirement, it has to be specified with the station, because it changes the configuration rather than adding a software option.

Does a flash system replace a telecentric lens inspection station?

No, it complements one. A flash system measures a calibrated field and suits dimensions that repeat at every site; a telecentric lens station concentrates pixels on one small region and suits sub-micron detail, defect review and features at the edge of resolution. On a wafer line the usual split is a whole-field pass for placement and street geometry plus a magnified pass for the features that need resolution.

Next step

Send the wafer size, the carrier outline, the dimension list with tolerances, and the smallest feature that has to be resolved through the contact page. The field check comes first and it is arithmetic, so a 300 mm wafer job is settled in one reply rather than in a quotation round. For the instrument itself, the IMAGE 3 image dimension measurement guide covers the series and its configurations, the configuration page covers what changes between models, and the flash measurement versus VMM comparison covers the case where a travelling stage is the better answer. For a connector-level or stamped-part job on the same line, the flash measurement for connectors page applies the same method to a smaller part.