Panel-Level Packaging Metrology: Line Scan, White-Light Interferometry and Film Thickness on One Platform

Panel-level packaging moves inspection off the wafer and onto a substrate that no single field of view can cover. A public reference point: Chroma's 7505-01 multi-functional optical profiling system specifies a 700 × 900 mm platform range and combines 2D line-scan inspection, 3D white-light interferometry and 1D film-thickness measurement in one machine. That combination is the part worth studying, because the three measurements fail for completely different reasons and only one of them is about pixels.

The three channels measure three different physical quantities

2D line scan returns lateral geometry: line width, pad position, bridging, missing features. What limits it is sampling in micrometres per pixel and the optical resolution of the lens at that magnification. 3D white-light interferometry returns height: bump height, coplanarity, warpage, die tilt. Its vertical resolution is largely independent of the lateral sampling, but it depends on surface slope, reflectivity and vibration. 1D film thickness returns the thickness of a layered stack, usually from spectral reflectance, at a single point: it is not topography and it does not produce an image. Treating all three as settings on one sensor is the fastest way to write a specification that no machine can meet.

What a 700 mm panel does to line-scan sampling

Work the numbers on the short axis of that platform range. A 16K line-scan camera over 700 mm gives 700 mm ÷ 16,384 pixels, about 42.7 micrometres per pixel. If the defect class you have to catch needs 5 micrometres per pixel, one 16K line is nowhere near enough: 700 mm ÷ 5 micrometres is 140,000 samples across the panel, which is roughly 8.5 such cameras, or multiple passes, or a tiled arrangement. This is arithmetic from the platform dimension, not a performance claim about any machine — and it is the number that decides whether a single-pass architecture is even possible. For how magnification, line rate and sensor format interact on this class of optics, see 16K line-scan lens selection: magnification, line rate and sensor format, and the 4K 35 mm line-scan lens for 32 mm image circle cameras.

Where white-light interferometry stops being the right tool

Interferometry needs a usable fringe. Steep sidewalls, rough mould compound, transparent films and surfaces with mixed reflectivity can return no fringe or a wrong height, and no amount of camera resolution fixes that. The second limit is field: an interferometric objective sees a small window, so a whole-panel coplanarity map becomes a stitching problem with the same arithmetic as any large-panel AOI. The third is environment: a long-travel panel stage and an interferometer are natural enemies, and the isolation requirement belongs in the machine specification rather than in the commissioning report. Where the height feature is large and the surface is difficult, a chromatic confocal point sensor trades lateral detail for a more forgiving measurement — the range-versus-resolution trade is set out in confocal sensor resolution: why it depends on range. A dedicated 3D instrument such as the PMS-WLI100 3D white-light interferometer is built for exactly this channel.

Film thickness is a spectral problem, not an imaging one

Thickness from spectral reflectance depends on the optical constants of the stack. If the film model — number of layers, n and k values — is wrong, the number is wrong, and it will be repeatably wrong, which is why this channel is the easiest to over-trust. It is also a point measurement: covering a panel with it is a stage-time problem with the same shape as any other scanning measurement. Five thickness points on a 600 mm panel are five data points, not a uniformity result, and the specification should say so explicitly.

Assign each tolerance to one channel before comparing machines

Lateral dimensions — line and space, pad size, overlay targets — belong to the 2D channel, where the questions are micrometres per pixel and optical resolution at the working distance. Height features — bump height, coplanarity, warpage, die tilt — belong to the 3D channel, where the questions are vertical resolution, maximum measurable slope and field size. Dielectric or resist thickness belongs to the film channel, where the questions are the stack model and the spot size. Anything that needs all three carries the cost of all three, and throughput is set by the slowest channel, not the fastest.

Six questions to answer before you request a quote

Ask which channel is quoted, at what sampling, at what stage speed. Ask whether throughput is per panel or per field. Ask for the maximum measurable slope and the numerical aperture of the 3D channel. Ask how many layers the film model supports and who supplies the optical constants. Ask for the vibration and isolation requirement. And ask what holds the panel flat, because a substrate this large does not stay flat by itself — flatness is a fixture problem that later shows up as a focus and a stitching problem. For the wider choice between measurement principles on in-line work, see in-line dimensional measurement: gauge, confocal or vision.

Frequently asked questions

Is a 700 × 900 mm platform the same as a 700 × 900 mm field of view?

No. Platform range is how far the stage travels; field of view is what the optics capture in one shot. A large panel is covered by moving the stage or the optics across many fields, and the number of fields drives the cycle time.

Can one sensor do 2D, 3D and thickness measurement?

The three measurements use different physics: lateral imaging, interference for height, and spectral reflectance for thickness. One machine can carry all three channels, but it cannot measure all three at the same instant at the same point.

How many pixels do I need for 5 micrometre sampling across a 700 mm panel?

700 mm divided by 5 micrometres is 140,000 samples across the panel. A 16K line-scan camera delivers 16,384 pixels, so a single line gives about 42.7 micrometres per pixel and roughly 8.5 lines' worth of pixels are needed to reach 5 micrometres per pixel.

Does white-light interferometry replace a confocal sensor?

Not universally. Interferometry gives dense height data over a small field on surfaces that produce a clean fringe; a confocal point sensor is more tolerant of difficult surfaces and larger ranges but returns fewer points. The choice follows the surface and the feature size.

Can film thickness be mapped across a whole panel?

It can be sampled across a panel one point at a time. Whether that counts as a map depends on how many points the cycle time allows, and on whether the film stack model is correct for every layer present.

What limits throughput when three channels share one stage?

The slowest channel plus the motion between fields. Adding a 3D scan to every field usually costs more time than the 2D capture itself, and per-field autofocus multiplies across every field in the panel.