On this page
Permanent artwork compensation is justified only when it removes a repeatable dimensional bias without hiding an unstable process. A ceramic thick film circuit normally starts with an already sintered substrate. Its later print-and-fire sequence must not be assigned the large densification shrinkage of green ceramic tape. The engineering task is to measure what actually changes between defined process states, distinguish substrate movement from printed-pattern movement, and test whether a stable inverse transformation improves the final functional geometry.
Key design decisions
- Identify the substrate and layer state being measured; do not transfer LTCC or green-ceramic shrinkage factors to fired alumina.
- Compensate only a reproducible mapping demonstrated on independent panels, not a single panel's defects.
- Preserve an uncompensated design master and verify final dimensions against it, not against the altered manufacturing artwork.
Specify which dimensional change is under investigation
There are several different comparisons: incoming ceramic before and after a thermal pass, wet print versus fired print, and one fired layer versus a later fired layer. Each comparison has a different physical meaning. Printed paste can change shape and thickness during drying and firing without implying that the dense alumina panel has undergone an equivalent in-plane shrinkage.
Define the before and after condition, including measurement temperature, fixture support and whether the panel has been singulated. A warm panel measured against a room-temperature baseline includes thermal expansion. Out-of-plane bow can also change the apparent projected positions in an optical measurement. Neither effect should be silently converted into permanent artwork shrink compensation. Material-grade and temperature-dependent behavior must come from the selected substrate information.
Follow the same features on the same panels
Assign a durable identity to each panel and measure stable ceramic features together with selected printed features at each stage. Hole centers, edge-related datums and printed marks serve different purposes; retain their identities rather than mixing them into one unnamed point set. A change shared by ceramic and printed features deserves a different investigation from a change found only in the printed layer.
Distribute points across the usable panel and include regions containing the tightest overlap requirements. Compare several panels and more than one controlled run. If all early measurements come from one material lot and all later measurements from another, an apparent process effect may actually be a starting-population difference. Paired records prevent that particular ambiguity and make outliers traceable to physical parts.
Remove metrology and seating effects from the compensation signal
Confirm the measurement scale using a suitable calibrated target over the required field. Repeatedly measure one panel with and without reseating to establish how much of the observed change comes from imaging and support. The calibration plane, magnification and orientation must remain appropriate when comparing layers of different thickness or a panel with different bow.
Inspect the raw coordinate tables before averaging. A swapped mark, reversed axis or altered threshold can create a coherent-looking transformation that has no material cause. Keep rejected observations and their reason for exclusion. Removing inconvenient points until a desired scale factor appears makes the correction impossible to defend when another panel fails.
Fit the forward mapping before calculating its inverse
Represent the measured relationship from an earlier coordinate state to the final coordinate state as a forward mapping. Start with translation and rotation, then consider affine scale and shear only if the remaining spatial pattern supports them. An affine model cannot describe arbitrary local warpage, curved distortion or isolated print defects.
Use a stated coordinate origin and units. The matrix coefficients and translation terms together define the mapping, so copying a percentage without its origin or transformation direction is incomplete. More fitting points than the minimum are needed to evaluate residuals. Keep independent verification locations outside the fit so a more flexible model must demonstrate predictive improvement rather than merely a smaller fitting error.
q = Ap + t; pcomp = A⁻¹(qtarget − t)
- p: manufacturing coordinate before the measured transformation
- q: resulting final coordinate
- A: fitted invertible in-plane affine matrix
- t: fitted translation vector
- qtarget: required final coordinate
- pcomp: proposed compensated manufacturing coordinate
The measured forward relationship remains stable for the proposed artwork and process range. The inverse is invalid for a singular or poorly conditioned fit and does not correct unmodeled local distortion.
Choose the correction that matches the demonstrated behavior
The decision is not simply compensate or reject. Some observations justify a setup adjustment, some justify a tooling investigation, and others support a controlled artwork revision. Separate those choices before changing the design data.
| Measured behavior | Appropriate response | Evidence still needed |
|---|---|---|
| Stable offset with unchanged dimensions | Correct setup or origin conversion | New print verifies expected displacement |
| Repeatable affine pattern with small independent residuals | Trial inverse compensation on manufacturing artwork | Independent panels retain functional margins |
| Large panel-to-panel changes in fitted scale | Investigate process or metrology variation | Stable distributions before permanent correction |
| Localized distortion near one support or feature | Review local support, print condition or geometry | Cause-specific comparison, not global scaling |
| Different movement for different fired layers | Use layer-specific process-state comparisons | Verified layer-to-layer overlap after the full sequence |
Recalculate feature dimensions and overlap after transformation
A compensated coordinate map can move feature centers correctly while changing line widths, spacing or opening dimensions in an unwanted way. State whether the transformation applies to complete polygons, feature centers, a step-and-repeat origin grid or only a registration layer. Those operations are not interchangeable when the layout contains narrow conductor necks, resistor bodies or bond pads.
Check the transformed manufacturing data against minimum manufacturable geometry and then check the resulting fired features against the original functional drawing. For resistors, altered length-to-width ratio can change the electrical value. For crossovers, different transformations of conductor and dielectric layers can consume insulating overlap. The correction package therefore needs both positional verification and dimension-sensitive electrical or insulation checks.
Validate on panels not used to derive the correction
Use a controlled trial that includes the original and proposed manufacturing data under comparable process conditions. Keep the part temperature, locating method, print state and measurement procedure consistent. Compare the final coordinates with the same uncompensated functional target, and report both average behavior and the worst relevant feature relationships.
A correction should improve the intended spatial bias without increasing unexplained scatter or shifting the failure to another layer. Evaluate whether it remains useful across the intended substrate lots, panel positions and thermal sequence. If only one panel benefits, the model may be fitting that panel rather than the process. Preserve unsuccessful trials because they define where the compensation does not transfer.
Keep design intent separate from manufacturing compensation
Retain the customer design as the dimensional authority and store compensation as a controlled manufacturing transformation. Record the source drawing revision, applicable layers, matrix or scalar factors, origin, units, evidence set and approved process conditions. A regenerated screen must receive the correct transformation once; applying it twice can reverse the apparent benefit and create a larger error.
Reassess the transformation after a significant change in screen construction, substrate family, panel format, paste system or thermal sequence. These are triggers to check transferability, not proof that the factor must change. The deliverable should make it possible to reproduce the uncompensated target, the transformed artwork and the final verification without relying on a technician's remembered percentage.
Submit the dimensional comparison package
Provide paired evidence before proposing a permanent artwork scale or distortion correction.
- Substrate family, panel format and confirmation of fired or green starting state.
- Matching feature coordinates before and after the specified print-and-fire stages, with panel identities.
- Original functional artwork, manufacturing data, units, origins and layer sequence.
- Calibration and support method, measurement temperatures, observed residual maps and limiting feature margins.
The drawing-upload form loads as you reach this section.