I recently had one of those moments of clarity that comes from asking different people the same questions and fitting all the varied answers together like a puzzle to come up with the big picture. In this case, the questions had to do with the various approaches being developed to achieve 3D integration using TSVs – namely, via-first, middle and last for via formation; and wafer-to-wafer (W2W) and die-to-wafer (D2W) for device stacking. One thing is for sure, there is no shortage on opinions out there surrounding these issues.
At the risk of oversimplification, I offer up the following analysis; perhaps the best way to understand how 3D integration using TSV will ultimately come together is to see it as a progression – a stepping stone approach. Even though that’s how the questions are often posed – it’s not likely to end up as an either-or situation.
D2W currently achieves better yields than W2W, but once the known-good-die issue is figured out, W2W will be a more cost-effective process because it is done in parallel. Via-middle will most likely be the TSV champion, once design and test tools are available; but until then, some predict that the first products will be built using via-last, which can be done with existing die. It’s a matter of convincing the industry as a whole (and there are still many who remain skeptical that TSV itself is the answer) that overcoming current limitations for W2W and via-middle is a worthy investment.
Regardless of how things shake out, the equipment and materials manufacturers are covering their bases by making sure their tools and chemistries adapt or exist for all possible variations. Because no matter which processes are adopted for volume production, there will always be niche applications that are better suited to the processes that don’t reach volume production. – F.v.T