Telecentric Lens for Lead-Frame Coplanarity: What Actually Resolves

Coplanarity is the one lead-frame measurement that a standard lens gets wrong in a way that looks right. A conventional lens projects a lead tip that sits 0.05 mm higher than its neighbours to a slightly different image position — not because the lead is out of place, but because the magnification changed with object distance. Object-side telecentricity removes that term. What it does not remove is everything else, and this page is mostly about the part it does not remove.

What coplanarity actually measures

The spec is usually stated as the maximum vertical distance between the lowest and the highest lead seating plane of a package, measured from a defined datum (typically the package body). It is a Z-axis quantity. A single 2D camera does not measure Z. What it measures is the projected position of each lead tip in a silhouette, and the Z information is inferred from that projection.

This is the reason the lens choice dominates the result. If the projection magnifies differently at different object distances, a lead that is perfectly seated but 0.05 mm further from the lens reads as laterally displaced. On a 0.5 mm pitch package that displacement can be a meaningful fraction of the tolerance.

Why object-side telecentricity is the non-negotiable part

In an object-side telecentric lens the aperture stop sits at the rear focal plane, so chief rays on the object side run parallel to the optical axis. The consequence is that magnification does not change as the object moves along Z within the depth of field. A lead tip 0.05 mm higher and a lead tip 0.05 mm lower project at the same magnification.

What telecentricity does not do: it does not make the depth of field larger, it does not improve resolution, and it does not turn a 2D image into a height map. Those three are the most common over-claims made for telecentric optics in coplanarity applications.

Telecentric models in the magnification range these packages use

Magnification between 1.0X and 2.0X covers SOIC through small QFP lead rows on 2/3″ and 1″ sensors. Sorted here by optical resolution, finest first — resolution is what sets how repeatably a lead edge can be extracted, and therefore how tight a coplanarity tolerance the station can hold. Full FOV arithmetic is in the calculation page.

Backlight geometry and what it hides

Coaxial or near-parallel backlight gives the high-contrast silhouette that makes edge extraction stable. It also collapses the part into a shadow, which is exactly where the measurement limits appear:

• Gull-wing leads — the silhouette shows the foot, which is the seating plane. This is the geometry coplanarity is defined on, and it reads well.
• J-lead — the seating surface is tucked under the body. The silhouette shows the outer bend, not the foot. What you measure is a proxy, and the offset to the true seating plane has to be established by correlation against a contact or confocal reference, not assumed.
• BGA and leadless — there is no lead to project. Ball coplanarity is a 3D measurement and needs a confocal, interferometric or laser profile sensor. See when a telecentric lens is the wrong choice.

Choosing magnification from the lead row, not from the lead

The instinct is to magnify until a single lead fills the frame. That is wrong. Coplanarity is a relative measurement across all leads on a package, so every lead has to be in the same image, in focus, at the same magnification. The field of view is set by the lead row width plus margin, not by one lead.

Worked backward: a QFP with a 14 mm lead row on a 2/3″ sensor (11.0 mm horizontal) needs roughly 11.0 / (14 × 1.3) ≈ 0.6X to sit comfortably with margin. A smaller SOIC with a 6 mm row can take 1.0X or higher. The magnification follows from the part; the resolution you get is then sensor pixel size divided by magnification. Use the FOV and magnification calculation rather than estimating.

Where the measurement actually fails

Three failure modes account for most coplanarity disputes between a vision system and a contact CMM:

• Part tilt. If the package is not parallel to the sensor plane, one end of the lead row is further away. Telecentricity makes the magnification immune to this, but the leads at the far end are now near the edge of the depth of field and their edges blur. Blur moves the extracted edge.
• Lead shadowing. On dense packages an out-of-plane lead can be occluded by its neighbour in the silhouette. The system reports fewer leads than exist.
• Datum drift. Coplanarity is measured from the package body, but the body is usually not in the silhouette at the same focus. If the datum is fixtured rather than imaged, the fixture becomes part of the measurement.

None of these are solved by buying a better lens. They are solved by fixturing, and by deciding up front that the system reports a proxy that has been correlated to a reference method.

Practical setup order

Fix the lighting first — a stable silhouette is worth more than an extra megapixel. Then set magnification from the lead row, then choose the sensor so that pixel size divided by magnification gives you the sub-pixel resolution the tolerance needs, and only then compare lens models. Reversing this order is the usual reason a coplanarity station gets rebuilt.

For the lighting decision specifically, the telecentric lighting guide covers backlight, coaxial and dark-field selection for measurement.

Questions engineers ask about this setup

Can a telecentric lens measure coplanarity directly in millimetres?

No. A single 2D telecentric view measures the projected position of lead tips. Coplanarity in millimetres is inferred from that projection. The inference is valid when the silhouette shows the actual seating surface — true for gull-wing leads, not for J-leads or BGA balls, where the measured feature is a proxy that must be correlated against a contact or confocal reference.

Why do my readings disagree with the CMM on the same parts?

The three usual causes are part tilt putting one end of the lead row at the edge of the depth of field, lead-on-lead occlusion in the silhouette, and a datum that is fixtured rather than imaged. Telecentricity removes the magnification error but does not fix any of these. Check the fixture before questioning the optics.

Should I use the highest magnification available?

No. Coplanarity is a relative measurement across the whole lead row, so every lead must be in one image at the same magnification. Field of view is set by the lead row width plus margin, which usually puts QFP-class packages near 0.5X to 1.0X and only small SOIC packages above 1.0X.

Does telecentricity increase depth of field?

No. Telecentricity makes magnification insensitive to object distance; depth of field is governed by aperture and magnification as in any lens. A part tilted enough to leave the depth of field will still produce blurred edges, and blurred edges move the extracted measurement.

What lighting is required for lead inspection?

A high-contrast silhouette, which means backlight with a collimated or near-collimated source. Coaxial front light is useful for inspecting lead surface condition and marking, not for coplanarity, because it does not produce a clean silhouette edge of the lead foot.

Can the same station measure J-lead packages?

It can image them, but the silhouette shows the outer bend rather than the seating foot, so the number it returns is a proxy. Treat it as a go/no-go correlated to a reference method, not as a direct coplanarity value, and validate the correlation on the actual package family.

Related reading

• Telecentric lens selection guide
• Magnification and FOV calculation
• Lighting for telecentric measurement
• What a telecentric lens is and when to use one
• When a telecentric lens is the wrong choice
• Machined parts: contour and spacing
• Battery electrode and tab
• Spring: wire, pitch and free length

Send one part and we will tell you what the measurement can hold

Specification arithmetic gets you to a shortlist, not to a verdict. Send one part with the drawing and the tolerance you have to hold, and we will run it and report what the system returns — including the features that come back as unmeasurable. A written feasibility note is worth more than another datasheet.