Summary

A capillary underfill can meet its DSC and TMA limits and still crack after thermal cycling because those tests describe the average thermal behavior of a prepared specimen. They do not reproduce every local condition beneath a flip-chip die.

The failure may begin at a die corner, beside an elongated void, within a filler-rich region, or along an interface weakened by contamination, moisture, or nonuniform cure. Raising Tg, lowering average CTE, or immediately changing the material may leave the controlling mechanism unchanged.

The investigation should begin by locating the earliest damage and confirming the crack path. Thermal-property data should then be reviewed against that evidence.

This article focuses on post-dispense capillary underfill used between a flip-chip die and substrate. No-flow, pre-applied, molded, and film underfill systems require different process checks.

What the Failure Looks Like

The incoming material passes viscosity, DSC, and TMA checks. Dispensing appears stable. Capillary filling is completed, the package is cured, and no obvious defect is visible.

Cracking or delamination appears only after thermal cycling.

The first useful question is not whether the package failed. It is where the earliest damage appeared and what structure it followed.

A void beside a crack is not automatically the crack origin. An interface that appears delaminated after the final cycle may have failed first, or it may have opened after a cohesive crack redistributed local stress.

Final-cycle inspection alone cannot always distinguish these sequences. Uncycled packages and samples removed at selected cycle intervals provide stronger evidence.

Why DSC and TMA Can Pass Without Predicting Package Reliability

DSC Does Not Reproduce the Complete In-Package Cure History

A DSC result depends on specimen preparation, cure history, heating program, sample mass, and the method used to assign Tg.

Material confined beneath a die experiences a different condition:

• heat transfer may vary across the die and fillet;
• filler concentration may change along the flow path;
• cure may begin before capillary filling is complete;
• contacted surfaces may affect local wetting or interfacial cure;
• production may include waiting periods, reflow, post-cure, or moisture exposure not represented by the DSC specimen.

A passing Tg confirms that the tested specimen met the selected method and acceptance rule. It does not prove that equivalent conversion was achieved at the dispense entrance, die center, far-flow edge, and corners.

The review should include:

• the cure profile used to prepare the specimen;
• first-heat and second-heat behavior, where relevant;
• residual exotherm after the production cure profile;
• cure onset and peak shape;
• heating rate and sample mass;
• time between dispensing and cure;
• reflow or post-cure history.

A small Tg increase should not be treated as corrective evidence unless the crack path or thermal-cycling result also changes.

TMA Measures a Prepared Specimen, Not Every Local Stress Field

ASTM E831-25 covers the determination of the technical coefficient of linear thermal expansion of solid materials by thermomechanical analysis. The result can support thermal-stress evaluation and specification control, but it remains a specimen-level measurement under defined conditions.

The interpretation should include:

• the temperature ranges used for CTE below and above Tg;
• specimen orientation;
• heating rate;
• probe condition and load;
• thermal history;
• transition width;
• possible softening or probe indentation.

Package cracking also depends on variables that average CTE cannot describe:

• modulus below and above Tg;
• die dimensions and distance from the neutral point;
• underfill thickness and fillet geometry;
• cure shrinkage and residual stress;
• filler-rich and resin-rich regions;
• void position and shape;
• adhesion to each contacted surface;
• fracture resistance and cyclic fatigue;
• absorbed moisture;
• prior reflow and conditioning.

NASA’s technical overview of flip-chip underfills also treats CTE, Tg, modulus, moisture absorption, cure shrinkage, filler behavior, voiding, and interface performance as connected design factors rather than independent pass/fail properties.

Two materials can produce similar TMA curves and still develop different crack paths.

Reducing CTE is not automatically beneficial. Higher filler loading may reduce bulk expansion while increasing viscosity, stiffness, filtration risk, agglomeration, or local resin starvation. The target should be selected from the observed failure mechanism, not from the lowest available CTE value.

Why D50 Alone Is Incomplete for Filler Approval

An average particle size is useful for general comparison, but it may hide the particles most likely to disturb narrow-gap filling.

A meaningful review should consider:

• D10, D50, and D90;
• maximum or oversized-particle control;
• agglomerate frequency;
• particle shape;
• fine-particle fraction;
• multimodal distribution;
• filler surface treatment;
• total loading;
• stability during storage and dispensing.

The upper tail may matter more than the average when the package gap approaches the size of coarse particles or agglomerates. A limited number of oversized features may interrupt flow, contact local structures, or create adjacent filler-rich and resin-rich regions.

Very fine filler is not automatically safer. Smaller particles increase total surface area, which can raise viscosity and make wetting, dispersion, and storage stability more sensitive.

A particle-size report and an in-package cross-section answer different questions. The first describes the tested sample preparation. The second shows whether the distribution remained uniform after storage, dispensing, capillary flow, and cure.

Useful sectioning locations include:

• the dispense entrance;
• the die center;
• the far-flow edge;
• die corners;
• regions beside voids;
• the fillet.

The review should look for agglomerates, filtration along the flow direction, local packing changes, resin-rich areas, and filler–matrix separation.

Specifications for polymers, coupling agents, fillers, and other functional components should be connected to the intended semiconductor and polymer material system, not approved as unrelated inputs.

Why Void Location Can Matter More Than Total Void Percentage

A single total-void value can hide the most relevant risk.

A small void near a die corner, bump, or weak interface may be more significant than a larger isolated void in a lower-stress region. Elongated, connected, and interface-aligned voids may create different crack paths from rounded internal voids.

Potential formation routes include:

• air introduced during handling or dispensing;
• insufficient wetting;
• moisture or volatile release during cure;
• flow-front convergence;
• unsuitable substrate preheat;
• premature viscosity increase;
• an ineffective dispense pattern;
• contamination or flux residue;
• excessive delay before cure;
• filler filtration that changes local rheology.

The first question is whether the void existed before thermal cycling.

Where the inspection method permits, compare the package:

1. after dispensing;
2. after cure;
3. after moisture conditioning;
4. after early cycle intervals;
5. after final failure.

A process change that lowers total void area but leaves the same crack origin may not have removed the controlling mechanism.

Interface Adhesion Must Be Evaluated After Aging

Capillary underfill may contact die passivation, solder mask, copper features, solder, organic substrate materials, and residues from earlier assembly steps.

Each surface presents a different adhesion condition.

Possible contributors to interfacial failure include:

• incomplete flux removal;
• variable solder-mask surface condition;
• plasma-treatment variation;
• delay between treatment and dispensing;
• absorbed moisture;
• organic or particulate contamination;
• insufficient wetting at the actual dispense temperature;
• coupling-agent incompatibility;
• interfacial cure inhibition;
• excessive cure shrinkage.

An adhesion result from an ideal laboratory coupon cannot represent every package interface.

A useful comparison should include:

• as-cured and conditioned specimens;
• actual die-passivation and substrate surfaces;
• the production cleaning sequence;
• realistic treatment-to-dispense delays;
• relevant moisture exposure;
• production cure and reflow history;
• failure location after aging.

Initial room-temperature adhesion may pass while the same interface loses resistance after moisture exposure or repeated thermal cycling.

Separate Material Variation From Process Compensation

Several variables are often changed together after cracking appears. A new formulation, higher preheat, longer cure, additional plasma treatment, and different moisture conditioning may produce a passing build, but the experiment cannot show which change mattered.

The factor groups should remain separate until the evidence connects them.

A material may meet viscosity, Tg, and CTE limits while its oversized-particle tail has shifted.

A process may appear stable at one optimized preheat condition while offering little tolerance around that setting.

A replacement formulation may pass because its flow behavior temporarily compensates for an uncontrolled surface condition. That result does not prove that the original cause has been corrected.

Recommended Investigation Sequence

1. Freeze the Failed Configuration

Document the exact:

• material and substrate lots;
• die dimensions and minimum gap;
• flow distance and bump arrangement;
• contacted surfaces;
• flux and cleaning method;
• storage and handling history;
• dispense program and substrate preheat;
• wait before cure;
• cure and reflow profiles;
• moisture conditioning;
• thermal-cycle conditions.

Die size, gap, flow distance, and interconnect layout matter because they determine whether a particle, void, or weak interface sits in a mechanically critical region.

Changing the material and process before documenting the failed condition destroys the reference needed for comparison.

2. Locate the Earliest Detectable Damage

Inspect uncycled packages and selected cycle intervals where practical.

“Crack origin” should refer to the earliest confirmed change, not merely the most visible feature in the final cross-section. Relevant evidence may include the first C-SAM indication, the first interface separation, or the earliest crack identified by staged destructive analysis.

Determine whether the first measurable event is:

• a pre-existing void;
• corner delamination;
• separation along one interface;
• cohesive cracking;
• filler–matrix debonding;
• package warpage;
• solder, passivation, or substrate damage.

This evidence should determine the next test.

3. Confirm the Crack Path and Local Microstructure

Use C-SAM, X-ray, or micro-CT to locate suspect regions. Follow with cross-sectioning or fractography when the failure mode cannot be identified nondestructively.

Classify the path as:

• adhesive;
• cohesive;
• mixed-mode;
• connected to a void;
• associated with an agglomerate;
• driven by another package feature.

Map filler and void distribution around the earliest damage. Compare several package locations and, where relevant, the beginning, middle, and end of the material-use window.

The final failure image should not be the sole basis for assigning root cause when earlier-stage evidence is available.

4. Recheck Thermal Data Against the Production History

Confirm:

• the specimen cure history;
• whether reported values are typical results or release limits;
• how Tg was assigned;
• whether residual cure was evaluated;
• the temperature ranges used for CTE;
• specimen orientation;
• heating rate and probe condition;
• whether the tested sample represents production processing.

A numerical shift matters only when it supports a plausible mechanism and changes the package result.

5. Challenge One Supported Variable at a Time

If the crack follows an interface, challenge cleaning, plasma condition, treatment delay, moisture state, or surface lot.

If damage follows voids, challenge dispense pattern, preheat, flow time, handling, and cure timing.

If the crack crosses filler-rich regions, compare PSD tails, dispersion, storage history, and package location.

A focused study produces stronger evidence than a broad trial in which formulation, surface treatment, cure, and conditioning change simultaneously.

Why Crack-Origin Evidence Should Outrank a Better Datasheet Number

Material discussions often move quickly toward a higher Tg, lower CTE, smaller D50, or replacement grade with more attractive typical values.

Those changes are easy to compare. They are not always connected to the observed failure.

The overlooked question is whether the proposed parameter can explain the crack path.

A crack beginning beside an elongated flow void is unlikely to be corrected by a modest Tg increase when the void was created before cure.

An interface that weakens only after moisture conditioning may continue to fail after CTE reduction if the dominant problem is contamination or aged adhesion.

A cohesive crack beside agglomerates may persist after a longer cure because the local microstructure remains unchanged.

My judgment is that three rules should guide this investigation:

1. Rank voids by location, shape, and connection to an interface—not only by total void area.
2. Do not approve a narrow-gap formulation from D50 alone when the oversized tail and agglomerates can control local flow.
3. Do not approve a replacement as a drop-in alternative until it reproduces the package result within the intended production window.

For R&D teams, this means selecting the next experiment from the crack origin rather than from the easiest material property to adjust.

For quality teams, it means checking whether incoming controls cover variables that can change local structure, not only viscosity, Tg, and average CTE.

For production teams, it means demonstrating tolerance around preheat, dispensing, cure, and surface-treatment conditions.

For procurement teams, it means separating a promising development sample from a reproducible commercial material.

Sample, Pilot, and Bulk Approval Require Different Evidence

A successful development sample does not establish a production window or prove commercial-batch consistency.

Development Sample

The sample should use the intended gap, flow distance, surfaces, dispense route, cure history, moisture condition, and thermal-cycle profile.

A larger-gap or shorter-flow test vehicle may hide filler filtration and voiding risk.

Pilot Build

Pilot work should test a practical operating window rather than repeat one optimized run.

The study should include normal variation in:

• substrate preheat;
• dispense timing;
• cure ramp;
• surface-treatment delay;
• material handling.

A pilot should pause when movement within the expected production range changes the crack origin, void morphology, or interface-separation pattern.

Commercial Qualification and Bulk Procurement

Commercial approval should confirm that routine production represents the qualified sample.

Relevant evidence includes:

• independent material and substrate lots;
• beginning-, middle-, and end-of-use samples where applicable;
• agreed control of oversized particles and agglomerates;
• clear distinction between typical values and release limits;
• storage and handling boundaries;
• traceable production and lot history;
• notification requirements for critical raw-material, formulation, and process changes.

The qualification of resin intermediates, polymer modifiers, and other functional inputs may also require review of electronic and high-tech material specifications before the finished formulation moves into a larger build.

A larger order does not reduce unresolved technical risk. It exposes more packages to it.

Questions That Can Change the Approval Decision

The technical discussion should focus on evidence that affects package approval:

• Does the PSD represent the finished formulation or only the incoming filler?
• How are D90, maximum particle size, and agglomerates controlled?
• Can filler distribution change during storage or dispensing?
• Which cure profile was used for DSC, TMA, DMA, and adhesion testing?
• Are Tg and CTE typical values or lot-release limits?
• Are the CTE temperature ranges and specimen conditions reported?
• Was residual cure checked after the production profile?
• Which package surfaces were used for adhesion evaluation?
• Was adhesion retested after moisture or thermal conditioning?
• How was voiding evaluated at the intended gap and flow distance?
• Which formulation and production changes require notification?
• Does routine commercial material follow the same route as the qualified sample?

These questions do not assume that every crack is caused by the underfill. They establish whether the available evidence can separate material variation from package and process variation.

Conclusion

When capillary underfill cracks after thermal cycling, passing DSC and TMA results should narrow the investigation, not close it.

The strongest sequence is to locate the earliest damage, confirm the crack path, map filler and void distribution, challenge the relevant interface, reconstruct the actual cure and handling history, and change only variables supported by that evidence.

Tg and CTE remain important qualification parameters. They should be interpreted together with crack-origin evidence, aged adhesion, local microstructure, and production-window stability.

For projects involving resin intermediates, functional additives, specified material parameters, samples, or technical documentation for electronic-material evaluation, ChemicalCell can review the requirements through a technical material RFQ.