On this page
Three non-collinear height references define a plane, but they do not necessarily define a useful measurement reference at every feature on a ceramic circuit. If the points occupy a narrow strip, a small height change at one point can produce a large plane-height change at a distant pad. The important quantity is the reference-plane contribution at the actual inspection location. Calculating that sensitivity helps distinguish an unstable reference construction from real ceramic bow or a changed printed-film height.
Key design decisions
- Define an explicitly selected three-point construction; do not substitute it for the datum-association method required by the drawing.
- Evaluate the plane at the functional feature, including whether the feature lies outside the reference triangle.
- Separate independent point noise, common height offset and actual surface form before reducing them to one uncertainty.
1. State which plane the measurement is constructing
Identify the three physical reference locations and why they are appropriate to the question. They might be dedicated reference patches used for an internal comparative measurement. A plane passing through those observations is a mathematical construction; it does not automatically represent a mechanically mating plane over an imperfect surface or the specified datum of a geometric-tolerancing requirement.
Keep the drawing's required association and contact conditions intact. If the requirement calls for a different fitted or constrained datum, apply that method and analyze its uncertainty separately. The simple construction below is useful precisely because its inputs and sensitivities are explicit. Changing to an arbitrary three-point plane to obtain a favorable result would change the measurand rather than improve the measurement.
2. Make the reference triangle visible in coordinates
For a transparent example, place the reference points at in-plane coordinates zero, zero; L, zero; and zero, B. Their measured heights are z1, z2 and z3. The horizontal spans L and B are nonzero and known sufficiently well for the initial analysis. The constructed plane slopes are the corresponding height differences divided by those spans.
At a target coordinate X, Y, the predicted reference height is a weighted sum of the three input heights. The weights always add to one. Within the triangular region, each weight is nonnegative. Outside it, at least one can become negative, so the construction extrapolates rather than interpolates. A wide rectangular product outline does not guarantee a wide reference triangle.
zref(X,Y) = w1z1+w2z2+w3z3; w2=X/L; w3=Y/B; w1=1−X/L−Y/B; u(zref)² = Σwi²u(zi)² + 2Σi<j wiwj cov(zi,zj)
- L and B: reference triangle spans in matching length units
- X and Y: target coordinates in that same in-plane system
- zi: measured heights at the three references
- wi: dimensionless sensitivity weights; u: standard uncertainty; cov: covariance
A plane through the three specified non-collinear points is the intended construction. In-plane position uncertainty is negligible in this initial model; significant position error, actual surface form and target-height measurement require additional terms.
3. Calculate a narrow-triangle example
Assume L is 40 mm and B is 5 mm, with a target at X equal to 20 mm and Y equal to 30 mm. The weights are minus 5.5, plus 0.5 and plus 6.0. A positive 2 micrometer error at the third reference alone raises the constructed plane at the target by 12 micrometers. It does not merely add 2 micrometers to the target result.
If the three heights each have independent standard uncertainty of 2 micrometers, the reference-height standard uncertainty is 2 times the square root of 30.25 plus 0.25 plus 36. This is approximately 16.31 micrometers. These are assumed measurement inputs, not substrate tolerances. The result describes the constructed reference at one target, not the uncertainty of every point in the inspection field.
4. Evaluate a layout change at the same target
Now retain L at 40 mm but move the third reference to B equal to 40 mm, if a suitable real reference region exists there. At the unchanged target, the weights become minus 0.25, plus 0.5 and plus 0.75. With the same independent 2 micrometer input uncertainty, the calculated reference-height uncertainty becomes approximately 1.87 micrometers.
The target is still just outside this right triangle because X divided by L plus Y divided by B is 1.25. Nevertheless, the extrapolation leverage is much smaller. This comparison shows why checking only whether a target is outside the triangle is insufficient: the actual weights quantify how far the construction is being extended. Relocating a reference must also preserve its intended material surface and accessibility.
5. Keep common motion distinct from point-to-point noise
Suppose all three reference heights receive the same unknown offset c. Because the weights sum to one, the predicted reference height receives c, not c multiplied by the large sum of absolute weights. It would be incorrect to treat one common offset as three independent random errors. Conversely, local height errors that differ between references can be strongly amplified by a narrow layout.
If the target height is measured with the same common offset, subtracting the reference can cancel that component under a valid simultaneous or stable-common-mode model. A target acquired later after instrument drift may not share it. Preserve acquisition order and covariance assumptions. The full uncertainty of target height minus reference also includes target uncertainty and its covariance with the reference, which the reference-only numerical example deliberately does not estimate.
6. Do not interpret a perfect three-point fit as surface flatness
A plane through exactly three non-collinear points has no residual at those points by construction. That perfect fit cannot reveal a local high spot elsewhere, ceramic bow between references or damage at an unsampled edge. Acquire independent check locations when the surface form could affect the decision; use them to test the construction rather than forcing every check point into the same zero-residual fit.
A change at a reference patch may be physical rather than instrumental. Contamination, a chipped support region or altered seating can tilt the constructed plane. Inspect the patch and support state before changing printed artwork or classifying a remote pad as too high. More repeated readings at the same narrow triangle can reduce random noise but cannot establish that the unsampled ceramic surface has the assumed form.
7. Select the response from the observed sensitivity
Use the plane calculation to choose a better measurement design. It is a diagnostic for the reference construction, not a license to move datums after seeing the result.
| Observation | Interpretation | Targeted response |
|---|---|---|
| Large positive and negative target weights | Extrapolation amplifies differential reference errors | Review spatial span or add an appropriate independently validated reference method |
| All reference heights move together | Common offset may dominate | Check target timing and covariance before combining uncertainties |
| Remote targets shift after one reference is cleaned | The reference patch changed the plane | Document surface state and repeat the affected comparison |
| Three-point residual is always zero | No independent form check exists | Measure additional locations outside the construction set |
| Drawing datum differs from chosen plane | The measurement definition has changed | Restore the required datum-association method |
8. Preserve the construction with the inspection result
Deliver the three coordinates and heights, target coordinates, weights, plane equation and uncertainty assumptions. Include the actual reference and target locations on an image or drawing so another engineer can reproduce the calculation. Record whether the specimen was free, supported or clamped and whether the reference surface changed during preparation.
For subsequent comparisons, maintain the same intended reference definition and review any necessary change before applying it to production decisions. Moving the inspected feature while keeping the same three references can alter uncertainty substantially. Evaluating every critical target against the existing triangle prevents a stable-looking setup from becoming an unrecognized source of error when a circuit layout grows or an inspection region moves.
Review a ceramic reference-plane construction
Provide the reference triangle and target locations to assess spatial leverage and measurement uncertainty.
- Drawing or coordinate map identifying three physical reference patches and every evaluated target.
- Reference and target height data, acquisition order, support condition and repeat measurements.
- Intended plane definition and any governing datum-association requirement.
- Independent surface checks, point uncertainties, common drift assumptions and decision limits.
The drawing-upload form loads as you reach this section.