Underfill for Chip Scale Packages: When and How

Chip-scale assembly has blurred a boundary that used to be neat. Flip chip and chip scale packages sit directly on the board, and the traditional separation between the semiconductor die, its package and the assembly process has largely disappeared. The advantage in density is real, but smaller features make the die, the joints and the package more sensitive to mechanical and thermal stress — and that is what underfill is for.

Why Underfill Exists

The original reason for dispensing material beneath a die is the mismatch in thermal expansion between silicon and the substrate it is mounted on.

Silicon expands at roughly 2.4 parts per million per degree; a typical board material expands at around 16. A ceramic substrate can be designed to match, but conventional alumina ceramic sits at about 6.3, so the mismatch persists even there.

In a conventionally packaged device, the leads absorb that differential movement. In direct attachment, such as a ball grid array, the solder joints themselves are the weakest structural element, and therefore the first thing to fail under stress. They are also the most critical, because a failure at any single joint destroys the function of the circuit.

Filling the gap between the die and the substrate does not eliminate the expansion difference. It distributes the stress across the whole die area by bonding to the die, the balls and the substrate together, which is a far better arrangement than concentrating it in the joints.

There is a second benefit: the fill also keeps moisture and other contamination away from the joint area. The costs are equally clear. Underfill adds a process step and makes rework far more difficult, which is why many manufacturers run a quick functional test after reflow and before dispensing, so that boards which fail early never receive an irreversible process.

Deciding Whether to Underfill

There are more than fifty distinct chip scale package designs in use, along with countless combinations of operating conditions, so no single rule decides when underfill is warranted. What the designer can do is assess the factors that drive stress.

CTE mismatch. The larger the difference in thermal expansion between die and substrate, the more the fill helps. Underfill is more necessary on board-based packages, although reliability improvements have also been demonstrated on ceramic substrates.

Die size. Larger dies concentrate more stress. One study found that increasing die size from 6.4 to 9.5 millimetres reduced the number of thermal cycles the joints survived, between minus forty and one hundred and twenty-five degrees Celsius, from about 1,500 to about 900.

Ball size and layout. Bigger joints tolerate stress better. A solder ball of 300 micrometres, typical of a chip scale package, is far stronger than the 75 micrometre joint used in flip chip. Assuming comparable relative shear, the stress seen by the CSP joint is roughly a quarter of that in the flip chip case. Consequently, some CSP designers regard the ball structure alone as sufficient — though later work shows underfill still brings a significant reliability advantage, particularly in portable products. Increasing pad size at the die corners can raise stress tolerance, but it is not always practical and not always enough.

Board thickness. Thicker boards are stiffer and resist bending better. One analysis found that increasing substrate thickness from 0.6 to 1.6 millimetres raised cycles-to-failure from about 600 to about 900. The trade is uneven: doubling substrate thickness roughly doubles the reliability gain, whereas doubling die size degrades it by roughly a factor of four. On ultra-small devices, increasing thickness is often not an option at all.

Use environment. This is usually the decisive factor. Portable products are commonly specified for a thousand cycles between minus forty and one hundred and twenty-five degrees, followed by twenty to thirty drops from a metre onto concrete, with the device still fully functional. Thermal cycling studies show underfill can increase cycle count fourfold, with some filled assemblies surviving more than 2,000 cycles. Set against the cost of field failures — returns, reputation, warranty — many manufacturers now treat underfill as reliability insurance.

The Dispensing Challenges

Once the decision is made, the process has to be made repeatable. The difficulties are consistent across applications: obtaining a complete, void-free flow of material beneath the die; dispensing around tightly packed chips; avoiding contamination of other components; dispensing through openings in an RF shield or cover; and controlling flux residue.

Getting Complete, Void-Free Flow

The material reaches the area under the die by capillary action, so the first requirement is to position the needle close enough for the flow to start. A recommended starting point is an X-Y offset from the die edge equal to half the needle’s outside diameter plus about seven thousandths of an inch, with the needle at eighty percent of the chip height above the substrate. Throughout dispensing, the position has to be controlled precisely enough to sustain flow without touching or contaminating the back of the die.

Dispensing along several edges at once improves throughput, but the direction of flow matters. Wave fronts arriving from opposite sides that meet at an acute angle can trap voids, so the dispense pattern should be designed so that fronts merge at obtuse angles.

Neighbouring components change the flow more than most designers expect. Where two dies can share a dispensing path, that is workable. A passive component sitting parallel to a die edge acts as a barrier; one placed at ninety degrees to the edge can draw material away from the die that needs filling. Cross capillary action from a neighbouring die or passive can pull fill out from beneath the intended target and leave voids under it.

Needle selection is a trade-off. Gauges of 21 or 22 are a good general choice for underfill; smaller needles resist the flow and slow dispensing, but they may be required to reduce fillet size or to keep material away from other components. Multi-head systems can be used with a higher-viscosity material that does not flow, dispensed as a dam around adjacent components before the main fill, so that the dam prevents unwanted capillary flow under them.

Dispensing Through Openings

Underfill is increasingly used in radio frequency assemblies, which raises the question of how to fill parts after the shield has been fitted. For throughput, the shield is usually placed with the other components and soldered in a single reflow pass, so product and process designers have to agree on openings in the shield for the fill.

Two constraints follow. The die must not sit too close to the shield, because capillary action or a fast dispense can carry material into the shield or over the top of the die. And where the gap between component and cover is small, the dispense rate has to be slowed to prevent material riding over the component — which slows the assembly and limits throughput. Moving to another opening and returning later introduces positional variation, and the extra machine movements cost throughput again.

Timing the Flow

The flow time is governed by viscosity, the distance the material must travel, the gap or ball height, the contact angle and the surface tension at the liquid-vapour interface. Because viscosity and surface tension are temperature dependent, the values used in any calculation have to be those at the dispensing temperature — typically around ninety degrees Celsius.

In practice, most pumps and valves deliver material to the die faster than it can flow underneath. What has to be determined is the volume or weight of fill required per die; once known, a first estimate of flow rate decides whether the full quantity can be dispensed in one shot or must be split. A typical arrangement dispenses at one component while material flows under the previous one, then returns to finish. If a die needs 20 milligrams of material in two cycles, the system must be able to dispense 10 milligrams accurately.

Controlling Flux Residue

Excess flux residue has a direct negative effect: the fill adheres to the residue rather than to the balls, die and substrate, producing voids, trailing and other discontinuities. Cleaning beneath the die before dispensing has been shown to improve thermal cycling performance by as much as five times, but adding that step runs against the direction of current assembly practice and reduces throughput.

A more practical alternative is better control of the flux applied in the first place, and of the reflow conditions that follow it, as described in this discussion of reflow quality and assembly stability. Selective flux jetting, with the quantity at each site controlled in software, allows the exact flux thickness required for each component type — particularly useful in mixed designs where flip chip and chip scale devices with different ball diameters, 75 and 300 micrometres, sit on the same board.

Precision Makes the Difference

Accurate, repeatable dispensing is what separates a good process from a marginal one, especially where repeated shots of around 10 milligrams are required. That calls for a pump whose output does not vary with viscosity or needle diameter: a positive-displacement pump, which meters an exact volume with each stroke regardless of shot size, provides that. A closed-loop feedback system using accurate specific gravity measurement adds real-time control of the dispensed volume, and a high-precision programmable motion system allows different dispense patterns without sacrificing throughput.

None of this is a single-vendor problem. Getting chip-scale underfill to work reliably depends on a partnership between the product designer, the process engineer, the material formulator and the dispensing equipment supplier — and on the kind of process verification described in this overview of PCB inspection after fabrication. The joints being protected are themselves produced by a process with its own constraints, as covered in this look at BGA assembly.

FAQ

Does underfill stop the die and board from expanding differently? No. It redistributes the resulting stress across the die area instead of concentrating it in the solder joints, which is what turns an early failure into an acceptable service life.

Why can underfill make a board unreworkable? Because the material is designed to bond permanently to the die, the joints and the substrate. Once cured, removing a component without damaging the board or the neighbouring joints is impractical, which is why functional testing happens before dispensing.

What causes voids under a die? Two common causes: wave fronts meeting at too acute an angle, and cross capillary flow from neighbouring components pulling material away mid-fill. Needle placement and dispense pattern address the first, and dams or spacing address the second.