I was recently asked to give a presentation and audit an assembly line regarding minimizing "tombstoning" of passives at a major electronic assembler.
As my presentation brought out, tomstoning can be caused by many factors: the reflow profile, the solder metal composition (for lead-free SAC 387 tends to tombstone more than SAC 305), off-center placement, nitrogen reflow atmosphere, buried vias, etc. After giving a two hour presentation, I toured the line that "had a problem with tombstoning."
As I started asking, it became clear that no one knew the magnitude of the problem. "How many passives are on each board?" I asked. No one knew. "How many DPMO (defects per million opportunities) for tombstones have you had recently." This metric was also unknown. As people scurried around to get data, it started to become clear that tombstoning might not be as big an issue as everyone thought. It was more of a local legend.
Finally, we got some data. Each board had about 1000 passives, they had produced 100 boards with a total of two tombstones in the last two hours. Tombstones were the only defect. Hmmmmm two bad boards out of 100 = 98% first pass yield, not bad! From a DPMO perspective, they had 2 defects per 100,000 opportunities or 20 DPMO, which is world class. This level of DPMO would be very difficult to improve on without massive engineering investment. It is "in the noise" and it is likely caused by "common cause" variation. I then asked how much money it costs to repair a tombstone, as expected no one knew. My guess is less than $2. This situation is the rare case where yields are so good, it may not pay to make engineering investment to improve them.
This isn't the point of the story however, whatever is done in a case like this one, needs to be "data driven." Only with the proper failure rate data, plotted in a Pareto chart and understanding costs can the appropriate action plan be developed.
Always be data driven!