| Hypothetical pattern | Total projected void area | Total fraction | Largest feature fraction |
|---|---|---|---|
| One central feature | 2.5 mm² of 25 mm² | 10% | 10% |
| Ten separate equal features | 2.5 mm² of 25 mm² | 10% | 1% |
A percentage needs a joint definition
There is no useful universal void percentage for every power die attached to a ceramic PCB. The relevant joint may be die to copper, substrate to baseplate or a package thermal pad, and those joints have different geometries and responsibilities. Begin by naming the exact interface, materials and operating conditions. The die-attachment guide provides construction context. An acceptance agreement should say what is measured, how it is measured and why the chosen limit supports the assembly's thermal, electrical and mechanical requirements.
Define the denominator before comparing reports
A report stating ten percent voiding is incomplete unless it identifies the reference area. It might divide by the intended die footprint, a visible solder outline or only a segmented wettable region. These definitions can produce different values from the same image. Specify the region of interest, treatment of fillets and exclusions, and how overlapping features are handled. Preserve that definition in the inspection program revision. Compare results only after confirming equivalent denominators. Otherwise, an apparent process improvement can be caused by a changed analysis boundary rather than a better joint.
Worked example: the same total can describe different patterns
Consider a hypothetical 5 by 5 mm attachment region with area 25 mm². A total projected void area of 2.5 mm² gives 10% by area. One joint might contain a single 2.5 mm² central feature; another might contain ten separate 0.25 mm² features. Both give the same total percentage, but the largest individual feature occupies 10% of the joint in the first case and 1% in the second. Their location relative to local heat generation and stress paths may differ as well. This example does not establish which joint passes; it explains why total area alone cannot describe the defect population.
Record size and location as separate descriptors
Useful descriptors can include total projected area, largest connected feature, location relative to the die footprint, and whether a feature intersects a critical edge or region. Define them only where the inspection method can support repeatable measurement. Infineon's general board-assembly guidance notes that void effects depend on the system and on void size and location. That is a reason to investigate the relevant geometry, not to copy a packaged-device limit into a bare-die joint. Keep the assembly process review linked to the defect definition so that process changes and inspection changes are reviewed together.
Distinguish what an X-ray projection can establish
A two-dimensional projection combines attenuation along the viewing direction. Features from different layers or overlapping structures can complicate segmentation. It does not automatically establish the depth, interface location or connectivity of every feature in the joint. Select imaging conditions, magnification and analysis settings appropriate to the construction, and use complementary investigation when the failure question requires information the projection cannot provide. Do not report a projected area as a three-dimensional void-volume fraction without a method that supports that conversion. Keep representative raw images with the analysis result.
Use a simplified thermal calculation only as a screening exercise
Suppose an ideal uniform joint has resistance R₀ = t/(kA). If one deliberately assumes that voids simply remove an area fraction f while the remaining paths retain the same thickness, conductivity and boundary temperatures, the toy model gives R/R₀ = 1/(1 − f). At f = 0.10, the ratio is 1.111; at f = 0.20 it is 1.25. This model omits lateral spreading, feature location, local heat flux, gas conduction and interfacial conditions. It cannot establish a permitted void percentage. Its value is showing the assumptions behind a common area-only estimate, not replacing thermal validation.
Do not transfer a package study without its boundaries
Onsemi's AN-9077 contains void-related thermal evaluation for its specific SPM 7 construction and board configurations. Such a study can demonstrate that a result belongs to a complete assembly and test condition. It does not establish acceptance for every ceramic-mounted power die. Before using any published experiment, compare the joint under study, heat-source distribution, material stack, mounting boundary and failure metric with your product. The thermal-management guide helps define that path. If important conditions differ, treat the paper as context and develop product-specific evidence.
Connect inspection to reliability evidence
Initial thermal behavior is only one consideration. Define whether the joint must also maintain electrical conduction, mechanical integrity or another function through service exposure. Select relevant qualification conditions and failure criteria for the actual product, keeping application requirements separate from a generic manufacturing preference. Compare the initial inspection record with post-test findings and with functional performance. The thermal-cycling guide explains why a cycling program needs a defined profile and failure criterion. Do not invent a cycle count or a confidence statement merely because an inspection threshold looks familiar.
Control the process and the measurement system together
When investigating excess voiding, document attachment material, surface preparation, deposition pattern, joint thickness and thermal process conditions. Infineon identifies several process contributors, including flux chemistry and reflow-related conditions; their relevance depends on the joint being evaluated. Change variables deliberately and keep inspection settings fixed when comparing trials. Check repeatability by reanalyzing representative images and, where practical, comparing inspectors or systems. A process distribution near the acceptance threshold is hard to manage if measurement variation has not been characterized. Retain both accepted and nonconforming examples with their disposition.
Write an acceptance clause that is auditable
The final agreement should identify the joint and drawing revision, inspection method, reference area, required descriptors, limits, sampling approach and disposition process. Link those limits to the component, customer or qualification evidence that makes them appropriate. State what happens when an image cannot be interpreted reliably and who approves an exception. Include the clause in the DFM checklist and supplier review. This produces a defensible acceptance decision; a standalone percentage copied from another package or an internet discussion does not.
Create a measurement definition before collecting a large dataset
A void database is useful only when the entries represent the same measured quantity. Record the image acquisition settings, region of interest, segmentation method and treatment of ambiguous features. Identify whether the result is measured automatically, edited by an inspector or derived from a destructive investigation. If the inspection program changes, retain the earlier program version with the affected data instead of comparing its output as though nothing changed. Keep the component and joint identifiers attached to each image. A filename containing only the production date can become ambiguous when multiple interfaces or reworked assemblies are inspected. Establishing this information first makes later process comparisons substantially more defensible than trying to reconstruct measurement conditions after a failure.
Test how segmentation changes the reported percentage
Consider an illustrative image with a reference joint area of twenty-five square millimeters. One analysis setting classifies 2.25 square millimeters as void, while another classifies 2.75. The reported fraction changes from nine to eleven percent even though the physical joint has not changed. This example does not prescribe an acceptable variation; it shows why an acceptance threshold close to a measured result needs a repeatable analysis method. Examine representative images containing small features, overlapping structures and uncertain boundaries when evaluating the method. Keep the raw image so that a later reviewer can distinguish a physical process change from a change in image processing. A tightly controlled numerical limit does not compensate for an unstable measurement definition.
Separate total voiding from an exclusion-zone requirement
Some assemblies may have a particular region whose integrity matters more than the average joint area. If engineering evidence justifies an exclusion zone, define its coordinates relative to the joint and describe the feature that is prohibited there. A total percentage requirement and a zone requirement then become separate acceptance items. Do not create a central exclusion zone merely because a drawing looks symmetric; the relevant heat generation or mechanical loading may be elsewhere. Similarly, a feature touching an edge should be described by the agreed inspection criteria rather than automatically treated as equivalent to an enclosed feature of the same projected area. The purpose is to express a functional concern in a measurable way, with a reason for the selected location.
Distinguish initial appearance from growth during exposure
An initial inspection establishes the joint's condition at that point in the process. It does not show whether a feature will grow, connect to another feature or interact with a different damage mechanism during service exposure. When qualification includes repeated inspection, register the images consistently so that the same locations can be compared. Record any changes in acquisition settings and consider whether they could explain an apparent difference. Pair the image comparison with the relevant functional measurement rather than treating every visible change as a quantified loss of performance. Conversely, an unchanged projection does not prove the absence of damage that the method cannot resolve. Choose the investigation method to match the failure question being asked.
Design a process experiment with interpretable comparisons
If the manufacturing team is changing deposition pattern or thermal processing to reduce voiding, define the trial variables and keep other influential conditions traceable. Changing material, surface preparation, joint thickness and inspection settings simultaneously can produce a better-looking result without showing which change mattered. A staged experiment can begin with a small number of controlled alternatives, then repeat the promising condition to examine reproducibility. Include functional outcomes such as the relevant thermal measurement and inspectability of the joint. Select trial quantities and statistical treatment appropriate to the decision; this guide does not provide a universal sample count. Document unsuccessful conditions as well as successful ones so that the same unproductive trial is not repeated after a personnel or supplier change.
Keep rework history in the acceptance decision
A reworked joint may have experienced an additional thermal cycle, different cleaning or a changed attachment process. Record those differences and identify the approved rework procedure before applying the original acceptance criteria. A lower void percentage after rework does not by itself demonstrate that every other requirement is still met. Inspect the features affected by the actual rework operation and retain the previous images when they help establish traceability. If the component or assembly documentation limits rework, follow the applicable controlled requirement rather than inferring permission from an acceptable final image. The acceptance record should make clear whether the assembly is as originally produced or has been repaired under an approved disposition.
Write a disposition that identifies the evidence
For a joint near or outside a specified limit, record the exact criterion, measured result and measurement method before discussing an exception. Identify whether the issue is a confirmed nonconformity, an ambiguous image or a measurement-system disagreement. An engineering concession should state which parts it covers, the supporting analysis or test, and whether it applies only to that lot. It should not silently change the standing process specification. When additional testing is used, explain how that test addresses the consequence of the observed feature. A room-temperature continuity result may not answer a thermal or fatigue question. The disposition should connect the detected condition to the requirement it could affect, rather than relying on a generic statement that the part appears usable.
Make supplier reporting concise but auditable
A practical inspection report can provide the drawing and program revisions, joint identification, total projected void fraction, largest-feature result where required, zone findings and links to retained images. State whether every joint was inspected or a defined sampling method was used. Report uninterpretable images separately from passing joints so that missing information is not hidden inside a favorable average. Include the disposition of nonconforming parts and the approval reference for any concession. This format gives the receiving team enough information to check the agreement without requiring a large unstructured image archive in every shipment. The underlying records should still be retrievable for process investigation and qualification comparisons over the agreed retention period.
Engineering example
The two patterns have equal total projected area but different feature distributions. Neither is assigned pass or fail. The separate area-only thermal model gives R/R₀ = 1/(1 − f), under restrictive assumptions that do not establish acceptance limits.
Before you release the design
- Identify the exact attachment interface and reference area.
- Specify repeatable size and location descriptors.
- Separate projected area from volume and interface-depth claims.
- Link acceptance to actual thermal and reliability evidence.
- Keep inspection settings and process revisions traceable.
Sources and further technical reading
Manufacturer references support the material and process context. Worked examples and checklists are engineering guidance; they are not test results or supplier guarantees.