Underfill Dispensing for Flip Chip and CSP Packages

A flip chip joint is a very short column of solder between a silicon die and a substrate, and the two materials expand at very different rates. Every thermal cycle loads that column in shear, and after enough cycles a crack starts at the corner where the stress is highest. Underfill exists to distribute that load across the whole interface instead of concentrating it at the joints, and applying it correctly is a process discipline in its own right.

What Underfill Does

Underfill is a filled epoxy that is drawn into the gap between a die and its substrate by capillary action, then cured into a rigid but compliant layer. It mechanically couples the die, the joints and the substrate, so that thermal expansion differences are shared across a large bonded area rather than applied to individual solder bumps.

The reliability gain is substantial. Properly underfilled assemblies routinely survive an order of magnitude more thermal cycles than identical assemblies without it, and the improvement is largest for the largest die. The cost is an extra process step, a cure schedule and a set of new defect modes that have to be controlled.

Why Flip Chip and CSP Need It

A flip chip places the active surface face down and connects through bumps directly to the substrate. Because there is no compliant lead to absorb movement, the joint carries the full differential expansion between silicon and laminate. Chip scale packages face the same issue on a smaller scale, and the trend toward thinner packages with finer pitch has made the problem more acute rather than less.

Ball grid array packages with large bodies benefit as well, though the mechanism is slightly different because the joint is taller. The decision to underfill is normally driven by the expected thermal cycle count, the delta between maximum and minimum temperature, and the size of the die. Where those three are modest, underfill may be unnecessary; where any is large, it becomes essential.

Capillary Flow Mechanics

Capillary flow depends on the balance between the surface energy driving the liquid into the gap and the viscous resistance opposing it. A narrow gap and a low-viscosity material favour fast filling, while a wide gap and a heavily filled material slow it down. The filler content that improves the cured properties also raises the viscosity, which is the central trade-off in material selection.

Flow also depends on the surfaces. Contamination, oxidation or flux residue in the gap changes the wetting behaviour and can stop the front before it reaches the far edge. That is why cleaning before underfill is not optional: a perfectly good material will fail to fill a contaminated gap, and the incomplete fill will be invisible from above.

Dispensing Patterns and Volume Control

The dispense pattern determines whether the material fills evenly. A single line along one edge or an L-shaped pattern along two edges is common; the shape is chosen so the front arrives at all corners at roughly the same time. If one path fills much faster than another, it can trap air in the slower region.

Volume control is equally important. Too little material leaves an incomplete fill and a weak fillet, while too much produces a large external fillet that can interfere with neighbouring components or wick onto pads. The correct volume is established by experiment and then maintained by checking the fillet dimensions on samples, because the dispensed weight alone does not prove the gap was filled.

Materials: Filler Content and CTE

Underfill materials are loaded with silica filler, typically at forty to seventy percent by weight. The filler lowers the coefficient of thermal expansion of the cured epoxy so it sits between that of silicon and the substrate, and it raises the modulus and the thermal conductivity. More filler improves the mechanical match but makes the material harder to flow.

Particle size matters because the filler must pass through the gap. A material with large particles will not fill a narrow standoff, and even when it flows, the particles can filter out and create a region of unfilled epoxy with different properties. Matching the maximum particle size to the gap height is a first-order material selection rule.

Cure Schedules and Process Windows

Cure is usually a two-stage affair: a short period at a moderate temperature to allow the material to flow and wet fully, followed by a higher temperature to complete the cross-linking. Starting the cure too early traps the material before it has reached the far edge, producing an incomplete fill that looks like a volume problem.

The cure window also has to respect the rest of the assembly. Solder joints, previously cured materials and the substrate all have thermal limits, and an aggressive high temperature cure can degrade any of them. Selecting a material with a lower cure temperature often costs little in final properties and removes a whole class of risk. Interactions with the wider thermal design are discussed in this guide to PCB thermal management.

Voids, Fillets and Common Defects

Voids are the most common defect and the hardest to see. They form where the flow front traps air, where outgassing occurs during cure, or where the dispense pattern leaves a pocket. A void directly under a bump concentrates stress rather than distributing it, so a small void in the wrong place can be worse than a uniformly thin layer.

Other defects include incomplete fill at a corner, filler separation near the edge, and an oversized fillet that creeps onto adjacent pads. Because the material is opaque, inspection relies on scanning acoustic microscopy for internal voids and on measured fillet dimensions for the external geometry. The techniques overlap with those used for hidden joint inspection, which are compared in this guide to X-ray and optical inspection.

Thermal Cycling Performance

The benefit of underfill is measured in cycles to failure under a defined thermal cycling profile. A typical qualification runs between minus forty and one hundred and twenty-five degrees Celsius and records the number of cycles until electrical continuity fails. Underfilled samples usually fail in the die or the substrate rather than at the joints, which is a sign that the joint is no longer the weak link.

The distribution of failures matters as much as the median. A wide spread suggests process variability in the underfill, particularly voids or incomplete fill, rather than a material limitation. That is why cycling results should always be reported alongside an acoustic inspection of the samples, so the failure population can be explained.

Inspection, Rework and Repair Limits

Once cured, underfill is effectively permanent. Removing a die means grinding or milling the material away, which destroys the package and risks the board. In practice a defective underfilled assembly is scrapped rather than repaired, and that economic fact is the strongest argument for getting the process right the first time.

Because rework is not an option, the inspection plan has to be conservative. Acoustic scanning of a sample from each lot, dimensional checks on the fillet, and process data such as dispense weight and cure temperature profile together provide the confidence that a repair cannot. Component-level reliability practice described in this overview of component tolerance and reliability is a useful reference for setting the sampling plan.

FAQ

Does every flip chip need underfill? No. Small die with modest thermal excursions can survive without it, and many commodity packages omit it entirely. Underfill becomes necessary as die size grows, as the temperature swing widens and as the expected cycle count rises. The decision should follow a qualification test rather than a general rule.

Why does underfill sometimes fail to fill completely? The usual causes are a contaminated gap, a material with particles too large for the standoff, insufficient dispensed volume, or a cure that starts before the flow front has travelled the full distance. Which of these applies is normally determined by inspecting the fill pattern and the material lot together.

Can a badly underfilled package be repaired? Practically speaking, no. Cured underfill cannot be selectively removed without destroying the die and often the substrate as well. The standard disposition is to scrap the assembly. That is why process validation, dispense weight monitoring and acoustic sampling matter more here than in processes where rework is available.