Package On Package Stacking Design

A package on package assembly places one packaged device on top of another, with the two joined by a perimeter array of solder balls that carries the signals between them. The arrangement is used where a logic device and a memory device have to be as close as possible, and where the routing between them has to be short.

This article covers how the stack is built, how height and warpage interact, and what the process demands of both packages.

How The Stack Is Built

The bottom package is a normal ball grid array, with its balls on the underside connecting to the motherboard. On its top face it carries a second set of pads arranged around the perimeter, and the top package carries a matching set of balls on its underside. The top package is placed onto the bottom package after the bottom one has been reflowed, and the second reflow joins the two.

Because the two reflows are separate, the bottom package and its joints see two thermal cycles and the top package sees one. That asymmetry matters for the ball alloy and for the components, and it is one reason why a low temperature alloy is often used for the second joint while the first uses a standard one. The sequence in which the parts are placed and reflowed is part of the general assembly plan described under placement order and pad positioning.

Height And Warpage

The stack height is fixed by the ball size of the perimeter array and by the thickness of the two packages, and it has to fit inside the space the product allows. The perimeter balls are larger than the balls of a normal device, because they have to span the gap between the two packages and they have to survive the reflow of the second joint without collapsing. Their size sets the standoff, and the standoff has to be enough that the underfill or the mould compound of the bottom package does not touch the top package.

Warpage is the difficulty that dominates the process. The bottom package has to be flat on its top face where the top package will land, and the top package has to be flat on its underside. Both packages curl when they are heated, and they curl in opposite senses when one has its die and mould on the top and the other on the bottom. If the two surfaces are not parallel at reflow temperature, the perimeter balls are compressed unevenly: some are squashed and some are stretched, and the ones at the extremes are the ones that crack. The flatness behaviour of the laminate and the way it changes with temperature are described under the quality characteristics of a board design.

Alloy And Joint Design

The second joint is usually made with a lower melting alloy than the first, so that the first joints do not remelt when the top package is attached. The temperature difference between the two alloys has to be large enough that the first joint is well below its solidus during the second reflow, and small enough that the second alloy is compatible with the surfaces and the fluxes already present.

The joint geometry is set by the ball size, the pad size on both packages and the standoff. A joint that is too short for its diameter has a high strain concentration at the pad and fails early in thermal cycling; a joint that is too tall is mechanically weak and can be displaced during placement. In practice the geometry is fixed by the package specification, and the assembly engineer’s freedom is in the process rather than in the design. The alloy selection that governs the choice follows the same reasoning as the comparison under lead free versus leaded solder.

Process Requirements

The placement of the top package has to be accurate, because the balls have to land on pads of a similar size and there is no self centring that can recover a large error when the ball count is high and the perimeter is long. The placement force has to be low enough not to deform the balls before reflow, and the flux has to be sufficient to remove the oxide on both surfaces without leaving residue that cannot escape.

The reflow profile for the second joint is measured at the joint itself, which is difficult because the joint is buried between the two packages. A thermocouple attached to a corner ball of a test assembly is the usual approach, and the profile is written to ensure that the perimeter joints reach the liquidus while the first joints stay below their solidus. Where the mass of the two packages is different, the difference in heating rate between the two has to be measured rather than assumed.

Inspection And Reliability

Inspection of the perimeter joints is possible from the side, and automated optical inspection can see the outer row. The joints further inside the perimeter cannot be seen, and X-ray from above is limited because the balls of the bottom package and the ones of the top package overlap in the image. In practice the inspection relies on the visible outer row and on the process being controlled.

Reliability testing of the stack concentrates on thermal cycling, because the failure mode is a cracked perimeter joint at the pad interface. The strain comes from the difference in expansion between the two packages, which is larger than in a single package joint because the two packages are different sizes and different materials. A drop test is also relevant, since the top package has mass and its joints are the only thing holding it.

Where The Approach Fits

The package on package approach buys a short connection between two devices at the cost of a more demanding assembly process and a limited inspection. It is used where that connection is the bottleneck, typically between a processor and its memory. Where the connection can be made on the board instead, a conventional side by side arrangement is easier to build and to inspect, and the decision is usually made on the routing requirement rather than on the assembly cost alone.

Where the stack is used, the two packages should be qualified together rather than separately. A bottom package that is flat on its own may not be flat with a top package attached, and a top package that survives its own reflow may not survive the second cycle at the temperature the bottom package requires. The qualification is a property of the pair, and it should be recorded as such.

FAQ

Does the bottom package reflow twice? Yes. It sees its own reflow when it is attached to the board and a second cycle when the top package is attached. The second alloy is chosen so that the first joints do not remelt during the second cycle.

How is the second joint inspected? The outer row can be seen optically from the side. The inner joints cannot be inspected reliably, so the process is controlled by profile measurement and by the package specifications rather than by post assembly inspection.

Why is warpage critical in a stack? Because the two mating surfaces have to be parallel at reflow temperature. Any difference in curvature loads the perimeter balls unevenly, and the joints at the extremes of the perimeter are the ones that fail.