Inline Measurement for Semiconductor Manufacturing

Semiconductor manufacturing is the one industry where "measure it later at the CMM" fails most expensively. Lead frames move through stamping and etching lines at strip speed, packages go through singulation and lead-forming in seconds per unit, and a rejected lot discovered at final inspection represents weeks of already-processed value. The POMEAS HM series inline image measuring heads exist for exactly this gap: a telecentric measurement head mounted on the production equipment itself, measuring critical dimensions on every part, in cycle, while the part is still where it was made.

This page is the semiconductor companion to two existing guides. The HM series system overview explains the measurement principle and the general architecture; the automotive powertrain guide works through shaft-class parts. Here we stay on semiconductor feature families: lead frames, package leads and bodies, wafer-edge and die-level features, and the precision equipment parts that semiconductor tool makers must verify. For context on the two sibling optical approaches, see telecentric lenses for semiconductor inspection and flash measurement for semiconductor wafers.

Why inline, specifically, on a semiconductor line

Three properties of semiconductor production push dimensional measurement onto the line rather than into a lab:

• Volume per feature. A lead frame strip carries dozens of leads; a cassette of packages carries hundreds of lead positions. Sampling 5 pieces per shift on an offline instrument cannot catch a punch that starts drifting at lead 40,000 of a run. An inline head measures every part, so the drift is a trend line, not a surprise.
• Feature fragility. Dam bars, tie bars and etched leads deform if they are re-clamped for offline measurement. Measuring in the original fixture, in the original orientation, removes one whole class of handling error.
• Cycle-time coupling. Singulation, forming and taping equipment already stops for a pick-and-place index. A head that measures inside that index adds no station, no operator and no WIP buffer.

The three feature families the HM heads actually cover

1. Lead frame features on strip

On a stamped or etched lead frame, the dimensions that decide yield are lead width and pitch, tie bar width, dam bar opening and the pad window. These features are small relative to the strip, and the strips themselves are commonly wider than any single optical field - so the honest framing is feature-family measurement, not whole-strip scanning. A head with a 40 mm or 65 mm field placed over one lead group measures that group completely; multiple heads, or one head on a programmable stage, cover the groups across the strip. What the head cannot do is replace a dedicated full-strip AOI system for cosmetic defect hunting; the HM head measures dimensions on features you point it at.

2. Package leads and body features

After molding and singulation, the measurable questions are body length and width, lead count and pitch, lead shoulder width on gull-wing parts, and J-lead hook geometry on the parts that use them. Most QFP and BGA body sizes sit comfortably inside a 40 mm to 65 mm field (typical industry package sizes; confirm against your own package drawings - POMEAS does not publish a package-size table). The same head also verifies coplanarity-adjacent 2D projections - lead tip offset, shoulder span - while leaving true 3D coplanarity to the dedicated metrology stations that already exist for it.

3. Wafer-edge, die-level and equipment-part features

A 300 mm wafer cannot fit in any HM field - the largest published field in the series is 120 mm - and the HM heads are not the tool for whole-wafer overlay or map-style inspection. Where they do fit: edge and bevel feature dimensions at the wafer periphery presented by the handling system, individual die features after dicing (scribe-line width, pad window on a single die), and the semiconductor equipment parts themselves - quartz rings, ceramic work holders, guide pins and ESD-safe fixture components, which are precision-machined parts like any other and need the same diameter and groove checks as automotive hardware.

Head-by-head fit: published fields and accuracy

The published figures below come from the current HM series specification sheet. Repeatability is the short-term scatter of the head on a stable target; position measurement accuracy is how close a measured dimension lies to a calibrated reference. Both are needed to judge whether a head can hold a given drawing tolerance, and the arithmetic in the next section shows why they must be added, not compared.

Reading the accuracy figures against semiconductor tolerances

Semiconductor dimensional tolerances are tight, and the honest way to size them up is budget arithmetic. Take a lead width tolerance of +/-10 um on a copper alloy frame - a common order of magnitude for stamped frames (confirm against your drawing; this page uses it as a worked example). On the HM-1040 (phi40 mm field), published repeatability +/-0.08 um plus position accuracy +/-1.0 um gives a combined instrument uncertainty of about +/-1.08 um (derived: 0.08 + 1.0), which is 10.8% of the +/-10 um band (derived: 1.08 / 10). That leaves roughly 89% of the tolerance for the process itself, which is the healthy direction: the measuring system should consume a small fraction of the band it polices.

Now tighten the feature. A scribe-line or lead-width spec at +/-2.5 um on the same head consumes 43% of the band (derived: 1.08 / 2.5), which is marginal; dropping to the HM-1006 (phi6 mm, +/-0.05 um + +/-0.2 um = +/-0.25 um combined, derived) brings consumption down to 10% of the same band (derived: 0.25 / 2.5). The trade is field of view: the tighter the tolerance, the smaller the field you should buy. This is the same rule the battery cell guide applies to tab dimensions, and it is the first question to settle in any HM selection: which field, for which feature family.

Repeatability is not the number that decides the tolerance. A head that repeats to +/-0.05 um can still read 1 um high if its scale or alignment is off; repeatability describes stability, position accuracy describes truth, and both appear additively in the budget. A head with +/-0.05 um repeatability can hold a +/-5 um tolerance on a line and be working exactly as specified even if its absolute reading needed a one-time calibration offset.

Environment: temperature, fixtures and what the spec sheet does not include

The published accuracy figures are measured under laboratory conditions on stable targets. On a semiconductor line, three environment terms decide whether those figures survive:

• Thermal expansion of the part. Copper alloys expand by roughly 17 um per meter per kelvin and silicon by roughly 2.6 um per meter per kelvin (standard textbook CTE values; per-part arithmetic is derived). A 100 mm copper frame feature shifts by about 1.7 um per kelvin of temperature difference (derived: 100 mm x 17 x 10^-6 x 1 K = 1.7 um). If the head was mastered at 22 C and the strip runs at 26 C, that 4 K difference is about 6.8 um on the same feature (derived: 1.7 x 4) - larger than the position accuracy of any head in the table. The fix is procedural, not optical: master the head at line temperature, or compensate thermally.
• Fixture repeatability. The head measures what the nest presents. If the part sits 0.3 mm deeper into the nest than the master piece did, every dimension referenced to the nest datum moves with it. Fixture repeatability should be quantified on the real tool, not assumed; the powertrain guide expands on the same fixture arithmetic.
• Edge definition on the process. Etched or plated lead edges, mold flash on a body edge, and dicing chippew on a die edge all move the apparent edge that the algorithm locks onto. These effects are real but not quantifiable from the specification sheet; they must be characterized on customer parts (process-specific - UNKNOWN until measured on your parts).

HM inline head vs flash measurement vs telecentric station

Three optical approaches overlap on semiconductor parts, and picking the wrong one costs a project more than any accuracy figure. The decision rule by job:

A practical line often ends up with all three: HM heads on the forming press, a flash system at the outgoing gate, and a telecentric or zoom station in the quality lab for review. They are complements, not competitors - the mistake is buying a lab instrument to do a line job.

Integration notes that decide success in practice

• Mount where the part is already held. The cheapest integration point is the station that already clamps the strip or package. Adding a measurement station with its own handling usually doubles the project cost.
• Match the field to the feature family, not the part. Buying the phi120 head to "cover everything" and then pointing it at 0.3 mm leads wastes both field and accuracy headroom; a two-head solution (one small, one large) usually beats one oversized head.
• Plan the mastering routine. A calibration artifact measured at line temperature at shift start converts the thermal discussion in the previous section from a risk into a checkbox.
• Wire results into SPC, not just pass/fail. The value of 100% measurement is the trend. The head's value is realized when lead width drift is visible two hours before it becomes scrap.

Frequently asked questions

Related products and reading

• POMEAS HM telecentric measurement system for in-line inspection - the series overview and configuration options
• HM-1006 inline measuring head (phi6 mm)
• HM-1015 inline measuring head (phi15 mm)
• HM-1040 inline measuring head (phi40 mm)
• HM-1065 inline measuring head (phi65 mm)
• HM-1120 inline measuring head (phi120 mm)
• Inline Dimensional Measurement System guide (HM series principle)
• Telecentric lenses for semiconductor inspection
• Flash measurement for semiconductor wafers
• Inline measurement for battery cell and tab inspection
• Inline measurement for automotive powertrain parts
• What is a flash measuring machine?