The Race To Mass Customization

The number of advanced packaging options continues to rise. The choices now include different materials for interposers, at least a half-dozen fan-outs, not to mention hybrid fan-out/3D stacking, system-in-package, flip-chip and die-to-die bridges.

There are several reasons for all of this activity. First, advanced packaging offers big improvements in performance and power that cannot be accomplished solely by shrinking features and increasing density. Throughput and shorter distances make it possible to achieve the benefits that only scaling could provide in the past.

Second, the cost of device scaling continues to rise with each new node. While density will continue to make a significant difference in digital logic for the foreseeable future—at least until quantum computing begins siphoning off development in the cloud—it doesn’t make sense to shrink the analog components in an SoC.

And third, faster time to market and flexibility need to happen concurrently because end markets are still immature. Those two approaches don’t mesh well on the same die, and the market is littered with companies that got it wrong. A common approach at the start of the IoT era was to use off-the-shelf IP to build solutions, producing among other things smart watches that didn’t last a full day before they needed to be charged. At the same time, a number of new markets are shifting so fast that companies cannot wait 18 months until a new design is ready.

Advanced packaging is a bridge across all of these factors, and that’s evident by the frenzy of activity in this market. Not all of the packaging approaches being developed will survive, of course. There are always more contestants at the beginning of a race than at the finish line.

Consider what’s happening in the automotive market, for example. Wally Rhines, president and CEO of Mentor, a Siemens Business, said at a recent event that 338 companies have announced plans to develop electric cars, and 127 indicated they’re developing driverless cars. “Ten years from now, most of those will not be in business, or they will not be doing what they were doing,” he said.

The same trend could be seen at the beginning of the automotive market. In 1909, there were 285 car companies. By 1930, that number had dropped to 27, and by 1960 there were essentially 3 left, Rhines said.

The difference in advanced packaging is that the cost of developing these packaging options is so high that it won’t necessarily cause a reduction in the number of companies working in this space. But the market will winnow out the best options, which is the point at which the industry will begin to develop economies of scale and standardized ways of putting different chips together.

Nor is this confined just to the packaging arena. Foundries have announced process nodes at every whole number down to 3nm, and there is work underway to shrink chips all the way to 1.3 or 1.2nm, by which point the naming conventions will likely shift to angstroms rather than nanometers. It’s uncertain if the chip industry will actually ever get to that point, but the path forward is clear, even if it appears prohibitively expensive at this point. Still, if logic becomes platform-ized, then it could very well make sense. A 1.3nm SoC sounds like a stretch, but a small 1.3nm processor designed with very regular shapes and lots of redundancy sounds much more feasible.

The advanced packaging race has begun. Place your bets for how the field will look in the next five years.