Telecentric Lenses for Semiconductor Inspection: Model Selection by the Numbers

Semiconductor inspection punishes optical shortcuts. A scribe line that drifts a few micrometres, a die misaligned by a fraction of its own pitch, or a lead frame that bows under coating - each of these turns into a pass/fail error if the optics magnify differently at different depths. This guide maps the POMEAS telecentric lens family onto the three inspection layers of a semiconductor line - wafer, chip and package - using only the measured specification values from the product datasheets, so every number below can be traced to a live product page.

Why a standard FA lens fails in semiconductor metrology

A conventional FA lens changes magnification when the object moves toward or away from the lens. On a polished wafer or a non-flat package surface that movement is guaranteed, and the resulting scale error is indistinguishable from a real dimension error. Telecentric designs hold magnification constant over a defined depth range. The POMEAS high-resolution telecentric series specifies object telecentricity below 0.02° (for example the PMS-LTCM15-110), and TV distortion between 0.01% and 0.04% across the models listed below. In practical terms: on a 2/3-inch sensor imaging an 8.8 mm wide field, a 0.01% TV distortion stays under 1 μm of geometric error at the field edge, while a typical non-telecentric design can exceed that by an order of magnitude. For dimensional checks where the tolerance band itself is only a few micrometres, that difference decides whether the station can make a valid measurement at all.

Magnification and object resolution: match the lens to the defect size

The second selection criterion is object resolution - the smallest feature the lens can resolve on the object side. The verified values across the POMEAS telecentric range:

The engineering reading of this table: defects at 15–30 μm scale (solder bridging, lead contamination) sit comfortably inside the 0.5X model's 12.43 μm resolution with room to spare; die-level defects at 5–10 μm call for the 1.0X to 1.5X range; sub-5 μm features such as fine crack propagation need the 2.0X models, where the PMS-LTC2-40 reaches 3 μm object resolution. As a derived rule of thumb, a defect should be at least three times larger than the lens resolution for reliable edge detection with a production-grade sensor - this multiplier is a common machine-vision practice, not a datasheet value.

Working distance and field of view on real sensor sizes

Field of view is set by the sensor, and every model above carries a 2/3-inch maximum sensor (the bi-telecentric PMS-10110CVP111T accepts up to a 1.1-inch format). On a 2/3-inch CCD that is an 8.8 × 6.6 mm object field at 1.0X, which covers a single die or a cluster of passive components in one shot. When the station must see a larger area - a full lead frame quadrant, a wafer sector - the long-working-distance models step the magnification down instead of trading resolution for field: the PMS-LTCM016-189 works at 0.16X with 189 mm WD and resolves 25.17 μm, and the PMS-LTCM0061-398 covers 0.061X at 398 mm WD with 87.99 μm resolution. Both hold TV distortion at 0.01%, so position measurements across a wide field stay honest. The PMS-LTC1-200 offers a middle path: 1.0X magnification at 200±10 mm working distance with 22.37 μm resolution for stations that need clearance for handling equipment.

Depth of field versus surface topography

Depth of field is where semiconductor parts are hardest on optics. The verified range runs from 0.2 mm on the PMS-LTC2-40 up to 171.97 mm on the PMS-LTCM0061-398 - the latter exists precisely because at 0.061X magnification the depth range becomes large enough to keep an entire wafer surface within focus despite bow and waviness. The inverse relationship between magnification and depth of field across the table above (4.53 μm resolution at 0.27 mm DOF versus 12.43 μm at 2.96 mm) is the fundamental optical trade-off, and it forces a station-level decision: high-resolution die inspection uses a Z-focus stage, while full-surface wafer checks pick a low-magnification long-DOF lens and accept coarser resolution. Specifying both at once is not physically available in any telecentric design, so the honest answer is to split the inspection into two stations.

Coaxial illumination for reflective semiconductor surfaces

Polished silicon, plated lead frames and tinned package pads all behave like mirrors at inspection wavelength. Direct ring or bar lighting creates hot spots that drown edge contrast. The PMS-LTCM1-110C is the coaxial variant of the 1.0X telecentric series: identical optical core (7.46 μm resolution, 0.89 mm DOF, 0.02% TV distortion, 110 mm WD) with built-in coaxial lighting that sends illumination down the optical axis and returns the specular reflection from flat surfaces back into the lens. This is the default choice for laser-marked die reading, polished substrate checks and any shiny-metal surface work. For non-coaxial stations, the separate guide on telecentric lighting selection covers point, ring and coaxial trade-offs in depth.

Recommended starting points by application layer

Mapping the verified models onto the three semiconductor inspection layers: at wafer level, start from the PMS-LTCM0061-398 (398 mm WD, 171.97 mm DOF) or PMS-LTCM016-189 for full-surface presence, coating coverage and large-feature checks. At chip and substrate level, the PMS-LTCM1-110 or the bi-telecentric PMS-10110CVP111T (1.1-inch sensor, Φ18.4 mm object FOV) handle die positioning, marking verification and pad dimensional checks with 5.4–7.5 μm object resolution. At package and lead level, the PMS-LTCM2-110 and PMS-LTC2-40 deliver the 3–4.5 μm resolution that fine-pitch dimensional work requires. These mappings are engineering recommendations derived from the specification table above, not laboratory benchmarks on specific semiconductor devices; final selection always needs a sample-image trial on the actual part.

From lens to production line

A telecentric lens is the front end of a dimensional station, not the whole station. POMEAS builds the complete chain: the HM telecentric measurement system integrates the optics into an in-line inspection unit, and the inline dimensional measurement platform shows how the same optics family extends to non-contact production gauging. For related industry work with the same lens family, see the PCB and FPC inspection guide, the battery electrode tab application, and the gear pitch measurement case. To browse the full optic range, start from the telecentric lens catalogue. Each of those pages carries the same verified-spec discipline as this one, so a station designer can move from semiconductor selection to the rest of the line without leaving verified numbers behind.

Frequently asked questions

Why use a telecentric lens instead of a standard FA lens for semiconductor inspection?

Because semiconductor parts are rarely flat, and a standard FA lens changes magnification with object distance, turning surface bow into fake dimension errors. Telecentric designs hold magnification constant: the POMEAS models specify object telecentricity under 0.02° and TV distortion of 0.01–0.04%, so edge positions stay comparable across the full depth of field.

Which magnification should I choose for chip-level defect inspection?

Match the object resolution to the defect scale: 15–30 μm defects work with the 0.5X PMS-LTCM05-110 (12.43 μm resolution), 5–10 μm die defects need the 1.0X to 1.5X models (7.46 and 5.59 μm), and sub-5 μm features require the 2.0X PMS-LTCM2-110 or PMS-LTC2-40 (4.53 and 3 μm). Keep the defect at least about three times the resolution for reliable edge detection.

Can one telecentric lens cover a full wafer inspection?

Not at inspection magnifications: the field of view on a 2/3-inch sensor is 8.8 × 6.6 mm at 1.0X. For wafer-scale coverage the long-WD models step magnification down - the PMS-LTCM016-189 (0.16X, 189 mm WD, 25.17 μm) and PMS-LTCM0061-398 (0.061X, 398 mm WD, 87.99 μm, 171.97 mm DOF) keep large areas in focus despite wafer bow. Full-surface metrology and fine-defect checks are normally split into two stations.

How does coaxial illumination help on reflective semiconductor surfaces?

Polished silicon and plated leads mirror light away from the lens, killing edge contrast. The PMS-LTCM1-110C adds built-in coaxial lighting along the optical axis with the same 1.0X optical core - 7.46 μm resolution, 0.89 mm DOF, 0.02% TV distortion at 110 mm WD - so flat specular surfaces return illumination evenly and edges stay measurable.

Turning this selection into a working station

The model choice above fixes what the lens can resolve; the station still depends on how the lens is mounted and lit. Two follow-on reads make the build predictable: the telecentric lens selection guide covers magnification, sensor format and working distance as one calculation rather than three separate picks, and the inline measurement platform page shows what changes when the same optics move from a bench to a production line. If you would rather check a specific part, send the part drawing, the defect size you must catch and the stand-off your fixture allows through the contact page.