This was a good question and one that interested me greatly so I contacted some of the component manufacturers and found the following info from the Carsem MLP application note dated April 2002.

Thanks to Ivan Chiang of Holt who really did the legwork and provided the info below

Holt Integrated Circuits
+1 (949) 859-8800 (office)
E-Mail--ichiang@holtic.com

“There are two basic designs for PCB land pads for the MLP: Copper Defined style (also known as Non Solder Mask Defined (NSMD)) and the Solder Mask Defined style (SMD). The industry has had some debate of the merits of both styles of land pads, and although we recommend the Copper Defined style land pad (NSMD), both styles are acceptable for use with the MLP package.

NSMD pads are recommended over SMD pads due to the tighter tolerance on copper etching than solder masking. NSDM by definition also provides a larger copper pad area and allows the solder to anchor to the edges of the copper pads thus providing improved solder joint reliability.

The thermal pad (D2th) should be greater than D2 of the MLP whenever possible, however adequate clearance (Cpl > 0.15mm) must be met to prevent solder bridging. If this clearance cannot be met, then D2th should be reduced in area. The formula would be: D2TH >D2 only if D2TH < Gmin - (2 x Cpl)

Figure 3-4 is an example of a PCB Land Pattern for a 3x3-16L with an Exposed pad (D2) of 1.45mm.

MLP Dimensions for 3x3-16L- Blue (Corresponding Land Pattern - Green) Figure 3-4

The MLP package is thermally and electrically efficient. This is enabled by the exposed die attach pad on the under side of the package which must be soldered down to the PCB or mother board substrate.

It is good practice to minimize the presence of voids within the exposed pad inter-connection. Total elimination is difficult but the design of the exposed pad stencil is key.

Figure 4-2 shows some suggested screen print patterns. For exposed die pad sizes less than 25 mm2, a single slotted square pattern is recommended. For larger areas, a matrix of squares will minimize voids and normalize the standoff height for the exposed pad and the terminals. (If large exposed pads are screened with excessive solder, the device may "float", thus causing a gap between the MLP terminal and the pcb land metalization. See Figure 4-3.) The proposed stencil designs enables out-gassing of the solder paste during reflow as well as regulating the finished solder thickness.

Suggest you look at IPC 7525 for aperture recommendations on the ground plane. Window pane aperture at about 50% area reduction. Also might look at 1.25 to 1.5 times aperture width for the 8 corner pads.

This is a pretty common problem with this type of component. The secret here would be to reduce the center aperature to 60%, but you want to do it in a “window pane” configuration. This will essentially create 4 large pads of solder in the center ground area, helping to maintain registration and alignment of the part during reflow.

Your suspicion is correct, the center heat sink pad is causing the movement. Printed solder paste tends raise parts in the z-axis during the initial phases of reflow and as it coalesces. This causing the QFN's to rock slightly and shift resulting in miss alignment. We generally recommend a 50% reduction and window pane design to reduce this movement.

Another idea that seemed to work on a recent design was to make sure the area of stencil apertures or the surrounding pads was greater than that of the heat sink pad apertures. The thought here was that the surface tension of the solder at the individual leads would help anchor the part in place. The feed back from the customer was that the new design solved the problem.

The fundamental reason for component shift is solder surface tension. All things being equal, the molten solder will try to reduce surface energy. It does this by wetting and spreading on high energy surfaces (substrate and component terminations cleaned by the flux) and reducing its own surface area. Since the component can move, it gets drawn into alignment and capillary gaps get closed down which is why we generally see self-alignment.

If component wetting is uneven at different sites on the component (different thermal demand on different parts of the components, different levels of solderability, different volumes of solder paste) or placement accuracy is very poor, then the surface energy may be reduced for only some of the areas to be soldered and this may move the component away from correct alignment. For example, array packages may "float" and rotate so not all terminations are on pads and others are on the wrong pads.

There are cases where components may be physically moved before the solder melts. This is caused by explosive outgassing from flux and other organics on the PCB. It may also happen if the tack force exerted by the printed paste is too low (it may have dried before or after component placement) and simple handling causes the component to move.

In short yes the problem could be related to the heat sink, uneven printing or offset printing or the design of the heat sink.

I’m not sure I have the exact answer, but my thoughts on this are that the (weight of the component) versus the (size of the solder contact), versus the (surface tension) of the molten metal create an equation which shows that the component does not weigh enough to stay in place, i.e. it floats on the surface of the molten solder.

The realignment difficulty in my opinion is based upon the amount of flux or not metallic elements in the paste that are moving about while the solder is molten and these liquids are floating on the surfaces of the molten liquid and moving the component at the same time. An analogy may be of a boat floating on the lake with rough currents beneath it, the boat will not stay in one place, it will move and I think the same thing is happening to your component.

For components to align themselves the size of the solder joint, solder pad or land area have to be smaller which is not the case with this particular component. The case for example would be BGA components, each solder joint is an independent floating devices but each termination has its own surface energies and when they are all molten they work together to reduce the liquid forces on each solder joint and the component realigns itself.

You don’t see this as much with the lead-free materials as they don’t wet as well and therefore the liquid stresses never find an equilibrium point and the components don’t realign themselves to the pads.

Make the pad on the board the same size as the heat-sink pad on the component would be an area that I would investigate first. My thought being that once the joint is made its total area is constant between the component and the land on the board and that should create an equilibrium condition between both the component and the board and this should help in the alignment of the component.

Hope this helps and looking forward to other answers from your request.